16ec11b Merge pull request #5 from bnema/feat/openai-fast-mode · zut

M README.md +3 -2


@@ -137,7 +137,7 @@ All data lives under `$ZOT_HOME`:
137	137
138	138		```
139	139		$ZOT_HOME/
140		-	├── config.json # last-used provider/model/theme, saved automatically
	140	+	├── config.json # last-used provider/model/theme and persistent settings
141	141		├── auth.json # api keys and oauth tokens (mode 0600)
142	142		├── sessions/ # jsonl transcripts, one dir per cwd
143	143		├── models-cache.json # live /v1/models discovery cache (6h ttl)

@@ -289,7 +289,7 @@ Slash command names are case-insensitive in the TUI and messaging backends; argu
289	289		\| `/unjail` \| Allow tools to touch paths outside again. \|
290	290		\| `/reload-ext` \| Hot-reload all extensions (re-read manifests, respawn subprocesses, rebuild tool registry). \|
291	291		\| `/telegram` \| Connect, disconnect, or show status of the Telegram bridge (takes `connect` / `disconnect` / `status` as an optional argument; opens a picker without one). When connected, DMs from the paired user become prompts in the running session and the assistant's replies are mirrored back to Telegram. Alias: `/tg`. \|
292		-	\| `/settings` \| Change persistent settings, including inline images, terminal alerts, auto-swarm, and the auto-compact threshold. Saved to `$ZOT_HOME/config.json`; takes effect immediately. \|
	292	+	\| `/settings` \| Change persistent settings, including inline images, terminal alerts, auto-swarm, fast mode, and the auto-compact threshold. Saved to `$ZOT_HOME/config.json`; takes effect immediately. \|
293	293		\| `/clear` \| Clear the chat transcript. \|
294	294		\| `/exit` \| Exit zot. \|
295	295

@@ -392,6 +392,7 @@ Opens a dialog with every persistent setting. `up`/`down` to navigate, `enter` o
392	392		- render images when supported — draw screenshots / `read`-returned images inline using the terminal's image protocol, or fall back to a text placeholder. Auto-detected from `TERM_PROGRAM`; the toggle overrides the detection. The row is greyed out and forced off on terminals that don't speak any image protocol.
393	393		- terminal alerts — emit a terminal bell when the main session stops with work that may need attention, or when an extension raises a structured alert. Enabled by default; changes apply immediately and persist as `terminal_alerts_enabled`. Terminal emulators may render the bell audibly, visually, or not at all.
394	394		- auto-swarm — let the main agent spawn background sub-agents in parallel via a built-in `swarm_spawn` tool. Off by default. When on, the tool is registered with the running agent, the system prompt gains a short addendum telling the model to delegate independent sub-tasks proactively, and available named profiles are rendered in `[subagents_list]`. The tool accepts a profile name through `agent` and a per-child `reasoning`/`thinking` override. zot watches every sub-agent the main agent spawns. As soon as the last sub-agent in a batch finishes its initial task, an `[auto-swarm update]` message is injected back into the chat with each agent's status / task / transcript tail, so the main agent can summarise the collective outcome. Flipping off mid-session removes the tool from the live agent and strips the addendum on the next turn — the model stops trying to delegate. See `/swarm` and [docs/subagents.md](docs/subagents.md) for details.
	395	+	- fast mode — request OpenAI's Fast service tier for OpenAI, OpenAI Responses, and OpenAI Codex models. Off by default; changes apply on the next model call and persist as `fast_mode`. Other providers return an unsupported-provider error. Fast mode may cost more and depends on the selected OpenAI model/account.
395	396		- auto-compact threshold — choose `off`, `70%`, `80%`, `85%` (default), or `90%` of the model's advertised context window. The selected percentage controls automatic compaction before and after interactive turns and persists as `auto_compact_threshold`. `off` disables percentage-based triggers but keeps manual `/compact` and automatic recovery from context-window and payload-too-large responses.
396	397		- jail new sessions by default: start every new agent with tools confined to its working directory. Off by default. The setting applies to interactive, print, JSON, RPC, and background-agent runs, persists as `jail_by_default`, and immediately updates the current interactive session. `/jail` and `/unjail` remain session-scoped overrides and do not change this default.
397	398		- compact transcript rendering: reduce visual chrome in the chat transcript. Tool calls render as a quiet header plus indented output instead of a bordered box, and sent messages render without padded background bubbles. Off by default. Changes apply immediately and persist to `config.json` as `compact_mode`.

 ```
 $ZOT_HOME/
-├── config.json         # last-used provider/model/theme, saved automatically
+├── config.json         # last-used provider/model/theme and persistent settings
 ├── auth.json           # api keys and oauth tokens (mode 0600)
 ├── sessions/           # jsonl transcripts, one dir per cwd
 ├── models-cache.json   # live /v1/models discovery cache (6h ttl)
 | `/unjail` | Allow tools to touch paths outside again. |
 | `/reload-ext` | Hot-reload all extensions (re-read manifests, respawn subprocesses, rebuild tool registry). |
 | `/telegram` | Connect, disconnect, or show status of the Telegram bridge (takes `connect` / `disconnect` / `status` as an optional argument; opens a picker without one). When connected, DMs from the paired user become prompts in the running session and the assistant's replies are mirrored back to Telegram. Alias: `/tg`. |
-| `/settings` | Change persistent settings, including inline images, terminal alerts, auto-swarm, and the auto-compact threshold. Saved to `$ZOT_HOME/config.json`; takes effect immediately. |
+| `/settings` | Change persistent settings, including inline images, terminal alerts, auto-swarm, fast mode, and the auto-compact threshold. Saved to `$ZOT_HOME/config.json`; takes effect immediately. |
 | `/clear` | Clear the chat transcript. |
 | `/exit` | Exit zot. |
 - **render images when supported** — draw screenshots / `read`-returned images inline using the terminal's image protocol, or fall back to a text placeholder. Auto-detected from `TERM_PROGRAM`; the toggle overrides the detection. The row is greyed out and forced off on terminals that don't speak any image protocol.
 - **terminal alerts** — emit a terminal bell when the main session stops with work that may need attention, or when an extension raises a structured alert. Enabled by default; changes apply immediately and persist as `terminal_alerts_enabled`. Terminal emulators may render the bell audibly, visually, or not at all.
 - **auto-swarm** — let the main agent spawn background sub-agents in parallel via a built-in `swarm_spawn` tool. Off by default. When on, the tool is registered with the running agent, the system prompt gains a short addendum telling the model to delegate independent sub-tasks proactively, and available named profiles are rendered in `[subagents_list]`. The tool accepts a profile name through `agent` and a per-child `reasoning`/`thinking` override. zot watches every sub-agent the main agent spawns. As soon as the last sub-agent in a batch finishes its initial task, an `[auto-swarm update]` message is injected back into the chat with each agent's status / task / transcript tail, so the main agent can summarise the collective outcome. Flipping off mid-session removes the tool from the live agent and strips the addendum on the next turn — the model stops trying to delegate. See `/swarm` and [docs/subagents.md](docs/subagents.md) for details.
+- **fast mode** — request OpenAI's Fast service tier for OpenAI, OpenAI Responses, and OpenAI Codex models. Off by default; changes apply on the next model call and persist as `fast_mode`. Other providers return an unsupported-provider error. Fast mode may cost more and depends on the selected OpenAI model/account.
 - **auto-compact threshold** — choose `off`, `70%`, `80%`, `85%` (default), or `90%` of the model's advertised context window. The selected percentage controls automatic compaction before and after interactive turns and persists as `auto_compact_threshold`. `off` disables percentage-based triggers but keeps manual `/compact` and automatic recovery from context-window and payload-too-large responses.
 - **jail new sessions by default**: start every new agent with tools confined to its working directory. Off by default. The setting applies to interactive, print, JSON, RPC, and background-agent runs, persists as `jail_by_default`, and immediately updates the current interactive session. `/jail` and `/unjail` remain session-scoped overrides and do not change this default.
 - **compact transcript rendering**: reduce visual chrome in the chat transcript. Tool calls render as a quiet header plus indented output instead of a bordered box, and sent messages render without padded background bubbles. Off by default. Changes apply immediately and persist to `config.json` as `compact_mode`.

M docs/providers.md +10 -0


@@ -115,6 +115,16 @@ show instructions and should be configured with environment variables.
115	115		\| Cloudflare AI Gateway \| `CLOUDFLARE_API_KEY` \| `cloudflare-ai-gateway` \|
116	116		\| Azure OpenAI Responses \| `AZURE_OPENAI_API_KEY` \| `azure-openai-responses` \|
117	117
	118	+	### OpenAI Fast mode
	119	+
	120	+	OpenAI, OpenAI Responses, and OpenAI Codex requests can opt into OpenAI's
	121	+	Fast service tier from `/settings` by enabling fast mode. The setting is
	122	+	off by default and is stored as `fast_mode` in `$ZOT_HOME/config.json`.
	123	+
	124	+	Fast mode is currently limited to those OpenAI provider IDs. Enabling it while
	125	+	using another provider returns an error instead of silently changing that
	126	+	provider's request. Swarm children inherit the setting from their parent.
	127	+
118	128		Example:
119	129
120	130		```bash


@@ -115,6 +115,16 @@ show instructions and should be configured with environment variables.
115	\| Cloudflare AI Gateway \| `CLOUDFLARE_API_KEY` \| `cloudflare-ai-gateway` \|	115		\| Cloudflare AI Gateway \| `CLOUDFLARE_API_KEY` \| `cloudflare-ai-gateway` \|
116	\| Azure OpenAI Responses \| `AZURE_OPENAI_API_KEY` \| `azure-openai-responses` \|	116		\| Azure OpenAI Responses \| `AZURE_OPENAI_API_KEY` \| `azure-openai-responses` \|
117		117
		118	+	### OpenAI Fast mode
		119	+
		120	+	OpenAI, OpenAI Responses, and OpenAI Codex requests can opt into OpenAI's
		121	+	Fast service tier from `/settings` by enabling fast mode. The setting is
		122	+	off by default and is stored as `fast_mode` in `$ZOT_HOME/config.json`.
		123	+
		124	+	Fast mode is currently limited to those OpenAI provider IDs. Enabling it while
		125	+	using another provider returns an error instead of silently changing that
		126	+	provider's request. Swarm children inherit the setting from their parent.
		127	+
118	Example:	128		Example:
119		129
120	```bash	130		```bash

M docs/subagents.md +1 -1


@@ -82,4 +82,4 @@ The interactive command also supports the same selection explicitly:
82	82		/swarm new --agent implementer --reasoning high Implement the parser change
83	83		```
84	84
85		-	All swarm children still share the host working directory. Named profiles change the child's instructions and configuration; they do not create a worktree, branch, or security sandbox.
	85	+	All swarm children still share the host working directory. Named profiles change the child's instructions and configuration; they do not create a worktree, branch, or security sandbox. When the host has fast mode enabled, every child inherits it; non-OpenAI child providers return an unsupported-provider error instead of silently ignoring the setting.


@@ -82,4 +82,4 @@ The interactive command also supports the same selection explicitly:
82		/swarm new --agent implementer --reasoning high Implement the parser change	82		/swarm new --agent implementer --reasoning high Implement the parser change
83		```	83		```
84			84
85	-	All swarm children still share the host working directory. Named profiles change the child's instructions and configuration; they do not create a worktree, branch, or security sandbox.	85	+	All swarm children still share the host working directory. Named profiles change the child's instructions and configuration; they do not create a worktree, branch, or security sandbox. When the host has fast mode enabled, every child inherits it; non-OpenAI child providers return an unsupported-provider error instead of silently ignoring the setting.

M packages/agent/args.go +14 -0


@@ -48,6 +48,14 @@ type Args struct {
48	48		Reasoning string
49	49		Temperature *float32
50	50
	51	+	// FastMode is an internal swarm-child propagation flag. Normal users
	52	+	// enable fast mode through the persisted config /settings toggle.
	53	+	FastMode bool
	54	+	// FastModeSet distinguishes an explicit child override (including
	55	+	// --no-fast-mode) from an omitted flag so persisted swarm settings
	56	+	// survive a parent config change.
	57	+	FastModeSet bool
	58	+
51	59		Continue bool
52	60		Resume bool
53	61		Session string

@@ -254,6 +262,12 @@ func ParseArgs(in []string) (Args, error) {
254	262		}
255	263		t := float32(f)
256	264		a.Temperature = &t
	265	+	case "--fast-mode":
	266	+	a.FastMode = true
	267	+	a.FastModeSet = true
	268	+	case "--no-fast-mode":
	269	+	a.FastMode = false
	270	+	a.FastModeSet = true
257	271		case "--stats":
258	272		v, err := want(&i, arg)
259	273		if err != nil {


@@ -48,6 +48,14 @@ type Args struct {
48	Reasoning string	48		Reasoning string
49	Temperature *float32	49		Temperature *float32
50		50
		51	+	// FastMode is an internal swarm-child propagation flag. Normal users
		52	+	// enable fast mode through the persisted config /settings toggle.
		53	+	FastMode bool
		54	+	// FastModeSet distinguishes an explicit child override (including
		55	+	// --no-fast-mode) from an omitted flag so persisted swarm settings
		56	+	// survive a parent config change.
		57	+	FastModeSet bool
		58	+
51	Continue bool	59		Continue bool
52	Resume bool	60		Resume bool
53	Session string	61		Session string

@@ -254,6 +262,12 @@ func ParseArgs(in []string) (Args, error) {
254	}	262		}
255	t := float32(f)	263		t := float32(f)
256	a.Temperature = &t	264		a.Temperature = &t
		265	+	case "--fast-mode":
		266	+	a.FastMode = true
		267	+	a.FastModeSet = true
		268	+	case "--no-fast-mode":
		269	+	a.FastMode = false
		270	+	a.FastModeSet = true
257	case "--stats":	271		case "--stats":
258	v, err := want(&i, arg)	272		v, err := want(&i, arg)
259	if err != nil {	273		if err != nil {

M packages/agent/build.go +9 -2


@@ -34,6 +34,7 @@ type Resolved struct {
34	34		CWD string
35	35		Reasoning string
36	36		Temperature *float32
	37	+	FastMode bool
37	38
38	39		ToolRegistry core.Registry
39	40		ToolSummary []ToolSummary

@@ -698,6 +699,10 @@ func Resolve(args Args, requireCred bool) (Resolved, error) {
698	699		if temperature == nil {
699	700		temperature = cfg.Temperature
700	701		}
	702	+	fastMode := cfg.FastMode != nil && *cfg.FastMode
	703	+	if args.FastModeSet {
	704	+	fastMode = args.FastMode
	705	+	}
701	706
702	707		max := args.MaxSteps // 0 = unlimited
703	708

@@ -712,6 +717,7 @@ func Resolve(args Args, requireCred bool) (Resolved, error) {
712	717		CWD: args.CWD,
713	718		Reasoning: reasoning,
714	719		Temperature: temperature,
	720	+	FastMode: fastMode,
715	721		ToolRegistry: reg,
716	722		ToolSummary: summaries,
717	723		SystemPrompt: sys,

@@ -840,13 +846,13 @@ func (r Resolved) NewClient() provider.Client {
840	846		wrap := r.withHTTPClient
841	847		switch r.Provider {
842	848		case "ollama":
843		-	return wrap(provider.NewOpenAI(r.Credential, r.BaseURL))
	849	+	return wrap(provider.NewOpenAICompat("ollama", r.Credential, r.BaseURL, ""))
844	850		case provider.LlamaCPPProviderID:
845	851		credential := r.Credential
846	852		if credential == "local" {
847	853		credential = ""
848	854		}
849		-	return wrap(provider.NewOpenAI(credential, r.BaseURL))
	855	+	return wrap(provider.NewOpenAICompat(provider.LlamaCPPProviderID, credential, r.BaseURL, ""))
850	856		case "kimi":
851	857		// kimi-coding speaks anthropic-messages on api.kimi.com/coding.
852	858		// Subscription OAuth (refreshed) wraps the same Anthropic-shaped client.

@@ -1034,6 +1040,7 @@ func (r Resolved) NewAgent() *core.Agent {
1034	1040		a.MaxTokens = r.MaxOutput
1035	1041		a.Reasoning = r.Reasoning
1036	1042		a.Temperature = r.Temperature
	1043	+	a.FastMode = r.FastMode
1037	1044		return a
1038	1045		}
1039	1046

 	CWD         string
 	Reasoning   string
 	Temperature *float32
+	FastMode    bool
 	ToolRegistry core.Registry
 	ToolSummary  []ToolSummary
 	if temperature == nil {
 		temperature = cfg.Temperature
 	}
+	fastMode := cfg.FastMode != nil && *cfg.FastMode
+	if args.FastModeSet {
+		fastMode = args.FastMode
+	}
 	max := args.MaxSteps // 0 = unlimited
 		CWD:              args.CWD,
 		Reasoning:        reasoning,
 		Temperature:      temperature,
+		FastMode:         fastMode,
 		ToolRegistry:     reg,
 		ToolSummary:      summaries,
 		SystemPrompt:     sys,
 	wrap := r.withHTTPClient
 	switch r.Provider {
 	case "ollama":
-		return wrap(provider.NewOpenAI(r.Credential, r.BaseURL))
+		return wrap(provider.NewOpenAICompat("ollama", r.Credential, r.BaseURL, ""))
 	case provider.LlamaCPPProviderID:
 		credential := r.Credential
 		if credential == "local" {
 			credential = ""
 		}
-		return wrap(provider.NewOpenAI(credential, r.BaseURL))
+		return wrap(provider.NewOpenAICompat(provider.LlamaCPPProviderID, credential, r.BaseURL, ""))
 	case "kimi":
 		// kimi-coding speaks anthropic-messages on api.kimi.com/coding.
 		// Subscription OAuth (refreshed) wraps the same Anthropic-shaped client.
 	a.MaxTokens = r.MaxOutput
 	a.Reasoning = r.Reasoning
 	a.Temperature = r.Temperature
+	a.FastMode = r.FastMode
 	return a
 }

M packages/agent/build_test.go +2 -2


@@ -405,8 +405,8 @@ func TestResolveLlamaCPPUsesRouterInferenceURL(t *testing.T) {
405	405		if r.Credential != "local" \|\| !r.HasCredential() {
406	406		t.Fatalf("credential = %q", r.Credential)
407	407		}
408		-	if got := r.NewClient().Name(); got != "openai" {
409		-	t.Fatalf("client name = %q", got)
	408	+	if got := r.NewClient().Name(); got != provider.LlamaCPPProviderID {
	409	+	t.Fatalf("client name = %q, want %q", got, provider.LlamaCPPProviderID)
410	410		}
411	411		}
412	412


@@ -405,8 +405,8 @@ func TestResolveLlamaCPPUsesRouterInferenceURL(t *testing.T) {
405		if r.Credential != "local" \|\| !r.HasCredential() {	405		if r.Credential != "local" \|\| !r.HasCredential() {
406		t.Fatalf("credential = %q", r.Credential)	406		t.Fatalf("credential = %q", r.Credential)
407		}	407		}
408	-	if got := r.NewClient().Name(); got != "openai" {	408	+	if got := r.NewClient().Name(); got != provider.LlamaCPPProviderID {
409	-	t.Fatalf("client name = %q", got)	409	+	t.Fatalf("client name = %q, want %q", got, provider.LlamaCPPProviderID)
410		}	410		}
411		}	411		}
412			412

M packages/agent/cli.go +3 -0


@@ -743,6 +743,7 @@ func runInteractive(ctx context.Context, args Args, version string) error {
743	743		swarmMgr = swarm.New(swarm.Config{
744	744		Root: filepath.Join(ZotHome(), "swarm"),
745	745		RepoRoot: r.CWD,
	746	+	FastMode: r.FastMode,
746	747		ResolveCredential: func(ctx context.Context, providerID string) (swarm.Credential, error) {
747	748		if providerID == "ollama" {
748	749		return swarm.Credential{Value: "ollama", Method: "apikey"}, nil

@@ -1291,6 +1292,7 @@ func runInteractive(ctx context.Context, args Args, version string) error {
1291	1292		}()
1292	1293
1293	1294		initialCfg, _ := LoadConfig()
	1295	+	fastMode := r.FastMode
1294	1296		quickModelShortcuts := make([]modes.QuickModelShortcut, len(initialCfg.QuickModelShortcuts))
1295	1297		for idx, s := range initialCfg.QuickModelShortcuts {
1296	1298		quickModelShortcuts[idx] = modes.QuickModelShortcut{Provider: s.Provider, Model: s.Model}

@@ -1331,6 +1333,7 @@ func runInteractive(ctx context.Context, args Args, version string) error {
1331	1333		InlineImagesEnabled: initialCfg.InlineImagesEnabled,
1332	1334		TerminalAlertsEnabled: initialCfg.TerminalAlertsEnabled,
1333	1335		AutoSwarmEnabled: initialCfg.AutoSwarmEnabled,
	1336	+	FastMode: &fastMode,
1334	1337		AutoCompactThreshold: initialCfg.AutoCompactThreshold,
1335	1338		JailByDefault: initialCfg.JailByDefault,
1336	1339		QuickModelShortcuts: quickModelShortcuts,


@@ -743,6 +743,7 @@ func runInteractive(ctx context.Context, args Args, version string) error {
743	swarmMgr = swarm.New(swarm.Config{	743		swarmMgr = swarm.New(swarm.Config{
744	Root: filepath.Join(ZotHome(), "swarm"),	744		Root: filepath.Join(ZotHome(), "swarm"),
745	RepoRoot: r.CWD,	745		RepoRoot: r.CWD,
		746	+	FastMode: r.FastMode,
746	ResolveCredential: func(ctx context.Context, providerID string) (swarm.Credential, error) {	747		ResolveCredential: func(ctx context.Context, providerID string) (swarm.Credential, error) {
747	if providerID == "ollama" {	748		if providerID == "ollama" {
748	return swarm.Credential{Value: "ollama", Method: "apikey"}, nil	749		return swarm.Credential{Value: "ollama", Method: "apikey"}, nil

@@ -1291,6 +1292,7 @@ func runInteractive(ctx context.Context, args Args, version string) error {
1291	}()	1292		}()
1292		1293
1293	initialCfg, _ := LoadConfig()	1294		initialCfg, _ := LoadConfig()
		1295	+	fastMode := r.FastMode
1294	quickModelShortcuts := make([]modes.QuickModelShortcut, len(initialCfg.QuickModelShortcuts))	1296		quickModelShortcuts := make([]modes.QuickModelShortcut, len(initialCfg.QuickModelShortcuts))
1295	for idx, s := range initialCfg.QuickModelShortcuts {	1297		for idx, s := range initialCfg.QuickModelShortcuts {
1296	quickModelShortcuts[idx] = modes.QuickModelShortcut{Provider: s.Provider, Model: s.Model}	1298		quickModelShortcuts[idx] = modes.QuickModelShortcut{Provider: s.Provider, Model: s.Model}

@@ -1331,6 +1333,7 @@ func runInteractive(ctx context.Context, args Args, version string) error {
1331	InlineImagesEnabled: initialCfg.InlineImagesEnabled,	1333		InlineImagesEnabled: initialCfg.InlineImagesEnabled,
1332	TerminalAlertsEnabled: initialCfg.TerminalAlertsEnabled,	1334		TerminalAlertsEnabled: initialCfg.TerminalAlertsEnabled,
1333	AutoSwarmEnabled: initialCfg.AutoSwarmEnabled,	1335		AutoSwarmEnabled: initialCfg.AutoSwarmEnabled,
		1336	+	FastMode: &fastMode,
1334	AutoCompactThreshold: initialCfg.AutoCompactThreshold,	1337		AutoCompactThreshold: initialCfg.AutoCompactThreshold,
1335	JailByDefault: initialCfg.JailByDefault,	1338		JailByDefault: initialCfg.JailByDefault,
1336	QuickModelShortcuts: quickModelShortcuts,	1339		QuickModelShortcuts: quickModelShortcuts,

M packages/agent/config.go +5 -0


@@ -30,6 +30,11 @@ type Config struct {
30	30		Temperature *float32 `json:"temperature,omitempty"`
31	31		Theme string `json:"theme"`
32	32
	33	+	// FastMode requests OpenAI's fast service tier. Off by default;
	34	+	// nil/missing means disabled. Other providers reject fast-mode
	35	+	// requests at the provider boundary.
	36	+	FastMode *bool `json:"fast_mode,omitempty"`
	37	+
33	38		// ToolRender selects how tool calls are drawn in interactive mode.
34	39		// "box" (default, or empty) wraps each call in a bordered panel;
35	40		// "flat" drops the frame for a quiet header line plus indented,


@@ -30,6 +30,11 @@ type Config struct {
30	Temperature *float32 `json:"temperature,omitempty"`	30		Temperature *float32 `json:"temperature,omitempty"`
31	Theme string `json:"theme"`	31		Theme string `json:"theme"`
32		32
		33	+	// FastMode requests OpenAI's fast service tier. Off by default;
		34	+	// nil/missing means disabled. Other providers reject fast-mode
		35	+	// requests at the provider boundary.
		36	+	FastMode *bool `json:"fast_mode,omitempty"`
		37	+
33	// ToolRender selects how tool calls are drawn in interactive mode.	38		// ToolRender selects how tool calls are drawn in interactive mode.
34	// "box" (default, or empty) wraps each call in a bordered panel;	39		// "box" (default, or empty) wraps each call in a bordered panel;
35	// "flat" drops the frame for a quiet header line plus indented,	40		// "flat" drops the frame for a quiet header line plus indented,

A packages/agent/fast_mode_test.go +152 -0

@@ -0,0 +1,152 @@
	1	+	package agent
	2	+
	3	+	import (
	4	+	"context"
	5	+	"encoding/json"
	6	+	"strings"
	7	+	"testing"
	8	+
	9	+	"github.com/patriceckhart/zot/packages/provider"
	10	+	)
	11	+
	12	+	func TestParseArgsFastModeChildFlag(t *testing.T) {
	13	+	args, err := ParseArgs([]string{"--fast-mode"})
	14	+	if err != nil {
	15	+	t.Fatalf("ParseArgs failed: %v", err)
	16	+	}
	17	+	if !args.FastMode {
	18	+	t.Fatal("FastMode = false, want true")
	19	+	}
	20	+	if !args.FastModeSet {
	21	+	t.Fatal("FastModeSet = false, want true")
	22	+	}
	23	+	}
	24	+
	25	+	func TestResolveFastModeDefaultsOff(t *testing.T) {
	26	+	t.Setenv("ZOT_HOME", t.TempDir())
	27	+	if err := SaveConfig(Config{Provider: "openai", Model: "gpt-5"}); err != nil {
	28	+	t.Fatal(err)
	29	+	}
	30	+
	31	+	r, err := Resolve(Args{}, false)
	32	+	if err != nil {
	33	+	t.Fatalf("Resolve failed: %v", err)
	34	+	}
	35	+	if r.FastMode {
	36	+	t.Fatal("FastMode = true by default, want false")
	37	+	}
	38	+	}
	39	+
	40	+	func TestResolveFastModeFromConfig(t *testing.T) {
	41	+	t.Setenv("ZOT_HOME", t.TempDir())
	42	+	enabled := true
	43	+	if err := SaveConfig(Config{Provider: "openai", Model: "gpt-5", FastMode: &enabled}); err != nil {
	44	+	t.Fatal(err)
	45	+	}
	46	+
	47	+	r, err := Resolve(Args{}, false)
	48	+	if err != nil {
	49	+	t.Fatalf("Resolve failed: %v", err)
	50	+	}
	51	+	if !r.FastMode {
	52	+	t.Fatal("FastMode = false, want true from config")
	53	+	}
	54	+	}
	55	+
	56	+	func TestResolveFastModeRemainsEnabledForUnsupportedProviderUntilRequest(t *testing.T) {
	57	+	t.Setenv("ZOT_HOME", t.TempDir())
	58	+	enabled := true
	59	+	if err := SaveConfig(Config{Provider: "anthropic", Model: "claude-sonnet-4-5", FastMode: &enabled}); err != nil {
	60	+	t.Fatal(err)
	61	+	}
	62	+
	63	+	r, err := Resolve(Args{}, false)
	64	+	if err != nil {
	65	+	t.Fatalf("Resolve failed: %v", err)
	66	+	}
	67	+	if !r.FastMode {
	68	+	t.Fatal("FastMode = false, want true from config")
	69	+	}
	70	+	}
	71	+
	72	+	func TestResolveFastModeRequiresExplicitChildOverride(t *testing.T) {
	73	+	t.Setenv("ZOT_HOME", t.TempDir())
	74	+	if err := SaveConfig(Config{Provider: "openai", Model: "gpt-5"}); err != nil {
	75	+	t.Fatal(err)
	76	+	}
	77	+
	78	+	r, err := Resolve(Args{FastMode: true}, false)
	79	+	if err != nil {
	80	+	t.Fatalf("Resolve failed: %v", err)
	81	+	}
	82	+	if r.FastMode {
	83	+	t.Fatal("FastMode = true, want unmarked programmatic value to be ignored")
	84	+	}
	85	+	}
	86	+
	87	+	func TestResolveFastModeExplicitChildDisableOverridesConfig(t *testing.T) {
	88	+	t.Setenv("ZOT_HOME", t.TempDir())
	89	+	enabled := true
	90	+	if err := SaveConfig(Config{Provider: "openai", Model: "gpt-5", FastMode: &enabled}); err != nil {
	91	+	t.Fatal(err)
	92	+	}
	93	+	args, err := ParseArgs([]string{"--no-fast-mode"})
	94	+	if err != nil {
	95	+	t.Fatalf("ParseArgs failed: %v", err)
	96	+	}
	97	+
	98	+	r, err := Resolve(args, false)
	99	+	if err != nil {
	100	+	t.Fatalf("Resolve failed: %v", err)
	101	+	}
	102	+	if r.FastMode {
	103	+	t.Fatal("FastMode = true, want explicit child disable to override config")
	104	+	}
	105	+	}
	106	+
	107	+	func TestConfigSettingsStorePersistsFastMode(t *testing.T) {
	108	+	t.Setenv("ZOT_HOME", t.TempDir())
	109	+	if err := SaveConfig(Config{Theme: "dark"}); err != nil {
	110	+	t.Fatal(err)
	111	+	}
	112	+
	113	+	if err := (configSettingsStore{}).SetFastMode(true); err != nil {
	114	+	t.Fatal(err)
	115	+	}
	116	+	cfg, err := LoadConfig()
	117	+	if err != nil {
	118	+	t.Fatal(err)
	119	+	}
	120	+	if cfg.FastMode == nil \|\| !*cfg.FastMode {
	121	+	t.Fatalf("fast_mode = %v, want true", cfg.FastMode)
	122	+	}
	123	+	if cfg.Theme != "dark" {
	124	+	t.Fatalf("unrelated config changed: theme = %q, want dark", cfg.Theme)
	125	+	}
	126	+	}
	127	+
	128	+	func TestFastModeConfigJSONUsesSnakeCase(t *testing.T) {
	129	+	enabled := true
	130	+	data, err := json.Marshal(Config{FastMode: &enabled})
	131	+	if err != nil {
	132	+	t.Fatal(err)
	133	+	}
	134	+	if !strings.Contains(string(data), `"fast_mode":true`) {
	135	+	t.Fatalf("config JSON = %s, want fast_mode", data)
	136	+	}
	137	+	}
	138	+
	139	+	func TestResolvedOllamaClientRejectsFastMode(t *testing.T) {
	140	+	client := (Resolved{
	141	+	Provider: "ollama",
	142	+	Credential: "ollama",
	143	+	BaseURL: "http://localhost:11434",
	144	+	}).NewClient()
	145	+	if client.Name() != "ollama" {
	146	+	t.Fatalf("client name = %q, want ollama", client.Name())
	147	+	}
	148	+	_, err := client.Stream(context.Background(), provider.Request{Model: "local", FastMode: true})
	149	+	if err == nil \|\| !strings.Contains(err.Error(), "only supported for OpenAI providers") {
	150	+	t.Fatalf("Stream error = %v, want unsupported-provider error", err)
	151	+	}
	152	+	}

@@ -0,0 +1,152 @@
	1	+	package agent
	2	+
	3	+	import (
	4	+	"context"
	5	+	"encoding/json"
	6	+	"strings"
	7	+	"testing"
	8	+
	9	+	"github.com/patriceckhart/zot/packages/provider"
	10	+	)
	11	+
	12	+	func TestParseArgsFastModeChildFlag(t *testing.T) {
	13	+	args, err := ParseArgs([]string{"--fast-mode"})
	14	+	if err != nil {
	15	+	t.Fatalf("ParseArgs failed: %v", err)
	16	+	}
	17	+	if !args.FastMode {
	18	+	t.Fatal("FastMode = false, want true")
	19	+	}
	20	+	if !args.FastModeSet {
	21	+	t.Fatal("FastModeSet = false, want true")
	22	+	}
	23	+	}
	24	+
	25	+	func TestResolveFastModeDefaultsOff(t *testing.T) {
	26	+	t.Setenv("ZOT_HOME", t.TempDir())
	27	+	if err := SaveConfig(Config{Provider: "openai", Model: "gpt-5"}); err != nil {
	28	+	t.Fatal(err)
	29	+	}
	30	+
	31	+	r, err := Resolve(Args{}, false)
	32	+	if err != nil {
	33	+	t.Fatalf("Resolve failed: %v", err)
	34	+	}
	35	+	if r.FastMode {
	36	+	t.Fatal("FastMode = true by default, want false")
	37	+	}
	38	+	}
	39	+
	40	+	func TestResolveFastModeFromConfig(t *testing.T) {
	41	+	t.Setenv("ZOT_HOME", t.TempDir())
	42	+	enabled := true
	43	+	if err := SaveConfig(Config{Provider: "openai", Model: "gpt-5", FastMode: &enabled}); err != nil {
	44	+	t.Fatal(err)
	45	+	}
	46	+
	47	+	r, err := Resolve(Args{}, false)
	48	+	if err != nil {
	49	+	t.Fatalf("Resolve failed: %v", err)
	50	+	}
	51	+	if !r.FastMode {
	52	+	t.Fatal("FastMode = false, want true from config")
	53	+	}
	54	+	}
	55	+
	56	+	func TestResolveFastModeRemainsEnabledForUnsupportedProviderUntilRequest(t *testing.T) {
	57	+	t.Setenv("ZOT_HOME", t.TempDir())
	58	+	enabled := true
	59	+	if err := SaveConfig(Config{Provider: "anthropic", Model: "claude-sonnet-4-5", FastMode: &enabled}); err != nil {
	60	+	t.Fatal(err)
	61	+	}
	62	+
	63	+	r, err := Resolve(Args{}, false)
	64	+	if err != nil {
	65	+	t.Fatalf("Resolve failed: %v", err)
	66	+	}
	67	+	if !r.FastMode {
	68	+	t.Fatal("FastMode = false, want true from config")
	69	+	}
	70	+	}
	71	+
	72	+	func TestResolveFastModeRequiresExplicitChildOverride(t *testing.T) {
	73	+	t.Setenv("ZOT_HOME", t.TempDir())
	74	+	if err := SaveConfig(Config{Provider: "openai", Model: "gpt-5"}); err != nil {
	75	+	t.Fatal(err)
	76	+	}
	77	+
	78	+	r, err := Resolve(Args{FastMode: true}, false)
	79	+	if err != nil {
	80	+	t.Fatalf("Resolve failed: %v", err)
	81	+	}
	82	+	if r.FastMode {
	83	+	t.Fatal("FastMode = true, want unmarked programmatic value to be ignored")
	84	+	}
	85	+	}
	86	+
	87	+	func TestResolveFastModeExplicitChildDisableOverridesConfig(t *testing.T) {
	88	+	t.Setenv("ZOT_HOME", t.TempDir())
	89	+	enabled := true
	90	+	if err := SaveConfig(Config{Provider: "openai", Model: "gpt-5", FastMode: &enabled}); err != nil {
	91	+	t.Fatal(err)
	92	+	}
	93	+	args, err := ParseArgs([]string{"--no-fast-mode"})
	94	+	if err != nil {
	95	+	t.Fatalf("ParseArgs failed: %v", err)
	96	+	}
	97	+
	98	+	r, err := Resolve(args, false)
	99	+	if err != nil {
	100	+	t.Fatalf("Resolve failed: %v", err)
	101	+	}
	102	+	if r.FastMode {
	103	+	t.Fatal("FastMode = true, want explicit child disable to override config")
	104	+	}
	105	+	}
	106	+
	107	+	func TestConfigSettingsStorePersistsFastMode(t *testing.T) {
	108	+	t.Setenv("ZOT_HOME", t.TempDir())
	109	+	if err := SaveConfig(Config{Theme: "dark"}); err != nil {
	110	+	t.Fatal(err)
	111	+	}
	112	+
	113	+	if err := (configSettingsStore{}).SetFastMode(true); err != nil {
	114	+	t.Fatal(err)
	115	+	}
	116	+	cfg, err := LoadConfig()
	117	+	if err != nil {
	118	+	t.Fatal(err)
	119	+	}
	120	+	if cfg.FastMode == nil \|\| !*cfg.FastMode {
	121	+	t.Fatalf("fast_mode = %v, want true", cfg.FastMode)
	122	+	}
	123	+	if cfg.Theme != "dark" {
	124	+	t.Fatalf("unrelated config changed: theme = %q, want dark", cfg.Theme)
	125	+	}
	126	+	}
	127	+
	128	+	func TestFastModeConfigJSONUsesSnakeCase(t *testing.T) {
	129	+	enabled := true
	130	+	data, err := json.Marshal(Config{FastMode: &enabled})
	131	+	if err != nil {
	132	+	t.Fatal(err)
	133	+	}
	134	+	if !strings.Contains(string(data), `"fast_mode":true`) {
	135	+	t.Fatalf("config JSON = %s, want fast_mode", data)
	136	+	}
	137	+	}
	138	+
	139	+	func TestResolvedOllamaClientRejectsFastMode(t *testing.T) {
	140	+	client := (Resolved{
	141	+	Provider: "ollama",
	142	+	Credential: "ollama",
	143	+	BaseURL: "http://localhost:11434",
	144	+	}).NewClient()
	145	+	if client.Name() != "ollama" {
	146	+	t.Fatalf("client name = %q, want ollama", client.Name())
	147	+	}
	148	+	_, err := client.Stream(context.Background(), provider.Request{Model: "local", FastMode: true})
	149	+	if err == nil \|\| !strings.Contains(err.Error(), "only supported for OpenAI providers") {
	150	+	t.Fatalf("Stream error = %v, want unsupported-provider error", err)
	151	+	}
	152	+	}

M packages/agent/modes/btw_dialog.go +1 -0


@@ -276,6 +276,7 @@ func newBtwAgent(main core.Agent, system, model string) core.Agent {
276	276		agent.MaxSteps = main.MaxSteps
277	277		agent.Reasoning = main.Reasoning
278	278		agent.Temperature = main.Temperature
	279	+	agent.FastMode = main.FastModeEnabled()
279	280		agent.MaxTokens = main.MaxTokens
280	281		agent.BeforeToolExecute = main.BeforeToolExecute
281	282		agent.BeforeTurn = main.BeforeTurn


@@ -276,6 +276,7 @@ func newBtwAgent(main core.Agent, system, model string) core.Agent {
276	agent.MaxSteps = main.MaxSteps	276		agent.MaxSteps = main.MaxSteps
277	agent.Reasoning = main.Reasoning	277		agent.Reasoning = main.Reasoning
278	agent.Temperature = main.Temperature	278		agent.Temperature = main.Temperature
		279	+	agent.FastMode = main.FastModeEnabled()
279	agent.MaxTokens = main.MaxTokens	280		agent.MaxTokens = main.MaxTokens
280	agent.BeforeToolExecute = main.BeforeToolExecute	281		agent.BeforeToolExecute = main.BeforeToolExecute
281	agent.BeforeTurn = main.BeforeTurn	282		agent.BeforeTurn = main.BeforeTurn

A packages/agent/modes/fast_mode_test.go +162 -0

@@ -0,0 +1,162 @@
	1	+	package modes
	2	+
	3	+	import (
	4	+	"context"
	5	+	"errors"
	6	+	"strings"
	7	+	"testing"
	8	+
	9	+	"github.com/patriceckhart/zot/packages/core"
	10	+	"github.com/patriceckhart/zot/packages/provider"
	11	+	)
	12	+
	13	+	type recordingFastModeSettingsStore struct {
	14	+	SettingsStore
	15	+	setCalls int
	16	+	last bool
	17	+	}
	18	+
	19	+	func (s *recordingFastModeSettingsStore) SetFastMode(enabled bool) error {
	20	+	s.setCalls++
	21	+	s.last = enabled
	22	+	return nil
	23	+	}
	24	+
	25	+	type failingFastModeSettingsStore struct {
	26	+	SettingsStore
	27	+	err error
	28	+	}
	29	+
	30	+	func (s *failingFastModeSettingsStore) SetFastMode(bool) error { return s.err }
	31	+
	32	+	type fastModeProviderClient struct{}
	33	+
	34	+	func (fastModeProviderClient) Name() string { return "openai" }
	35	+	func (fastModeProviderClient) Stream(_ context.Context, _ provider.Request) (<-chan provider.Event, error) {
	36	+	out := make(chan provider.Event)
	37	+	close(out)
	38	+	return out, nil
	39	+	}
	40	+
	41	+	func TestSettingsDialogOffersFastModeOffByDefault(t *testing.T) {
	42	+	interactive := NewInteractive(InteractiveConfig{Provider: "openai", Model: "gpt-5"})
	43	+	interactive.openSettingsDialog()
	44	+
	45	+	item := findSettingsItem(interactive.settingsDialog.items, "fast_mode")
	46	+	if item == nil {
	47	+	t.Fatal("settings dialog is missing fast mode")
	48	+	}
	49	+	if item.value {
	50	+	t.Fatal("fast mode is enabled by default")
	51	+	}
	52	+	}
	53	+
	54	+	func TestApplyFastModeUpdatesAgentAndStore(t *testing.T) {
	55	+	store := &recordingFastModeSettingsStore{}
	56	+	agent := core.NewAgent(fastModeProviderClient{}, "gpt-5", "", nil)
	57	+	interactive := NewInteractive(InteractiveConfig{
	58	+	Provider: "openai",
	59	+	Model: "gpt-5",
	60	+	SettingsStore: store,
	61	+	Agent: agent,
	62	+	})
	63	+	interactive.openSettingsDialog()
	64	+	interactive.rend = nil
	65	+
	66	+	interactive.applySettingToggle("fast_mode", true)
	67	+
	68	+	if interactive.cfg.FastMode == nil \|\| !*interactive.cfg.FastMode {
	69	+	t.Fatal("interactive FastMode was not enabled")
	70	+	}
	71	+	if !agent.FastMode {
	72	+	t.Fatal("agent FastMode was not enabled")
	73	+	}
	74	+	if store.setCalls != 1 \|\| !store.last {
	75	+	t.Fatalf("store calls = %d/%v, want one enabled write", store.setCalls, store.last)
	76	+	}
	77	+	}
	78	+
	79	+	func TestApplyFastModeRejectsUnsupportedProvider(t *testing.T) {
	80	+	store := &recordingFastModeSettingsStore{}
	81	+	interactive := NewInteractive(InteractiveConfig{
	82	+	Provider: "anthropic",
	83	+	Model: "claude-sonnet-4-5",
	84	+	SettingsStore: store,
	85	+	})
	86	+	interactive.openSettingsDialog()
	87	+	interactive.rend = nil
	88	+
	89	+	interactive.applySettingToggle("fast_mode", true)
	90	+
	91	+	if interactive.cfg.FastMode != nil && *interactive.cfg.FastMode {
	92	+	t.Fatal("unsupported fast mode was enabled")
	93	+	}
	94	+	if store.setCalls != 0 {
	95	+	t.Fatalf("store calls = %d, want none", store.setCalls)
	96	+	}
	97	+	if !strings.Contains(interactive.statusErr, "only supported for OpenAI providers") {
	98	+	t.Fatalf("status error = %q, want unsupported-provider error", interactive.statusErr)
	99	+	}
	100	+	}
	101	+
	102	+	func TestApplyFastModeCanDisableUnsupportedProvider(t *testing.T) {
	103	+	enabled := true
	104	+	store := &recordingFastModeSettingsStore{}
	105	+	agent := core.NewAgent(fastModeProviderClient{}, "claude-sonnet-4-5", "", nil)
	106	+	agent.FastMode = true
	107	+	interactive := NewInteractive(InteractiveConfig{
	108	+	Provider: "anthropic",
	109	+	Model: "claude-sonnet-4-5",
	110	+	FastMode: &enabled,
	111	+	SettingsStore: store,
	112	+	Agent: agent,
	113	+	})
	114	+	interactive.openSettingsDialog()
	115	+	interactive.rend = nil
	116	+
	117	+	interactive.applySettingToggle("fast_mode", false)
	118	+
	119	+	if interactive.cfg.FastMode == nil \|\| *interactive.cfg.FastMode {
	120	+	t.Fatal("fast mode was not disabled")
	121	+	}
	122	+	if agent.FastMode {
	123	+	t.Fatal("agent fast mode was not disabled")
	124	+	}
	125	+	if store.setCalls != 1 \|\| store.last {
	126	+	t.Fatalf("store calls = %d/%v, want one disabled write", store.setCalls, store.last)
	127	+	}
	128	+	}
	129	+
	130	+	func TestApplyFastModeLeavesLiveStateWhenPersistenceFails(t *testing.T) {
	131	+	store := &failingFastModeSettingsStore{err: errors.New("disk full")}
	132	+	agent := core.NewAgent(fastModeProviderClient{}, "gpt-5", "", nil)
	133	+	interactive := NewInteractive(InteractiveConfig{
	134	+	Provider: "openai",
	135	+	Model: "gpt-5",
	136	+	SettingsStore: store,
	137	+	Agent: agent,
	138	+	})
	139	+	interactive.openSettingsDialog()
	140	+	interactive.rend = nil
	141	+
	142	+	interactive.applySettingToggle("fast_mode", true)
	143	+
	144	+	if interactive.cfg.FastMode != nil && *interactive.cfg.FastMode {
	145	+	t.Fatal("fast mode changed despite persistence failure")
	146	+	}
	147	+	if agent.FastMode {
	148	+	t.Fatal("agent fast mode changed despite persistence failure")
	149	+	}
	150	+	if !strings.Contains(interactive.statusErr, "disk full") {
	151	+	t.Fatalf("status error = %q, want persistence error", interactive.statusErr)
	152	+	}
	153	+	}
	154	+
	155	+	func findSettingsItem(items []settingsItem, key string) *settingsItem {
	156	+	for idx := range items {
	157	+	if items[idx].key == key {
	158	+	return &items[idx]
	159	+	}
	160	+	}
	161	+	return nil
	162	+	}

@@ -0,0 +1,162 @@
	1	+	package modes
	2	+
	3	+	import (
	4	+	"context"
	5	+	"errors"
	6	+	"strings"
	7	+	"testing"
	8	+
	9	+	"github.com/patriceckhart/zot/packages/core"
	10	+	"github.com/patriceckhart/zot/packages/provider"
	11	+	)
	12	+
	13	+	type recordingFastModeSettingsStore struct {
	14	+	SettingsStore
	15	+	setCalls int
	16	+	last bool
	17	+	}
	18	+
	19	+	func (s *recordingFastModeSettingsStore) SetFastMode(enabled bool) error {
	20	+	s.setCalls++
	21	+	s.last = enabled
	22	+	return nil
	23	+	}
	24	+
	25	+	type failingFastModeSettingsStore struct {
	26	+	SettingsStore
	27	+	err error
	28	+	}
	29	+
	30	+	func (s *failingFastModeSettingsStore) SetFastMode(bool) error { return s.err }
	31	+
	32	+	type fastModeProviderClient struct{}
	33	+
	34	+	func (fastModeProviderClient) Name() string { return "openai" }
	35	+	func (fastModeProviderClient) Stream(_ context.Context, _ provider.Request) (<-chan provider.Event, error) {
	36	+	out := make(chan provider.Event)
	37	+	close(out)
	38	+	return out, nil
	39	+	}
	40	+
	41	+	func TestSettingsDialogOffersFastModeOffByDefault(t *testing.T) {
	42	+	interactive := NewInteractive(InteractiveConfig{Provider: "openai", Model: "gpt-5"})
	43	+	interactive.openSettingsDialog()
	44	+
	45	+	item := findSettingsItem(interactive.settingsDialog.items, "fast_mode")
	46	+	if item == nil {
	47	+	t.Fatal("settings dialog is missing fast mode")
	48	+	}
	49	+	if item.value {
	50	+	t.Fatal("fast mode is enabled by default")
	51	+	}
	52	+	}
	53	+
	54	+	func TestApplyFastModeUpdatesAgentAndStore(t *testing.T) {
	55	+	store := &recordingFastModeSettingsStore{}
	56	+	agent := core.NewAgent(fastModeProviderClient{}, "gpt-5", "", nil)
	57	+	interactive := NewInteractive(InteractiveConfig{
	58	+	Provider: "openai",
	59	+	Model: "gpt-5",
	60	+	SettingsStore: store,
	61	+	Agent: agent,
	62	+	})
	63	+	interactive.openSettingsDialog()
	64	+	interactive.rend = nil
	65	+
	66	+	interactive.applySettingToggle("fast_mode", true)
	67	+
	68	+	if interactive.cfg.FastMode == nil \|\| !*interactive.cfg.FastMode {
	69	+	t.Fatal("interactive FastMode was not enabled")
	70	+	}
	71	+	if !agent.FastMode {
	72	+	t.Fatal("agent FastMode was not enabled")
	73	+	}
	74	+	if store.setCalls != 1 \|\| !store.last {
	75	+	t.Fatalf("store calls = %d/%v, want one enabled write", store.setCalls, store.last)
	76	+	}
	77	+	}
	78	+
	79	+	func TestApplyFastModeRejectsUnsupportedProvider(t *testing.T) {
	80	+	store := &recordingFastModeSettingsStore{}
	81	+	interactive := NewInteractive(InteractiveConfig{
	82	+	Provider: "anthropic",
	83	+	Model: "claude-sonnet-4-5",
	84	+	SettingsStore: store,
	85	+	})
	86	+	interactive.openSettingsDialog()
	87	+	interactive.rend = nil
	88	+
	89	+	interactive.applySettingToggle("fast_mode", true)
	90	+
	91	+	if interactive.cfg.FastMode != nil && *interactive.cfg.FastMode {
	92	+	t.Fatal("unsupported fast mode was enabled")
	93	+	}
	94	+	if store.setCalls != 0 {
	95	+	t.Fatalf("store calls = %d, want none", store.setCalls)
	96	+	}
	97	+	if !strings.Contains(interactive.statusErr, "only supported for OpenAI providers") {
	98	+	t.Fatalf("status error = %q, want unsupported-provider error", interactive.statusErr)
	99	+	}
	100	+	}
	101	+
	102	+	func TestApplyFastModeCanDisableUnsupportedProvider(t *testing.T) {
	103	+	enabled := true
	104	+	store := &recordingFastModeSettingsStore{}
	105	+	agent := core.NewAgent(fastModeProviderClient{}, "claude-sonnet-4-5", "", nil)
	106	+	agent.FastMode = true
	107	+	interactive := NewInteractive(InteractiveConfig{
	108	+	Provider: "anthropic",
	109	+	Model: "claude-sonnet-4-5",
	110	+	FastMode: &enabled,
	111	+	SettingsStore: store,
	112	+	Agent: agent,
	113	+	})
	114	+	interactive.openSettingsDialog()
	115	+	interactive.rend = nil
	116	+
	117	+	interactive.applySettingToggle("fast_mode", false)
	118	+
	119	+	if interactive.cfg.FastMode == nil \|\| *interactive.cfg.FastMode {
	120	+	t.Fatal("fast mode was not disabled")
	121	+	}
	122	+	if agent.FastMode {
	123	+	t.Fatal("agent fast mode was not disabled")
	124	+	}
	125	+	if store.setCalls != 1 \|\| store.last {
	126	+	t.Fatalf("store calls = %d/%v, want one disabled write", store.setCalls, store.last)
	127	+	}
	128	+	}
	129	+
	130	+	func TestApplyFastModeLeavesLiveStateWhenPersistenceFails(t *testing.T) {
	131	+	store := &failingFastModeSettingsStore{err: errors.New("disk full")}
	132	+	agent := core.NewAgent(fastModeProviderClient{}, "gpt-5", "", nil)
	133	+	interactive := NewInteractive(InteractiveConfig{
	134	+	Provider: "openai",
	135	+	Model: "gpt-5",
	136	+	SettingsStore: store,
	137	+	Agent: agent,
	138	+	})
	139	+	interactive.openSettingsDialog()
	140	+	interactive.rend = nil
	141	+
	142	+	interactive.applySettingToggle("fast_mode", true)
	143	+
	144	+	if interactive.cfg.FastMode != nil && *interactive.cfg.FastMode {
	145	+	t.Fatal("fast mode changed despite persistence failure")
	146	+	}
	147	+	if agent.FastMode {
	148	+	t.Fatal("agent fast mode changed despite persistence failure")
	149	+	}
	150	+	if !strings.Contains(interactive.statusErr, "disk full") {
	151	+	t.Fatalf("status error = %q, want persistence error", interactive.statusErr)
	152	+	}
	153	+	}
	154	+
	155	+	func findSettingsItem(items []settingsItem, key string) *settingsItem {
	156	+	for idx := range items {
	157	+	if items[idx].key == key {
	158	+	return &items[idx]
	159	+	}
	160	+	}
	161	+	return nil
	162	+	}

M packages/agent/modes/interactive.go +64 -0


@@ -68,6 +68,11 @@ type InteractiveConfig struct {
68	68		// re-reading config.json on every open.
69	69		AutoSwarmEnabled *bool
70	70
	71	+	// FastMode mirrors the persisted OpenAI fast-mode flag at startup.
	72	+	// nil/missing means disabled. Unsupported providers reject attempts
	73	+	// to enable it and reject requests when it remains enabled.
	74	+	FastMode *bool
	75	+
71	76		// AutoCompactThreshold is the context-window percentage that triggers
72	77		// automatic compaction. nil means 85; zero disables percentage-based
73	78		// triggers while preserving payload-too-large recovery.

@@ -370,6 +375,10 @@ type autoCompactThresholdSettingsStore interface {
370	375		SetAutoCompactThreshold(percent int) error
371	376		}
372	377
	378	+	type fastModeSettingsStore interface {
	379	+	SetFastMode(enabled bool) error
	380	+	}
	381	+
373	382		type Interactive struct {
374	383		cfg InteractiveConfig
375	384		view *tui.View

@@ -3411,6 +3420,12 @@ func (i *Interactive) openSettingsDialog() {
3411	3420		autoSwarmHint = "swarm supervisor not available in this mode"
3412	3421		}
3413	3422
	3423	+	fastMode := i.cfg.FastMode != nil && *i.cfg.FastMode
	3424	+	fastModeHint := "OpenAI service tier"
	3425	+	if !provider.SupportsFastMode(i.cfg.Provider) {
	3426	+	fastModeHint = "only supported for OpenAI providers"
	3427	+	}
	3428	+
3414	3429		jailByDefault := i.cfg.JailByDefault != nil && *i.cfg.JailByDefault
3415	3430		recursiveFiles := i.cfg.RecursiveFileSuggest != nil && *i.cfg.RecursiveFileSuggest
3416	3431		respectGitignore := i.cfg.RespectGitignore == nil \|\| *i.cfg.RespectGitignore

@@ -3523,6 +3538,13 @@ func (i *Interactive) openSettingsDialog() {
3523	3538		disabled: autoSwarmDisabled,
3524	3539		hint: autoSwarmHint,
3525	3540		},
	3541	+	{
	3542	+	key: "fast_mode",
	3543	+	label: "fast mode",
	3544	+	desc: "request OpenAI's fast service tier; unsupported providers return an error",
	3545	+	value: fastMode,
	3546	+	hint: fastModeHint,
	3547	+	},
3526	3548		{
3527	3549		key: "auto_compact_threshold",
3528	3550		label: "auto-compact threshold",

@@ -3821,6 +3843,18 @@ func (i *Interactive) refreshQuickModelSettingsItem(slot int) {
3821	3843		}
3822	3844		}
3823	3845
	3846	+	func (i *Interactive) resetSettingsToggle(key string, value bool) {
	3847	+	if i.settingsDialog == nil {
	3848	+	return
	3849	+	}
	3850	+	for idx := range i.settingsDialog.items {
	3851	+	if i.settingsDialog.items[idx].key == key {
	3852	+	i.settingsDialog.items[idx].value = value
	3853	+	return
	3854	+	}
	3855	+	}
	3856	+	}
	3857	+
3824	3858		func (i *Interactive) applySettingToggle(key string, value bool) {
3825	3859		// Every setting toggle forces a full repaint at the end — same
3826	3860		// effect as the user pressing Ctrl+L — so any per-setting visual

@@ -3891,6 +3925,36 @@ func (i *Interactive) applySettingToggle(key string, value bool) {
3891	3925		i.statusOK = "auto-swarm " + onOff(value)
3892	3926		i.statusErr = ""
3893	3927		i.mu.Unlock()
	3928	+	case "fast_mode":
	3929	+	previous := i.cfg.FastMode != nil && *i.cfg.FastMode
	3930	+	if err := provider.ValidateFastMode(i.cfg.Provider, value); err != nil {
	3931	+	i.resetSettingsToggle(key, previous)
	3932	+	i.mu.Lock()
	3933	+	i.statusErr = "settings: " + err.Error()
	3934	+	i.mu.Unlock()
	3935	+	return
	3936	+	}
	3937	+	if store, ok := i.cfg.SettingsStore.(fastModeSettingsStore); ok {
	3938	+	if err := store.SetFastMode(value); err != nil {
	3939	+	i.resetSettingsToggle(key, previous)
	3940	+	i.mu.Lock()
	3941	+	i.statusErr = "settings: " + err.Error()
	3942	+	i.mu.Unlock()
	3943	+	return
	3944	+	}
	3945	+	}
	3946	+	val := value
	3947	+	i.cfg.FastMode = &val
	3948	+	if i.agent != nil {
	3949	+	i.agent.SetFastMode(value)
	3950	+	}
	3951	+	if i.cfg.Swarm != nil {
	3952	+	i.cfg.Swarm.SetFastMode(value)
	3953	+	}
	3954	+	i.mu.Lock()
	3955	+	i.statusOK = "fast mode " + onOff(value)
	3956	+	i.statusErr = ""
	3957	+	i.mu.Unlock()
3894	3958		case "jail_by_default":
3895	3959		val := value
3896	3960		i.cfg.JailByDefault = &val


@@ -68,6 +68,11 @@ type InteractiveConfig struct {
68	// re-reading config.json on every open.	68		// re-reading config.json on every open.
69	AutoSwarmEnabled *bool	69		AutoSwarmEnabled *bool
70		70
		71	+	// FastMode mirrors the persisted OpenAI fast-mode flag at startup.
		72	+	// nil/missing means disabled. Unsupported providers reject attempts
		73	+	// to enable it and reject requests when it remains enabled.
		74	+	FastMode *bool
		75	+
71	// AutoCompactThreshold is the context-window percentage that triggers	76		// AutoCompactThreshold is the context-window percentage that triggers
72	// automatic compaction. nil means 85; zero disables percentage-based	77		// automatic compaction. nil means 85; zero disables percentage-based
73	// triggers while preserving payload-too-large recovery.	78		// triggers while preserving payload-too-large recovery.

@@ -370,6 +375,10 @@ type autoCompactThresholdSettingsStore interface {
370	SetAutoCompactThreshold(percent int) error	375		SetAutoCompactThreshold(percent int) error
371	}	376		}
372		377
		378	+	type fastModeSettingsStore interface {
		379	+	SetFastMode(enabled bool) error
		380	+	}
		381	+
373	type Interactive struct {	382		type Interactive struct {
374	cfg InteractiveConfig	383		cfg InteractiveConfig
375	view *tui.View	384		view *tui.View

@@ -3411,6 +3420,12 @@ func (i *Interactive) openSettingsDialog() {
3411	autoSwarmHint = "swarm supervisor not available in this mode"	3420		autoSwarmHint = "swarm supervisor not available in this mode"
3412	}	3421		}
3413		3422
		3423	+	fastMode := i.cfg.FastMode != nil && *i.cfg.FastMode
		3424	+	fastModeHint := "OpenAI service tier"
		3425	+	if !provider.SupportsFastMode(i.cfg.Provider) {
		3426	+	fastModeHint = "only supported for OpenAI providers"
		3427	+	}
		3428	+
3414	jailByDefault := i.cfg.JailByDefault != nil && *i.cfg.JailByDefault	3429		jailByDefault := i.cfg.JailByDefault != nil && *i.cfg.JailByDefault
3415	recursiveFiles := i.cfg.RecursiveFileSuggest != nil && *i.cfg.RecursiveFileSuggest	3430		recursiveFiles := i.cfg.RecursiveFileSuggest != nil && *i.cfg.RecursiveFileSuggest
3416	respectGitignore := i.cfg.RespectGitignore == nil \|\| *i.cfg.RespectGitignore	3431		respectGitignore := i.cfg.RespectGitignore == nil \|\| *i.cfg.RespectGitignore

@@ -3523,6 +3538,13 @@ func (i *Interactive) openSettingsDialog() {
3523	disabled: autoSwarmDisabled,	3538		disabled: autoSwarmDisabled,
3524	hint: autoSwarmHint,	3539		hint: autoSwarmHint,
3525	},	3540		},
		3541	+	{
		3542	+	key: "fast_mode",
		3543	+	label: "fast mode",
		3544	+	desc: "request OpenAI's fast service tier; unsupported providers return an error",
		3545	+	value: fastMode,
		3546	+	hint: fastModeHint,
		3547	+	},
3526	{	3548		{
3527	key: "auto_compact_threshold",	3549		key: "auto_compact_threshold",
3528	label: "auto-compact threshold",	3550		label: "auto-compact threshold",

@@ -3821,6 +3843,18 @@ func (i *Interactive) refreshQuickModelSettingsItem(slot int) {
3821	}	3843		}
3822	}	3844		}
3823		3845
		3846	+	func (i *Interactive) resetSettingsToggle(key string, value bool) {
		3847	+	if i.settingsDialog == nil {
		3848	+	return
		3849	+	}
		3850	+	for idx := range i.settingsDialog.items {
		3851	+	if i.settingsDialog.items[idx].key == key {
		3852	+	i.settingsDialog.items[idx].value = value
		3853	+	return
		3854	+	}
		3855	+	}
		3856	+	}
		3857	+
3824	func (i *Interactive) applySettingToggle(key string, value bool) {	3858		func (i *Interactive) applySettingToggle(key string, value bool) {
3825	// Every setting toggle forces a full repaint at the end — same	3859		// Every setting toggle forces a full repaint at the end — same
3826	// effect as the user pressing Ctrl+L — so any per-setting visual	3860		// effect as the user pressing Ctrl+L — so any per-setting visual

@@ -3891,6 +3925,36 @@ func (i *Interactive) applySettingToggle(key string, value bool) {
3891	i.statusOK = "auto-swarm " + onOff(value)	3925		i.statusOK = "auto-swarm " + onOff(value)
3892	i.statusErr = ""	3926		i.statusErr = ""
3893	i.mu.Unlock()	3927		i.mu.Unlock()
		3928	+	case "fast_mode":
		3929	+	previous := i.cfg.FastMode != nil && *i.cfg.FastMode
		3930	+	if err := provider.ValidateFastMode(i.cfg.Provider, value); err != nil {
		3931	+	i.resetSettingsToggle(key, previous)
		3932	+	i.mu.Lock()
		3933	+	i.statusErr = "settings: " + err.Error()
		3934	+	i.mu.Unlock()
		3935	+	return
		3936	+	}
		3937	+	if store, ok := i.cfg.SettingsStore.(fastModeSettingsStore); ok {
		3938	+	if err := store.SetFastMode(value); err != nil {
		3939	+	i.resetSettingsToggle(key, previous)
		3940	+	i.mu.Lock()
		3941	+	i.statusErr = "settings: " + err.Error()
		3942	+	i.mu.Unlock()
		3943	+	return
		3944	+	}
		3945	+	}
		3946	+	val := value
		3947	+	i.cfg.FastMode = &val
		3948	+	if i.agent != nil {
		3949	+	i.agent.SetFastMode(value)
		3950	+	}
		3951	+	if i.cfg.Swarm != nil {
		3952	+	i.cfg.Swarm.SetFastMode(value)
		3953	+	}
		3954	+	i.mu.Lock()
		3955	+	i.statusOK = "fast mode " + onOff(value)
		3956	+	i.statusErr = ""
		3957	+	i.mu.Unlock()
3894	case "jail_by_default":	3958		case "jail_by_default":
3895	val := value	3959		val := value
3896	i.cfg.JailByDefault = &val	3960		i.cfg.JailByDefault = &val

M packages/agent/settings_store.go +9 -0


@@ -59,6 +59,15 @@ func (configSettingsStore) SetAutoSwarm(enabled bool) error {
59	59		return SaveConfig(cfg)
60	60		}
61	61
	62	+	func (configSettingsStore) SetFastMode(enabled bool) error {
	63	+	cfg, err := LoadConfig()
	64	+	if err != nil {
	65	+	return err
	66	+	}
	67	+	cfg.FastMode = &enabled
	68	+	return SaveConfig(cfg)
	69	+	}
	70	+
62	71		func (configSettingsStore) SetAutoCompactThreshold(percent int) error {
63	72		cfg, err := LoadConfig()
64	73		if err != nil {


@@ -59,6 +59,15 @@ func (configSettingsStore) SetAutoSwarm(enabled bool) error {
59	return SaveConfig(cfg)	59		return SaveConfig(cfg)
60	}	60		}
61		61
		62	+	func (configSettingsStore) SetFastMode(enabled bool) error {
		63	+	cfg, err := LoadConfig()
		64	+	if err != nil {
		65	+	return err
		66	+	}
		67	+	cfg.FastMode = &enabled
		68	+	return SaveConfig(cfg)
		69	+	}
		70	+
62	func (configSettingsStore) SetAutoCompactThreshold(percent int) error {	71		func (configSettingsStore) SetAutoCompactThreshold(percent int) error {
63	cfg, err := LoadConfig()	72		cfg, err := LoadConfig()
64	if err != nil {	73		if err != nil {

M packages/agent/swarm/agent.go +1 -0


@@ -25,6 +25,7 @@ type Agent struct {
25	25		Model string
26	26		Provider string
27	27		Reasoning string
	28	+	FastMode bool
28	29		Subagent string
29	30
30	31		// SessionID, when non-empty, scopes the agent to a particular


@@ -25,6 +25,7 @@ type Agent struct {
25	Model string	25		Model string
26	Provider string	26		Provider string
27	Reasoning string	27		Reasoning string
		28	+	FastMode bool
28	Subagent string	29		Subagent string
29		30
30	// SessionID, when non-empty, scopes the agent to a particular	31		// SessionID, when non-empty, scopes the agent to a particular

A packages/agent/swarm/fast_mode_test.go +136 -0

@@ -0,0 +1,136 @@
	1	+	package swarm
	2	+
	3	+	import (
	4	+	"context"
	5	+	"strings"
	6	+	"testing"
	7	+	)
	8	+
	9	+	func TestSwarmFastModePropagatesToChild(t *testing.T) {
	10	+	root := t.TempDir()
	11	+	f := New(Config{
	12	+	Root: root,
	13	+	RepoRoot: root,
	14	+	FastMode: true,
	15	+	NewRunner: func(a *Agent) Runner {
	16	+	return RunnerFunc(func(ctx context.Context, _ Sink) error {
	17	+	<-ctx.Done()
	18	+	return ctx.Err()
	19	+	})
	20	+	},
	21	+	})
	22	+
	23	+	a, err := f.SpawnReq(context.Background(), SpawnRequest{Task: "x"})
	24	+	if err != nil {
	25	+	t.Fatalf("spawn: %v", err)
	26	+	}
	27	+	cleanedUp := false
	28	+	t.Cleanup(func() {
	29	+	if cleanedUp {
	30	+	return
	31	+	}
	32	+	if err := f.Stop(a.ID); err != nil {
	33	+	t.Errorf("cleanup stop: %v", err)
	34	+	}
	35	+	a.Wait()
	36	+	})
	37	+	if !a.FastMode {
	38	+	t.Fatal("spawned agent FastMode = false, want true")
	39	+	}
	40	+	args := defaultChildArgs("/zot", a, "/session", "/inbox")
	41	+	if !containsArg(args, "--fast-mode") {
	42	+	t.Fatalf("child args = %v, want --fast-mode", args)
	43	+	}
	44	+
	45	+	if err := f.Stop(a.ID); err != nil {
	46	+	t.Fatalf("stop: %v", err)
	47	+	}
	48	+	a.Wait()
	49	+	cleanedUp = true
	50	+
	51	+	g := New(Config{Root: root, RepoRoot: root})
	52	+	if loaded, errs := g.Reload(); loaded != 1 \|\| len(errs) != 0 {
	53	+	t.Fatalf("reload loaded=%d errs=%v", loaded, errs)
	54	+	}
	55	+	reloaded := g.Get(a.ID)
	56	+	if reloaded == nil \|\| !reloaded.FastMode {
	57	+	t.Fatalf("reloaded FastMode = %v, want true", reloaded != nil && reloaded.FastMode)
	58	+	}
	59	+	}
	60	+
	61	+	func TestSwarmReloadPreservesExplicitFastModeOff(t *testing.T) {
	62	+	root := t.TempDir()
	63	+	newSwarm := func(fastMode bool) *Swarm {
	64	+	return New(Config{
	65	+	Root: root,
	66	+	RepoRoot: root,
	67	+	FastMode: fastMode,
	68	+	NewRunner: func(a *Agent) Runner {
	69	+	return RunnerFunc(func(ctx context.Context, _ Sink) error {
	70	+	<-ctx.Done()
	71	+	return ctx.Err()
	72	+	})
	73	+	},
	74	+	})
	75	+	}
	76	+
	77	+	f := newSwarm(false)
	78	+	a, err := f.SpawnReq(context.Background(), SpawnRequest{Task: "x"})
	79	+	if err != nil {
	80	+	t.Fatalf("spawn: %v", err)
	81	+	}
	82	+	if a.FastMode {
	83	+	t.Fatal("spawned agent FastMode = true, want false")
	84	+	}
	85	+	if err := f.Stop(a.ID); err != nil {
	86	+	t.Fatalf("stop: %v", err)
	87	+	}
	88	+	a.Wait()
	89	+
	90	+	g := newSwarm(true)
	91	+	if loaded, errs := g.Reload(); loaded != 1 \|\| len(errs) != 0 {
	92	+	t.Fatalf("reload loaded=%d errs=%v", loaded, errs)
	93	+	}
	94	+	if reloaded := g.Get(a.ID); reloaded == nil \|\| reloaded.FastMode {
	95	+	t.Fatalf("reloaded FastMode = %v, want false", reloaded != nil && reloaded.FastMode)
	96	+	}
	97	+	}
	98	+
	99	+	func TestSwarmAgentArgsOmitFastModeWhenUnset(t *testing.T) {
	100	+	args := swarmAgentArgs(swarmAgentArgsOpts{Exe: "/zot", Dir: "/wt", SessionPath: "/s", InboxPath: "/i"})
	101	+	if strings.Contains(strings.Join(args, " "), "--fast-mode") \|\| strings.Contains(strings.Join(args, " "), "--no-fast-mode") {
	102	+	t.Fatalf("child args = %v, want fast mode omitted when unset", args)
	103	+	}
	104	+	}
	105	+
	106	+	func TestSwarmAgentArgsExplicitlyDisablesFastMode(t *testing.T) {
	107	+	args := swarmAgentArgs(swarmAgentArgsOpts{
	108	+	Exe: "/zot",
	109	+	Dir: "/wt",
	110	+	SessionPath: "/s",
	111	+	InboxPath: "/i",
	112	+	FastModeSet: true,
	113	+	})
	114	+	if !containsArg(args, "--no-fast-mode") {
	115	+	t.Fatalf("child args = %v, want --no-fast-mode", args)
	116	+	}
	117	+	if containsArg(args, "--fast-mode") {
	118	+	t.Fatalf("child args = %v, want no positive fast-mode flag", args)
	119	+	}
	120	+	}
	121	+
	122	+	func TestDefaultChildArgsExplicitlyDisablesFastMode(t *testing.T) {
	123	+	args := defaultChildArgs("/zot", &Agent{Dir: "/wt"}, "/s", "/i")
	124	+	if !containsArg(args, "--no-fast-mode") {
	125	+	t.Fatalf("child args = %v, want --no-fast-mode", args)
	126	+	}
	127	+	}
	128	+
	129	+	func containsArg(args []string, want string) bool {
	130	+	for _, arg := range args {
	131	+	if arg == want {
	132	+	return true
	133	+	}
	134	+	}
	135	+	return false
	136	+	}

@@ -0,0 +1,136 @@
	1	+	package swarm
	2	+
	3	+	import (
	4	+	"context"
	5	+	"strings"
	6	+	"testing"
	7	+	)
	8	+
	9	+	func TestSwarmFastModePropagatesToChild(t *testing.T) {
	10	+	root := t.TempDir()
	11	+	f := New(Config{
	12	+	Root: root,
	13	+	RepoRoot: root,
	14	+	FastMode: true,
	15	+	NewRunner: func(a *Agent) Runner {
	16	+	return RunnerFunc(func(ctx context.Context, _ Sink) error {
	17	+	<-ctx.Done()
	18	+	return ctx.Err()
	19	+	})
	20	+	},
	21	+	})
	22	+
	23	+	a, err := f.SpawnReq(context.Background(), SpawnRequest{Task: "x"})
	24	+	if err != nil {
	25	+	t.Fatalf("spawn: %v", err)
	26	+	}
	27	+	cleanedUp := false
	28	+	t.Cleanup(func() {
	29	+	if cleanedUp {
	30	+	return
	31	+	}
	32	+	if err := f.Stop(a.ID); err != nil {
	33	+	t.Errorf("cleanup stop: %v", err)
	34	+	}
	35	+	a.Wait()
	36	+	})
	37	+	if !a.FastMode {
	38	+	t.Fatal("spawned agent FastMode = false, want true")
	39	+	}
	40	+	args := defaultChildArgs("/zot", a, "/session", "/inbox")
	41	+	if !containsArg(args, "--fast-mode") {
	42	+	t.Fatalf("child args = %v, want --fast-mode", args)
	43	+	}
	44	+
	45	+	if err := f.Stop(a.ID); err != nil {
	46	+	t.Fatalf("stop: %v", err)
	47	+	}
	48	+	a.Wait()
	49	+	cleanedUp = true
	50	+
	51	+	g := New(Config{Root: root, RepoRoot: root})
	52	+	if loaded, errs := g.Reload(); loaded != 1 \|\| len(errs) != 0 {
	53	+	t.Fatalf("reload loaded=%d errs=%v", loaded, errs)
	54	+	}
	55	+	reloaded := g.Get(a.ID)
	56	+	if reloaded == nil \|\| !reloaded.FastMode {
	57	+	t.Fatalf("reloaded FastMode = %v, want true", reloaded != nil && reloaded.FastMode)
	58	+	}
	59	+	}
	60	+
	61	+	func TestSwarmReloadPreservesExplicitFastModeOff(t *testing.T) {
	62	+	root := t.TempDir()
	63	+	newSwarm := func(fastMode bool) *Swarm {
	64	+	return New(Config{
	65	+	Root: root,
	66	+	RepoRoot: root,
	67	+	FastMode: fastMode,
	68	+	NewRunner: func(a *Agent) Runner {
	69	+	return RunnerFunc(func(ctx context.Context, _ Sink) error {
	70	+	<-ctx.Done()
	71	+	return ctx.Err()
	72	+	})
	73	+	},
	74	+	})
	75	+	}
	76	+
	77	+	f := newSwarm(false)
	78	+	a, err := f.SpawnReq(context.Background(), SpawnRequest{Task: "x"})
	79	+	if err != nil {
	80	+	t.Fatalf("spawn: %v", err)
	81	+	}
	82	+	if a.FastMode {
	83	+	t.Fatal("spawned agent FastMode = true, want false")
	84	+	}
	85	+	if err := f.Stop(a.ID); err != nil {
	86	+	t.Fatalf("stop: %v", err)
	87	+	}
	88	+	a.Wait()
	89	+
	90	+	g := newSwarm(true)
	91	+	if loaded, errs := g.Reload(); loaded != 1 \|\| len(errs) != 0 {
	92	+	t.Fatalf("reload loaded=%d errs=%v", loaded, errs)
	93	+	}
	94	+	if reloaded := g.Get(a.ID); reloaded == nil \|\| reloaded.FastMode {
	95	+	t.Fatalf("reloaded FastMode = %v, want false", reloaded != nil && reloaded.FastMode)
	96	+	}
	97	+	}
	98	+
	99	+	func TestSwarmAgentArgsOmitFastModeWhenUnset(t *testing.T) {
	100	+	args := swarmAgentArgs(swarmAgentArgsOpts{Exe: "/zot", Dir: "/wt", SessionPath: "/s", InboxPath: "/i"})
	101	+	if strings.Contains(strings.Join(args, " "), "--fast-mode") \|\| strings.Contains(strings.Join(args, " "), "--no-fast-mode") {
	102	+	t.Fatalf("child args = %v, want fast mode omitted when unset", args)
	103	+	}
	104	+	}
	105	+
	106	+	func TestSwarmAgentArgsExplicitlyDisablesFastMode(t *testing.T) {
	107	+	args := swarmAgentArgs(swarmAgentArgsOpts{
	108	+	Exe: "/zot",
	109	+	Dir: "/wt",
	110	+	SessionPath: "/s",
	111	+	InboxPath: "/i",
	112	+	FastModeSet: true,
	113	+	})
	114	+	if !containsArg(args, "--no-fast-mode") {
	115	+	t.Fatalf("child args = %v, want --no-fast-mode", args)
	116	+	}
	117	+	if containsArg(args, "--fast-mode") {
	118	+	t.Fatalf("child args = %v, want no positive fast-mode flag", args)
	119	+	}
	120	+	}
	121	+
	122	+	func TestDefaultChildArgsExplicitlyDisablesFastMode(t *testing.T) {
	123	+	args := defaultChildArgs("/zot", &Agent{Dir: "/wt"}, "/s", "/i")
	124	+	if !containsArg(args, "--no-fast-mode") {
	125	+	t.Fatalf("child args = %v, want --no-fast-mode", args)
	126	+	}
	127	+	}
	128	+
	129	+	func containsArg(args []string, want string) bool {
	130	+	for _, arg := range args {
	131	+	if arg == want {
	132	+	return true
	133	+	}
	134	+	}
	135	+	return false
	136	+	}

M packages/agent/swarm/persist.go +12 -1


@@ -45,6 +45,7 @@ type agentMeta struct {
45	45		Model string `json:"model,omitempty"`
46	46		Provider string `json:"provider,omitempty"`
47	47		Reasoning string `json:"reasoning,omitempty"`
	48	+	FastMode *bool `json:"fast_mode,omitempty"`
48	49		Subagent string `json:"subagent,omitempty"`
49	50		InboxPath string `json:"inbox_path"`
50	51		EventLogPath string `json:"event_log_path"`

@@ -66,6 +67,7 @@ func metaPath(stateDir string) string { return filepath.Join(stateDir, "meta.jso
66	67		// write is atomic (tmp + rename) so a crash mid-write can't leave a
67	68		// half-parsable file that fails Reload.
68	69		func writeAgentMeta(stateDir string, a *Agent) error {
	70	+	fastMode := a.FastMode
69	71		m := agentMeta{
70	72		ID: a.ID,
71	73		Task: a.Task,

@@ -74,6 +76,7 @@ func writeAgentMeta(stateDir string, a *Agent) error {
74	76		Model: a.Model,
75	77		Provider: a.Provider,
76	78		Reasoning: a.Reasoning,
	79	+	FastMode: &fastMode,
77	80		Subagent: a.Subagent,
78	81		InboxPath: a.InboxPath,
79	82		EventLogPath: a.EventLogPath,

@@ -202,6 +205,10 @@ func (f Swarm) buildDetachedAgent(m agentMeta) Agent {
202	205		if f.cfg.RepoRoot != "" {
203	206		dir = f.cfg.RepoRoot
204	207		}
	208	+	fastMode := f.cfg.FastMode
	209	+	if m.FastMode != nil {
	210	+	fastMode = *m.FastMode
	211	+	}
205	212		a := &Agent{
206	213		ID: m.ID,
207	214		Task: m.Task,

@@ -210,6 +217,7 @@ func (f Swarm) buildDetachedAgent(m agentMeta) Agent {
210	217		Model: m.Model,
211	218		Provider: m.Provider,
212	219		Reasoning: m.Reasoning,
	220	+	FastMode: fastMode,
213	221		Subagent: m.Subagent,
214	222		InboxPath: m.InboxPath,
215	223		EventLogPath: m.EventLogPath,

@@ -348,11 +356,13 @@ func (f Swarm) Resume(ctx context.Context, id string) (Agent, error) {
348	356		if err != nil {
349	357		return nil, fmt.Errorf("swarm inbox path: %w", err)
350	358		}
	359	+	fastMode := existing.FastMode
351	360		m := agentMeta{
352	361		ID: existing.ID, Task: existing.Task,
353	362		Dir: existing.Dir, Started: existing.Started,
354	363		Model: existing.Model, Provider: existing.Provider,
355		-	Reasoning: existing.Reasoning, Subagent: existing.Subagent,
	364	+	Reasoning: existing.Reasoning, FastMode: &fastMode,
	365	+	Subagent: existing.Subagent,
356	366		SessionID: existing.SessionID,
357	367		InboxPath: inboxPath, EventLogPath: existing.EventLogPath,
358	368		SessionPath: existing.SessionPath,

@@ -366,6 +376,7 @@ func (f Swarm) Resume(ctx context.Context, id string) (Agent, error) {
366	376		Model: m.Model,
367	377		Provider: m.Provider,
368	378		Reasoning: m.Reasoning,
	379	+	FastMode: fastMode,
369	380		Subagent: m.Subagent,
370	381		SessionID: m.SessionID,
371	382		InboxPath: m.InboxPath,

 	Model        string    `json:"model,omitempty"`
 	Provider     string    `json:"provider,omitempty"`
 	Reasoning    string    `json:"reasoning,omitempty"`
+	FastMode     *bool     `json:"fast_mode,omitempty"`
 	Subagent     string    `json:"subagent,omitempty"`
 	InboxPath    string    `json:"inbox_path"`
 	EventLogPath string    `json:"event_log_path"`
 // write is atomic (tmp + rename) so a crash mid-write can't leave a
 // half-parsable file that fails Reload.
 func writeAgentMeta(stateDir string, a *Agent) error {
+	fastMode := a.FastMode
 	m := agentMeta{
 		ID:           a.ID,
 		Task:         a.Task,
 		Model:        a.Model,
 		Provider:     a.Provider,
 		Reasoning:    a.Reasoning,
+		FastMode:     &fastMode,
 		Subagent:     a.Subagent,
 		InboxPath:    a.InboxPath,
 		EventLogPath: a.EventLogPath,
 	if f.cfg.RepoRoot != "" {
 		dir = f.cfg.RepoRoot
 	}
+	fastMode := f.cfg.FastMode
+	if m.FastMode != nil {
+		fastMode = *m.FastMode
+	}
 	a := &Agent{
 		ID:           m.ID,
 		Task:         m.Task,
 		Model:        m.Model,
 		Provider:     m.Provider,
 		Reasoning:    m.Reasoning,
+		FastMode:     fastMode,
 		Subagent:     m.Subagent,
 		InboxPath:    m.InboxPath,
 		EventLogPath: m.EventLogPath,
 	if err != nil {
 		return nil, fmt.Errorf("swarm inbox path: %w", err)
 	}
+	fastMode := existing.FastMode
 	m := agentMeta{
 		ID: existing.ID, Task: existing.Task,
 		Dir: existing.Dir, Started: existing.Started,
 		Model: existing.Model, Provider: existing.Provider,
-		Reasoning: existing.Reasoning, Subagent: existing.Subagent,
+		Reasoning: existing.Reasoning, FastMode: &fastMode,
+		Subagent:  existing.Subagent,
 		SessionID: existing.SessionID,
 		InboxPath: inboxPath, EventLogPath: existing.EventLogPath,
 		SessionPath: existing.SessionPath,
 		Model:        m.Model,
 		Provider:     m.Provider,
 		Reasoning:    m.Reasoning,
+		FastMode:     fastMode,
 		Subagent:     m.Subagent,
 		SessionID:    m.SessionID,
 		InboxPath:    m.InboxPath,

M packages/agent/swarm/runner.go +11 -0


@@ -67,6 +67,8 @@ type swarmAgentArgsOpts struct {
67	67		Model string
68	68		Provider string
69	69		Reasoning string
	70	+	FastMode bool
	71	+	FastModeSet bool
70	72		Subagent string
71	73		}
72	74

@@ -98,6 +100,8 @@ func defaultChildArgs(exe string, a *Agent, sessionPath, inboxPath string) []str
98	100		Model: a.Model,
99	101		Provider: a.Provider,
100	102		Reasoning: a.Reasoning,
	103	+	FastMode: a.FastMode,
	104	+	FastModeSet: true,
101	105		Subagent: a.Subagent,
102	106		})
103	107		}

@@ -121,6 +125,13 @@ func swarmAgentArgs(opts swarmAgentArgsOpts) []string {
121	125		if opts.Reasoning != "" {
122	126		args = append(args, "--reasoning", opts.Reasoning)
123	127		}
	128	+	if opts.FastModeSet {
	129	+	if opts.FastMode {
	130	+	args = append(args, "--fast-mode")
	131	+	} else {
	132	+	args = append(args, "--no-fast-mode")
	133	+	}
	134	+	}
124	135		if opts.Subagent != "" {
125	136		args = append(args, "--subagent", opts.Subagent)
126	137		}


@@ -67,6 +67,8 @@ type swarmAgentArgsOpts struct {
67	Model string	67		Model string
68	Provider string	68		Provider string
69	Reasoning string	69		Reasoning string
		70	+	FastMode bool
		71	+	FastModeSet bool
70	Subagent string	72		Subagent string
71	}	73		}
72		74

@@ -98,6 +100,8 @@ func defaultChildArgs(exe string, a *Agent, sessionPath, inboxPath string) []str
98	Model: a.Model,	100		Model: a.Model,
99	Provider: a.Provider,	101		Provider: a.Provider,
100	Reasoning: a.Reasoning,	102		Reasoning: a.Reasoning,
		103	+	FastMode: a.FastMode,
		104	+	FastModeSet: true,
101	Subagent: a.Subagent,	105		Subagent: a.Subagent,
102	})	106		})
103	}	107		}

@@ -121,6 +125,13 @@ func swarmAgentArgs(opts swarmAgentArgsOpts) []string {
121	if opts.Reasoning != "" {	125		if opts.Reasoning != "" {
122	args = append(args, "--reasoning", opts.Reasoning)	126		args = append(args, "--reasoning", opts.Reasoning)
123	}	127		}
		128	+	if opts.FastModeSet {
		129	+	if opts.FastMode {
		130	+	args = append(args, "--fast-mode")
		131	+	} else {
		132	+	args = append(args, "--no-fast-mode")
		133	+	}
		134	+	}
124	if opts.Subagent != "" {	135		if opts.Subagent != "" {
125	args = append(args, "--subagent", opts.Subagent)	136		args = append(args, "--subagent", opts.Subagent)
126	}	137		}

M packages/agent/swarm/runner_test.go +3 -5


@@ -204,14 +204,12 @@ func TestDefaultChildArgsResumeOmitsTask(t *testing.T) {
204	204		t.Fatalf("resume argv contains the task; it would re-fire as a duplicate turn\n%v", args)
205	205		}
206	206		}
207		-	// And no trailing positional at all: the last arg should be a
208		-	// flag value, not a stray empty string.
	207	+	// And no trailing positional at all: every final token must be a
	208	+	// recognized boolean flag or a value belonging to a preceding flag,
	209	+	// never a stray empty string or the original task.
209	210		if got := args[len(args)-1]; got == "" {
210	211		t.Fatalf("resume argv ends with an empty positional: %v", args)
211	212		}
212		-	if strings.HasPrefix(args[len(args)-1], "--") {
213		-	t.Fatalf("resume argv ends on a bare flag (no value): %v", args)
214		-	}
215	213		}
216	214
217	215		func TestNotifyPromptTurnEndIgnoresProviderToolLoopTurnEnd(t *testing.T) {

 			t.Fatalf("resume argv contains the task; it would re-fire as a duplicate turn\n%v", args)
 		}
 	}
-	// And no trailing positional at all: the last arg should be a
+	// And no trailing positional at all: every final token must be a
-	// flag value, not a stray empty string.
+	// recognized boolean flag or a value belonging to a preceding flag,
+	// never a stray empty string or the original task.
 	if got := args[len(args)-1]; got == "" {
 		t.Fatalf("resume argv ends with an empty positional: %v", args)
 	}
-	if strings.HasPrefix(args[len(args)-1], "--") {
-		t.Fatalf("resume argv ends on a bare flag (no value): %v", args)
-	}
 }
 func TestNotifyPromptTurnEndIgnoresProviderToolLoopTurnEnd(t *testing.T) {

M packages/agent/swarm/swarm.go +24 -0


@@ -58,6 +58,10 @@ type Config struct {
58	58		// per-agent isolation: agents edit the host's files directly.
59	59		RepoRoot string
60	60
	61	+	// FastMode is copied to every newly spawned child and persisted so
	62	+	// child subprocesses use the same OpenAI service-tier setting.
	63	+	FastMode bool
	64	+
61	65		// NewRunner produces the Runner for an Agent. If nil, the default
62	66		// `zot --swarm-agent ...` exec runner is used. Tests inject a fake
63	67		// here.

@@ -156,6 +160,22 @@ func (f *Swarm) RepoRoot() string {
156	160		return f.cfg.RepoRoot
157	161		}
158	162
	163	+	// SetFastMode updates the fast-mode setting for subsequently spawned
	164	+	// children. Existing agents keep the setting captured at their spawn
	165	+	// boundary so resumes remain deterministic.
	166	+	func (f *Swarm) SetFastMode(enabled bool) {
	167	+	f.mu.Lock()
	168	+	f.cfg.FastMode = enabled
	169	+	f.mu.Unlock()
	170	+	}
	171	+
	172	+	// FastMode reports the setting used for subsequently spawned children.
	173	+	func (f *Swarm) FastMode() bool {
	174	+	f.mu.Lock()
	175	+	defer f.mu.Unlock()
	176	+	return f.cfg.FastMode
	177	+	}
	178	+
159	179		// SetActiveSession scopes the dashboard view (and Spawn stamping)
160	180		// to a particular host zot session id. Pass empty to clear the
161	181		// scope and revert to "show every agent" (the original behaviour).

@@ -235,6 +255,7 @@ func (f Swarm) SpawnReq(ctx context.Context, req SpawnRequest) (Agent, error)
235	255		// concurrent /cd cannot race a spawn or produce a mixed snapshot.
236	256		f.mu.Lock()
237	257		dir := f.cfg.RepoRoot
	258	+	fastMode := f.cfg.FastMode
238	259		sessionID := f.activeSession
239	260		f.mu.Unlock()
240	261

@@ -260,6 +281,7 @@ func (f Swarm) SpawnReq(ctx context.Context, req SpawnRequest) (Agent, error)
260	281		Model: strings.TrimSpace(req.Model),
261	282		Provider: strings.TrimSpace(req.Provider),
262	283		Reasoning: strings.TrimSpace(req.Reasoning),
	284	+	FastMode: fastMode,
263	285		Subagent: strings.TrimSpace(req.Subagent),
264	286		SessionID: sessionID,
265	287		InboxPath: inboxPath,

@@ -475,6 +497,7 @@ type AgentSnapshot struct {
475	497		Model string
476	498		Provider string
477	499		Reasoning string
	500	+	FastMode bool
478	501		Subagent string
479	502
480	503		// Paths to the agent's durable state. Surface them in the

@@ -506,6 +529,7 @@ func (a *Agent) Snapshot() AgentSnapshot {
506	529		Model: a.Model,
507	530		Provider: a.Provider,
508	531		Reasoning: a.Reasoning,
	532	+	FastMode: a.FastMode,
509	533		Subagent: a.Subagent,
510	534		InboxPath: a.InboxPath,
511	535		EventLogPath: a.EventLogPath,


@@ -58,6 +58,10 @@ type Config struct {
58	// per-agent isolation: agents edit the host's files directly.	58		// per-agent isolation: agents edit the host's files directly.
59	RepoRoot string	59		RepoRoot string
60		60
		61	+	// FastMode is copied to every newly spawned child and persisted so
		62	+	// child subprocesses use the same OpenAI service-tier setting.
		63	+	FastMode bool
		64	+
61	// NewRunner produces the Runner for an Agent. If nil, the default	65		// NewRunner produces the Runner for an Agent. If nil, the default
62	// `zot --swarm-agent ...` exec runner is used. Tests inject a fake	66		// `zot --swarm-agent ...` exec runner is used. Tests inject a fake
63	// here.	67		// here.

@@ -156,6 +160,22 @@ func (f *Swarm) RepoRoot() string {
156	return f.cfg.RepoRoot	160		return f.cfg.RepoRoot
157	}	161		}
158		162
		163	+	// SetFastMode updates the fast-mode setting for subsequently spawned
		164	+	// children. Existing agents keep the setting captured at their spawn
		165	+	// boundary so resumes remain deterministic.
		166	+	func (f *Swarm) SetFastMode(enabled bool) {
		167	+	f.mu.Lock()
		168	+	f.cfg.FastMode = enabled
		169	+	f.mu.Unlock()
		170	+	}
		171	+
		172	+	// FastMode reports the setting used for subsequently spawned children.
		173	+	func (f *Swarm) FastMode() bool {
		174	+	f.mu.Lock()
		175	+	defer f.mu.Unlock()
		176	+	return f.cfg.FastMode
		177	+	}
		178	+
159	// SetActiveSession scopes the dashboard view (and Spawn stamping)	179		// SetActiveSession scopes the dashboard view (and Spawn stamping)
160	// to a particular host zot session id. Pass empty to clear the	180		// to a particular host zot session id. Pass empty to clear the
161	// scope and revert to "show every agent" (the original behaviour).	181		// scope and revert to "show every agent" (the original behaviour).

@@ -235,6 +255,7 @@ func (f Swarm) SpawnReq(ctx context.Context, req SpawnRequest) (Agent, error)
235	// concurrent /cd cannot race a spawn or produce a mixed snapshot.	255		// concurrent /cd cannot race a spawn or produce a mixed snapshot.
236	f.mu.Lock()	256		f.mu.Lock()
237	dir := f.cfg.RepoRoot	257		dir := f.cfg.RepoRoot
		258	+	fastMode := f.cfg.FastMode
238	sessionID := f.activeSession	259		sessionID := f.activeSession
239	f.mu.Unlock()	260		f.mu.Unlock()
240		261

@@ -260,6 +281,7 @@ func (f Swarm) SpawnReq(ctx context.Context, req SpawnRequest) (Agent, error)
260	Model: strings.TrimSpace(req.Model),	281		Model: strings.TrimSpace(req.Model),
261	Provider: strings.TrimSpace(req.Provider),	282		Provider: strings.TrimSpace(req.Provider),
262	Reasoning: strings.TrimSpace(req.Reasoning),	283		Reasoning: strings.TrimSpace(req.Reasoning),
		284	+	FastMode: fastMode,
263	Subagent: strings.TrimSpace(req.Subagent),	285		Subagent: strings.TrimSpace(req.Subagent),
264	SessionID: sessionID,	286		SessionID: sessionID,
265	InboxPath: inboxPath,	287		InboxPath: inboxPath,

@@ -475,6 +497,7 @@ type AgentSnapshot struct {
475	Model string	497		Model string
476	Provider string	498		Provider string
477	Reasoning string	499		Reasoning string
		500	+	FastMode bool
478	Subagent string	501		Subagent string
479		502
480	// Paths to the agent's durable state. Surface them in the	503		// Paths to the agent's durable state. Surface them in the

@@ -506,6 +529,7 @@ func (a *Agent) Snapshot() AgentSnapshot {
506	Model: a.Model,	529		Model: a.Model,
507	Provider: a.Provider,	530		Provider: a.Provider,
508	Reasoning: a.Reasoning,	531		Reasoning: a.Reasoning,
		532	+	FastMode: a.FastMode,
509	Subagent: a.Subagent,	533		Subagent: a.Subagent,
510	InboxPath: a.InboxPath,	534		InboxPath: a.InboxPath,
511	EventLogPath: a.EventLogPath,	535		EventLogPath: a.EventLogPath,

M packages/agent/tools/swarm_spawn.go +4 -0


@@ -202,6 +202,9 @@ func (t *SwarmSpawnTool) Execute(ctx context.Context, raw json.RawMessage, progr
202	202		if reasoning != "" {
203	203		fmt.Fprintf(&sb, "reasoning: %s\n", reasoning)
204	204		}
	205	+	if agent.FastMode {
	206	+	sb.WriteString("fast mode: enabled\n")
	207	+	}
205	208		sb.WriteString("\nThe sub-agent is running in the background. Use /swarm in the TUI to monitor it. ")
206	209		sb.WriteString("This conversation continues immediately; do not wait for the sub-agent to finish before working on the next thing.")
207	210		return core.ToolResult{

@@ -213,6 +216,7 @@ func (t *SwarmSpawnTool) Execute(ctx context.Context, raw json.RawMessage, progr
213	216		"model": model,
214	217		"provider": providerID,
215	218		"reasoning": reasoning,
	219	+	"fast_mode": agent.FastMode,
216	220		},
217	221		}, nil
218	222		}


@@ -202,6 +202,9 @@ func (t *SwarmSpawnTool) Execute(ctx context.Context, raw json.RawMessage, progr
202	if reasoning != "" {	202		if reasoning != "" {
203	fmt.Fprintf(&sb, "reasoning: %s\n", reasoning)	203		fmt.Fprintf(&sb, "reasoning: %s\n", reasoning)
204	}	204		}
		205	+	if agent.FastMode {
		206	+	sb.WriteString("fast mode: enabled\n")
		207	+	}
205	sb.WriteString("\nThe sub-agent is running in the background. Use /swarm in the TUI to monitor it. ")	208		sb.WriteString("\nThe sub-agent is running in the background. Use /swarm in the TUI to monitor it. ")
206	sb.WriteString("This conversation continues immediately; do not wait for the sub-agent to finish before working on the next thing.")	209		sb.WriteString("This conversation continues immediately; do not wait for the sub-agent to finish before working on the next thing.")
207	return core.ToolResult{	210		return core.ToolResult{

@@ -213,6 +216,7 @@ func (t *SwarmSpawnTool) Execute(ctx context.Context, raw json.RawMessage, progr
213	"model": model,	216		"model": model,
214	"provider": providerID,	217		"provider": providerID,
215	"reasoning": reasoning,	218		"reasoning": reasoning,
		219	+	"fast_mode": agent.FastMode,
216	},	220		},
217	}, nil	221		}, nil
218	}	222		}

M packages/core/agent.go +24 -0


@@ -22,6 +22,9 @@ type Agent struct {
22	22		MaxSteps int
23	23		Reasoning string
24	24		Temperature *float32
	25	+	// FastMode requests the OpenAI fast service tier. The provider boundary
	26	+	// rejects it for providers that do not support that contract.
	27	+	FastMode bool
25	28
26	29		// MaxTokens caps the model's output tokens per turn. Zero leaves
27	30		// the field unset on the provider request, letting each provider

@@ -120,6 +123,22 @@ func NewAgent(client provider.Client, model, system string, tools Registry) *Age
120	123		}
121	124		}
122	125
	126	+	// SetFastMode changes the service-tier preference for the next model call.
	127	+	// It is safe to use while a turn is in flight; the active turn keeps the
	128	+	// value it already captured.
	129	+	func (a *Agent) SetFastMode(enabled bool) {
	130	+	a.mu.Lock()
	131	+	a.FastMode = enabled
	132	+	a.mu.Unlock()
	133	+	}
	134	+
	135	+	// FastModeEnabled returns the service-tier preference for a model call.
	136	+	func (a *Agent) FastModeEnabled() bool {
	137	+	a.mu.Lock()
	138	+	defer a.mu.Unlock()
	139	+	return a.FastMode
	140	+	}
	141	+
123	142		// QueueMessage queues text to be injected as a user message at the
124	143		// next safe boundary of the active agent loop. It is non-blocking in
125	144		// the sense that it never waits for model/tool work; it only takes

@@ -540,6 +559,10 @@ func (a *Agent) dropLastAssistantMessage() {
540	559		// oneTurn calls the LLM once, forwards events, returns the stop reason
541	560		// and the assembled assistant message (already appended to the transcript).
542	561		func (a *Agent) oneTurn(ctx context.Context, sink func(AgentEvent)) (provider.StopReason, provider.Message, error) {
	562	+	fastMode := a.FastModeEnabled()
	563	+	if err := provider.ValidateFastMode(a.Client.Name(), fastMode); err != nil {
	564	+	return provider.StopError, provider.Message{}, err
	565	+	}
543	566		req := provider.Request{
544	567		Model: a.Model,
545	568		System: a.System,

@@ -554,6 +577,7 @@ func (a *Agent) oneTurn(ctx context.Context, sink func(AgentEvent)) (provider.St
554	577		Messages: repairToolUseResultPairs(a.Messages()),
555	578		Tools: a.Tools.Specs(),
556	579		Reasoning: a.Reasoning,
	580	+	FastMode: fastMode,
557	581		MaxTokens: a.MaxTokens,
558	582		Temperature: a.Temperature,
559	583		}


@@ -22,6 +22,9 @@ type Agent struct {
22	MaxSteps int	22		MaxSteps int
23	Reasoning string	23		Reasoning string
24	Temperature *float32	24		Temperature *float32
		25	+	// FastMode requests the OpenAI fast service tier. The provider boundary
		26	+	// rejects it for providers that do not support that contract.
		27	+	FastMode bool
25		28
26	// MaxTokens caps the model's output tokens per turn. Zero leaves	29		// MaxTokens caps the model's output tokens per turn. Zero leaves
27	// the field unset on the provider request, letting each provider	30		// the field unset on the provider request, letting each provider

@@ -120,6 +123,22 @@ func NewAgent(client provider.Client, model, system string, tools Registry) *Age
120	}	123		}
121	}	124		}
122		125
		126	+	// SetFastMode changes the service-tier preference for the next model call.
		127	+	// It is safe to use while a turn is in flight; the active turn keeps the
		128	+	// value it already captured.
		129	+	func (a *Agent) SetFastMode(enabled bool) {
		130	+	a.mu.Lock()
		131	+	a.FastMode = enabled
		132	+	a.mu.Unlock()
		133	+	}
		134	+
		135	+	// FastModeEnabled returns the service-tier preference for a model call.
		136	+	func (a *Agent) FastModeEnabled() bool {
		137	+	a.mu.Lock()
		138	+	defer a.mu.Unlock()
		139	+	return a.FastMode
		140	+	}
		141	+
123	// QueueMessage queues text to be injected as a user message at the	142		// QueueMessage queues text to be injected as a user message at the
124	// next safe boundary of the active agent loop. It is non-blocking in	143		// next safe boundary of the active agent loop. It is non-blocking in
125	// the sense that it never waits for model/tool work; it only takes	144		// the sense that it never waits for model/tool work; it only takes

@@ -540,6 +559,10 @@ func (a *Agent) dropLastAssistantMessage() {
540	// oneTurn calls the LLM once, forwards events, returns the stop reason	559		// oneTurn calls the LLM once, forwards events, returns the stop reason
541	// and the assembled assistant message (already appended to the transcript).	560		// and the assembled assistant message (already appended to the transcript).
542	func (a *Agent) oneTurn(ctx context.Context, sink func(AgentEvent)) (provider.StopReason, provider.Message, error) {	561		func (a *Agent) oneTurn(ctx context.Context, sink func(AgentEvent)) (provider.StopReason, provider.Message, error) {
		562	+	fastMode := a.FastModeEnabled()
		563	+	if err := provider.ValidateFastMode(a.Client.Name(), fastMode); err != nil {
		564	+	return provider.StopError, provider.Message{}, err
		565	+	}
543	req := provider.Request{	566		req := provider.Request{
544	Model: a.Model,	567		Model: a.Model,
545	System: a.System,	568		System: a.System,

@@ -554,6 +577,7 @@ func (a *Agent) oneTurn(ctx context.Context, sink func(AgentEvent)) (provider.St
554	Messages: repairToolUseResultPairs(a.Messages()),	577		Messages: repairToolUseResultPairs(a.Messages()),
555	Tools: a.Tools.Specs(),	578		Tools: a.Tools.Specs(),
556	Reasoning: a.Reasoning,	579		Reasoning: a.Reasoning,
		580	+	FastMode: fastMode,
557	MaxTokens: a.MaxTokens,	581		MaxTokens: a.MaxTokens,
558	Temperature: a.Temperature,	582		Temperature: a.Temperature,
559	}	583		}

M packages/core/compact.go +6 -0


@@ -40,6 +40,11 @@ func (a *Agent) Compact(ctx context.Context, keepTail int, sink func(delta strin
40	40		return "", fmt.Errorf("nothing to compact: keep-tail covers the whole transcript")
41	41		}
42	42
	43	+	fastMode := a.FastModeEnabled()
	44	+	if err := provider.ValidateFastMode(a.Client.Name(), fastMode); err != nil {
	45	+	return "", err
	46	+	}
	47	+
43	48		// Serialize the summarizable transcript to text and wrap it in tags
44	49		// so the model treats it as material to summarize, not to continue.
45	50		transcript := serializeTranscript(summarizable)

@@ -51,6 +56,7 @@ func (a *Agent) Compact(ctx context.Context, keepTail int, sink func(delta strin
51	56		System: summarizationSystem,
52	57		MaxTokens: 4096,
53	58		Temperature: a.Temperature,
	59	+	FastMode: fastMode,
54	60		Messages: []provider.Message{
55	61		{
56	62		Role: provider.RoleUser,


@@ -40,6 +40,11 @@ func (a *Agent) Compact(ctx context.Context, keepTail int, sink func(delta strin
40	return "", fmt.Errorf("nothing to compact: keep-tail covers the whole transcript")	40		return "", fmt.Errorf("nothing to compact: keep-tail covers the whole transcript")
41	}	41		}
42		42
		43	+	fastMode := a.FastModeEnabled()
		44	+	if err := provider.ValidateFastMode(a.Client.Name(), fastMode); err != nil {
		45	+	return "", err
		46	+	}
		47	+
43	// Serialize the summarizable transcript to text and wrap it in tags	48		// Serialize the summarizable transcript to text and wrap it in tags
44	// so the model treats it as material to summarize, not to continue.	49		// so the model treats it as material to summarize, not to continue.
45	transcript := serializeTranscript(summarizable)	50		transcript := serializeTranscript(summarizable)

@@ -51,6 +56,7 @@ func (a *Agent) Compact(ctx context.Context, keepTail int, sink func(delta strin
51	System: summarizationSystem,	56		System: summarizationSystem,
52	MaxTokens: 4096,	57		MaxTokens: 4096,
53	Temperature: a.Temperature,	58		Temperature: a.Temperature,
		59	+	FastMode: fastMode,
54	Messages: []provider.Message{	60		Messages: []provider.Message{
55	{	61		{
56	Role: provider.RoleUser,	62		Role: provider.RoleUser,

A packages/core/fast_mode_test.go +92 -0

@@ -0,0 +1,92 @@
	1	+	package core
	2	+
	3	+	import (
	4	+	"context"
	5	+	"strings"
	6	+	"testing"
	7	+
	8	+	"github.com/patriceckhart/zot/packages/provider"
	9	+	)
	10	+
	11	+	type fastModeTestClient struct {
	12	+	name string
	13	+	lastReq provider.Request
	14	+	called bool
	15	+	}
	16	+
	17	+	func (c *fastModeTestClient) Name() string { return c.name }
	18	+
	19	+	func (c *fastModeTestClient) Stream(_ context.Context, req provider.Request) (<-chan provider.Event, error) {
	20	+	c.called = true
	21	+	c.lastReq = req
	22	+	out := make(chan provider.Event, 2)
	23	+	out <- provider.EventTextDelta{Delta: "ok"}
	24	+	out <- provider.EventDone{Stop: provider.StopEnd}
	25	+	close(out)
	26	+	return out, nil
	27	+	}
	28	+
	29	+	func TestAgentPropagatesFastMode(t *testing.T) {
	30	+	client := &fastModeTestClient{name: "openai"}
	31	+	agent := NewAgent(client, "gpt-5", "", nil)
	32	+	agent.FastMode = true
	33	+
	34	+	if err := agent.Prompt(context.Background(), "hello", nil, nil); err != nil {
	35	+	t.Fatalf("Prompt returned %v", err)
	36	+	}
	37	+	if !client.called {
	38	+	t.Fatal("provider was not called")
	39	+	}
	40	+	if !client.lastReq.FastMode {
	41	+	t.Fatal("request FastMode = false, want true")
	42	+	}
	43	+	}
	44	+
	45	+	func TestAgentPropagatesFastModeToCompaction(t *testing.T) {
	46	+	client := &fastModeTestClient{name: "openai"}
	47	+	agent := NewAgent(client, "gpt-5", "", nil)
	48	+	agent.FastMode = true
	49	+	agent.SetMessages([]provider.Message{
	50	+	{Role: provider.RoleUser, Content: []provider.Content{provider.TextBlock{Text: "old"}}},
	51	+	{Role: provider.RoleAssistant, Content: []provider.Content{provider.TextBlock{Text: "answer"}}},
	52	+	})
	53	+
	54	+	if _, err := agent.Compact(context.Background(), 1, nil); err != nil {
	55	+	t.Fatalf("Compact returned %v", err)
	56	+	}
	57	+	if !client.lastReq.FastMode {
	58	+	t.Fatal("compaction request FastMode = false, want true")
	59	+	}
	60	+	}
	61	+
	62	+	func TestAgentRejectsFastModeForNonOpenAIProvider(t *testing.T) {
	63	+	client := &fastModeTestClient{name: "anthropic"}
	64	+	agent := NewAgent(client, "claude-sonnet-4-5", "", nil)
	65	+	agent.FastMode = true
	66	+
	67	+	err := agent.Prompt(context.Background(), "hello", nil, nil)
	68	+	if err == nil \|\| !strings.Contains(err.Error(), "only supported for OpenAI providers") {
	69	+	t.Fatalf("Prompt error = %v, want unsupported-provider error", err)
	70	+	}
	71	+	if client.called {
	72	+	t.Fatal("provider was called despite unsupported fast mode")
	73	+	}
	74	+	}
	75	+
	76	+	func TestAgentRejectsFastModeForNonOpenAIProviderDuringCompaction(t *testing.T) {
	77	+	client := &fastModeTestClient{name: "anthropic"}
	78	+	agent := NewAgent(client, "claude-sonnet-4-5", "", nil)
	79	+	agent.FastMode = true
	80	+	agent.SetMessages([]provider.Message{
	81	+	{Role: provider.RoleUser, Content: []provider.Content{provider.TextBlock{Text: "old"}}},
	82	+	{Role: provider.RoleAssistant, Content: []provider.Content{provider.TextBlock{Text: "answer"}}},
	83	+	})
	84	+
	85	+	_, err := agent.Compact(context.Background(), 1, nil)
	86	+	if err == nil \|\| !strings.Contains(err.Error(), "only supported for OpenAI providers") {
	87	+	t.Fatalf("Compact error = %v, want unsupported-provider error", err)
	88	+	}
	89	+	if client.called {
	90	+	t.Fatal("provider was called despite unsupported fast mode")
	91	+	}
	92	+	}

@@ -0,0 +1,92 @@
	1	+	package core
	2	+
	3	+	import (
	4	+	"context"
	5	+	"strings"
	6	+	"testing"
	7	+
	8	+	"github.com/patriceckhart/zot/packages/provider"
	9	+	)
	10	+
	11	+	type fastModeTestClient struct {
	12	+	name string
	13	+	lastReq provider.Request
	14	+	called bool
	15	+	}
	16	+
	17	+	func (c *fastModeTestClient) Name() string { return c.name }
	18	+
	19	+	func (c *fastModeTestClient) Stream(_ context.Context, req provider.Request) (<-chan provider.Event, error) {
	20	+	c.called = true
	21	+	c.lastReq = req
	22	+	out := make(chan provider.Event, 2)
	23	+	out <- provider.EventTextDelta{Delta: "ok"}
	24	+	out <- provider.EventDone{Stop: provider.StopEnd}
	25	+	close(out)
	26	+	return out, nil
	27	+	}
	28	+
	29	+	func TestAgentPropagatesFastMode(t *testing.T) {
	30	+	client := &fastModeTestClient{name: "openai"}
	31	+	agent := NewAgent(client, "gpt-5", "", nil)
	32	+	agent.FastMode = true
	33	+
	34	+	if err := agent.Prompt(context.Background(), "hello", nil, nil); err != nil {
	35	+	t.Fatalf("Prompt returned %v", err)
	36	+	}
	37	+	if !client.called {
	38	+	t.Fatal("provider was not called")
	39	+	}
	40	+	if !client.lastReq.FastMode {
	41	+	t.Fatal("request FastMode = false, want true")
	42	+	}
	43	+	}
	44	+
	45	+	func TestAgentPropagatesFastModeToCompaction(t *testing.T) {
	46	+	client := &fastModeTestClient{name: "openai"}
	47	+	agent := NewAgent(client, "gpt-5", "", nil)
	48	+	agent.FastMode = true
	49	+	agent.SetMessages([]provider.Message{
	50	+	{Role: provider.RoleUser, Content: []provider.Content{provider.TextBlock{Text: "old"}}},
	51	+	{Role: provider.RoleAssistant, Content: []provider.Content{provider.TextBlock{Text: "answer"}}},
	52	+	})
	53	+
	54	+	if _, err := agent.Compact(context.Background(), 1, nil); err != nil {
	55	+	t.Fatalf("Compact returned %v", err)
	56	+	}
	57	+	if !client.lastReq.FastMode {
	58	+	t.Fatal("compaction request FastMode = false, want true")
	59	+	}
	60	+	}
	61	+
	62	+	func TestAgentRejectsFastModeForNonOpenAIProvider(t *testing.T) {
	63	+	client := &fastModeTestClient{name: "anthropic"}
	64	+	agent := NewAgent(client, "claude-sonnet-4-5", "", nil)
	65	+	agent.FastMode = true
	66	+
	67	+	err := agent.Prompt(context.Background(), "hello", nil, nil)
	68	+	if err == nil \|\| !strings.Contains(err.Error(), "only supported for OpenAI providers") {
	69	+	t.Fatalf("Prompt error = %v, want unsupported-provider error", err)
	70	+	}
	71	+	if client.called {
	72	+	t.Fatal("provider was called despite unsupported fast mode")
	73	+	}
	74	+	}
	75	+
	76	+	func TestAgentRejectsFastModeForNonOpenAIProviderDuringCompaction(t *testing.T) {
	77	+	client := &fastModeTestClient{name: "anthropic"}
	78	+	agent := NewAgent(client, "claude-sonnet-4-5", "", nil)
	79	+	agent.FastMode = true
	80	+	agent.SetMessages([]provider.Message{
	81	+	{Role: provider.RoleUser, Content: []provider.Content{provider.TextBlock{Text: "old"}}},
	82	+	{Role: provider.RoleAssistant, Content: []provider.Content{provider.TextBlock{Text: "answer"}}},
	83	+	})
	84	+
	85	+	_, err := agent.Compact(context.Background(), 1, nil)
	86	+	if err == nil \|\| !strings.Contains(err.Error(), "only supported for OpenAI providers") {
	87	+	t.Fatalf("Compact error = %v, want unsupported-provider error", err)
	88	+	}
	89	+	if client.called {
	90	+	t.Fatal("provider was called despite unsupported fast mode")
	91	+	}
	92	+	}

M packages/provider/amazon_bedrock.go +3 -0


@@ -572,6 +572,9 @@ func bedrockGeoPrefixForRegion(region string) string {
572	572		}
573	573
574	574		func (c *bedrockClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {
	575	+	if err := ValidateFastMode(c.Name(), req.FastMode); err != nil {
	576	+	return nil, err
	577	+	}
575	578		wire, err := c.buildRequest(req)
576	579		if err != nil {
577	580		return nil, err


@@ -572,6 +572,9 @@ func bedrockGeoPrefixForRegion(region string) string {
572	}	572		}
573		573
574	func (c *bedrockClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {	574		func (c *bedrockClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {
		575	+	if err := ValidateFastMode(c.Name(), req.FastMode); err != nil {
		576	+	return nil, err
		577	+	}
575	wire, err := c.buildRequest(req)	578		wire, err := c.buildRequest(req)
576	if err != nil {	579		if err != nil {
577	return nil, err	580		return nil, err

M packages/provider/anthropic.go +3 -0


@@ -523,6 +523,9 @@ func anthBuildToolResultContent(blocks []Content) (json.RawMessage, error) {
523	523		// ---- streaming ----
524	524
525	525		func (c *anthropicClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {
	526	+	if err := ValidateFastMode(c.Name(), req.FastMode); err != nil {
	527	+	return nil, err
	528	+	}
526	529		wire, err := c.buildRequest(req)
527	530		if err != nil {
528	531		return nil, err


@@ -523,6 +523,9 @@ func anthBuildToolResultContent(blocks []Content) (json.RawMessage, error) {
523	// ---- streaming ----	523		// ---- streaming ----
524		524
525	func (c *anthropicClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {	525		func (c *anthropicClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {
		526	+	if err := ValidateFastMode(c.Name(), req.FastMode); err != nil {
		527	+	return nil, err
		528	+	}
526	wire, err := c.buildRequest(req)	529		wire, err := c.buildRequest(req)
527	if err != nil {	530		if err != nil {
528	return nil, err	531		return nil, err

M packages/provider/extra_providers.go +3 -0


@@ -233,6 +233,9 @@ type unimplementedClient struct {
233	233		func (c *unimplementedClient) Name() string { return c.name }
234	234
235	235		func (c *unimplementedClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {
	236	+	if err := ValidateFastMode(c.Name(), req.FastMode); err != nil {
	237	+	return nil, err
	238	+	}
236	239		return nil, fmt.Errorf("provider %q not yet implemented: %s", c.name, c.hint)
237	240		}
238	241


@@ -233,6 +233,9 @@ type unimplementedClient struct {
233	func (c *unimplementedClient) Name() string { return c.name }	233		func (c *unimplementedClient) Name() string { return c.name }
234		234
235	func (c *unimplementedClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {	235		func (c *unimplementedClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {
		236	+	if err := ValidateFastMode(c.Name(), req.FastMode); err != nil {
		237	+	return nil, err
		238	+	}
236	return nil, fmt.Errorf("provider %q not yet implemented: %s", c.name, c.hint)	239		return nil, fmt.Errorf("provider %q not yet implemented: %s", c.name, c.hint)
237	}	240		}
238		241

A packages/provider/fast_mode_test.go +62 -0

@@ -0,0 +1,62 @@
	1	+	package provider
	2	+
	3	+	import (
	4	+	"context"
	5	+	"strings"
	6	+	"testing"
	7	+	)
	8	+
	9	+	func TestOpenAIFastModeAddsServiceTier(t *testing.T) {
	10	+	client := NewOpenAI("key", "").(*openaiClient)
	11	+
	12	+	wire, err := client.buildRequest(Request{Model: "gpt-5", FastMode: true})
	13	+	if err != nil {
	14	+	t.Fatal(err)
	15	+	}
	16	+	if wire.ServiceTier != fastModeServiceTier {
	17	+	t.Fatalf("service_tier = %q, want %q", wire.ServiceTier, fastModeServiceTier)
	18	+	}
	19	+
	20	+	wire, err = client.buildRequest(Request{Model: "gpt-5"})
	21	+	if err != nil {
	22	+	t.Fatal(err)
	23	+	}
	24	+	if wire.ServiceTier != "" {
	25	+	t.Fatalf("service_tier = %q with fast mode disabled, want omitted", wire.ServiceTier)
	26	+	}
	27	+	}
	28	+
	29	+	func TestOpenAICodexFastModeAddsServiceTier(t *testing.T) {
	30	+	client := NewOpenAICodex("token", "account", "").(*codexClient)
	31	+
	32	+	wire, err := client.buildRequest(Request{Model: "gpt-5.6-sol", FastMode: true})
	33	+	if err != nil {
	34	+	t.Fatal(err)
	35	+	}
	36	+	if wire.ServiceTier != fastModeServiceTier {
	37	+	t.Fatalf("service_tier = %q, want %q", wire.ServiceTier, fastModeServiceTier)
	38	+	}
	39	+	}
	40	+
	41	+	func TestFastModeIsRestrictedToOpenAIProviders(t *testing.T) {
	42	+	for _, providerName := range []string{"openai", "openai-codex", "openai-responses"} {
	43	+	if err := ValidateFastMode(providerName, true); err != nil {
	44	+	t.Errorf("ValidateFastMode(%q, true) = %v", providerName, err)
	45	+	}
	46	+	}
	47	+
	48	+	for _, providerName := range []string{"anthropic", "google", "openrouter", "azure-openai-responses"} {
	49	+	err := ValidateFastMode(providerName, true)
	50	+	if err == nil \|\| !strings.Contains(err.Error(), "only supported for OpenAI providers") {
	51	+	t.Errorf("ValidateFastMode(%q, true) = %v, want unsupported-provider error", providerName, err)
	52	+	}
	53	+	}
	54	+	}
	55	+
	56	+	func TestFastModeRejectedBeforeNonOpenAIRequest(t *testing.T) {
	57	+	client := NewOpenAICompat("openrouter", "key", "", "")
	58	+	_, err := client.Stream(context.Background(), Request{Model: "gpt-5", FastMode: true})
	59	+	if err == nil \|\| !strings.Contains(err.Error(), "only supported for OpenAI providers") {
	60	+	t.Fatalf("Stream error = %v, want unsupported-provider error", err)
	61	+	}
	62	+	}

@@ -0,0 +1,62 @@
	1	+	package provider
	2	+
	3	+	import (
	4	+	"context"
	5	+	"strings"
	6	+	"testing"
	7	+	)
	8	+
	9	+	func TestOpenAIFastModeAddsServiceTier(t *testing.T) {
	10	+	client := NewOpenAI("key", "").(*openaiClient)
	11	+
	12	+	wire, err := client.buildRequest(Request{Model: "gpt-5", FastMode: true})
	13	+	if err != nil {
	14	+	t.Fatal(err)
	15	+	}
	16	+	if wire.ServiceTier != fastModeServiceTier {
	17	+	t.Fatalf("service_tier = %q, want %q", wire.ServiceTier, fastModeServiceTier)
	18	+	}
	19	+
	20	+	wire, err = client.buildRequest(Request{Model: "gpt-5"})
	21	+	if err != nil {
	22	+	t.Fatal(err)
	23	+	}
	24	+	if wire.ServiceTier != "" {
	25	+	t.Fatalf("service_tier = %q with fast mode disabled, want omitted", wire.ServiceTier)
	26	+	}
	27	+	}
	28	+
	29	+	func TestOpenAICodexFastModeAddsServiceTier(t *testing.T) {
	30	+	client := NewOpenAICodex("token", "account", "").(*codexClient)
	31	+
	32	+	wire, err := client.buildRequest(Request{Model: "gpt-5.6-sol", FastMode: true})
	33	+	if err != nil {
	34	+	t.Fatal(err)
	35	+	}
	36	+	if wire.ServiceTier != fastModeServiceTier {
	37	+	t.Fatalf("service_tier = %q, want %q", wire.ServiceTier, fastModeServiceTier)
	38	+	}
	39	+	}
	40	+
	41	+	func TestFastModeIsRestrictedToOpenAIProviders(t *testing.T) {
	42	+	for _, providerName := range []string{"openai", "openai-codex", "openai-responses"} {
	43	+	if err := ValidateFastMode(providerName, true); err != nil {
	44	+	t.Errorf("ValidateFastMode(%q, true) = %v", providerName, err)
	45	+	}
	46	+	}
	47	+
	48	+	for _, providerName := range []string{"anthropic", "google", "openrouter", "azure-openai-responses"} {
	49	+	err := ValidateFastMode(providerName, true)
	50	+	if err == nil \|\| !strings.Contains(err.Error(), "only supported for OpenAI providers") {
	51	+	t.Errorf("ValidateFastMode(%q, true) = %v, want unsupported-provider error", providerName, err)
	52	+	}
	53	+	}
	54	+	}
	55	+
	56	+	func TestFastModeRejectedBeforeNonOpenAIRequest(t *testing.T) {
	57	+	client := NewOpenAICompat("openrouter", "key", "", "")
	58	+	_, err := client.Stream(context.Background(), Request{Model: "gpt-5", FastMode: true})
	59	+	if err == nil \|\| !strings.Contains(err.Error(), "only supported for OpenAI providers") {
	60	+	t.Fatalf("Stream error = %v, want unsupported-provider error", err)
	61	+	}
	62	+	}

M packages/provider/gemini.go +3 -0


@@ -468,6 +468,9 @@ func geminiThinkingConfig(modelID, level string) *gemThinkingConfig {
468	468		// ---- streaming ----
469	469
470	470		func (c *geminiClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {
	471	+	if err := ValidateFastMode(c.Name(), req.FastMode); err != nil {
	472	+	return nil, err
	473	+	}
471	474		wire, modelID, err := c.buildRequest(req)
472	475		if err != nil {
473	476		return nil, err


@@ -468,6 +468,9 @@ func geminiThinkingConfig(modelID, level string) *gemThinkingConfig {
468	// ---- streaming ----	468		// ---- streaming ----
469		469
470	func (c *geminiClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {	470		func (c *geminiClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {
		471	+	if err := ValidateFastMode(c.Name(), req.FastMode); err != nil {
		472	+	return nil, err
		473	+	}
471	wire, modelID, err := c.buildRequest(req)	474		wire, modelID, err := c.buildRequest(req)
472	if err != nil {	475		if err != nil {
473	return nil, err	476		return nil, err

M packages/provider/google_vertex.go +3 -0


@@ -367,5 +367,8 @@ type renamedClient struct {
367	367
368	368		func (r *renamedClient) Name() string { return r.name }
369	369		func (r *renamedClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {
	370	+	if err := ValidateFastMode(r.Name(), req.FastMode); err != nil {
	371	+	return nil, err
	372	+	}
370	373		return r.inner.Stream(ctx, req)
371	374		}


@@ -367,5 +367,8 @@ type renamedClient struct {
367		367
368	func (r *renamedClient) Name() string { return r.name }	368		func (r *renamedClient) Name() string { return r.name }
369	func (r *renamedClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {	369		func (r *renamedClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {
		370	+	if err := ValidateFastMode(r.Name(), req.FastMode); err != nil {
		371	+	return nil, err
		372	+	}
370	return r.inner.Stream(ctx, req)	373		return r.inner.Stream(ctx, req)
371	}	374		}

M packages/provider/model_router.go +3 -0


@@ -30,6 +30,9 @@ func NewModelRouter(name string, fallback Client, byAPI map[string]Client) Clien
30	30		func (c *modelRouter) Name() string { return c.name }
31	31
32	32		func (c *modelRouter) Stream(ctx context.Context, req Request) (<-chan Event, error) {
	33	+	if err := ValidateFastMode(c.Name(), req.FastMode); err != nil {
	34	+	return nil, err
	35	+	}
33	36		client := c.fallback
34	37		if model, err := FindModel(c.name, req.Model); err == nil && model.API != "" {
35	38		routed := c.byAPI[model.API]


@@ -30,6 +30,9 @@ func NewModelRouter(name string, fallback Client, byAPI map[string]Client) Clien
30	func (c *modelRouter) Name() string { return c.name }	30		func (c *modelRouter) Name() string { return c.name }
31		31
32	func (c *modelRouter) Stream(ctx context.Context, req Request) (<-chan Event, error) {	32		func (c *modelRouter) Stream(ctx context.Context, req Request) (<-chan Event, error) {
		33	+	if err := ValidateFastMode(c.Name(), req.FastMode); err != nil {
		34	+	return nil, err
		35	+	}
33	client := c.fallback	36		client := c.fallback
34	if model, err := FindModel(c.name, req.Model); err == nil && model.API != "" {	37		if model, err := FindModel(c.name, req.Model); err == nil && model.API != "" {
35	routed := c.byAPI[model.API]	38		routed := c.byAPI[model.API]

M packages/provider/openai.go +7 -0


@@ -176,11 +176,15 @@ type oaiRequest struct {
176	176		MaxTokens *int `json:"max_tokens,omitempty"`
177	177		MaxCompletionTok *int `json:"max_completion_tokens,omitempty"`
178	178		ReasoningEffort string `json:"reasoning_effort,omitempty"`
	179	+	ServiceTier string `json:"service_tier,omitempty"`
179	180		}
180	181
181	182		// ---- request building ----
182	183
183	184		func (c openaiClient) buildRequest(req Request) (oaiRequest, error) {
	185	+	if err := ValidateFastMode(c.Name(), req.FastMode); err != nil {
	186	+	return nil, err
	187	+	}
184	188		// The OpenAI wire client is shared by many providers, so prefer its own
185	189		// provider namespace before falling back to a provider-agnostic lookup.
186	190		m, err := FindModel(c.Name(), req.Model)

@@ -203,6 +207,9 @@ func (c openaiClient) buildRequest(req Request) (oaiRequest, error) {
203	207		StreamOptions: &oaiStreamOptions{IncludeUsage: true},
204	208		Temperature: req.Temperature,
205	209		}
	210	+	if req.FastMode {
	211	+	out.ServiceTier = fastModeServiceTier
	212	+	}
206	213
207	214		maxTok := req.MaxTokens
208	215		if maxTok <= 0 {


@@ -176,11 +176,15 @@ type oaiRequest struct {
176	MaxTokens *int `json:"max_tokens,omitempty"`	176		MaxTokens *int `json:"max_tokens,omitempty"`
177	MaxCompletionTok *int `json:"max_completion_tokens,omitempty"`	177		MaxCompletionTok *int `json:"max_completion_tokens,omitempty"`
178	ReasoningEffort string `json:"reasoning_effort,omitempty"`	178		ReasoningEffort string `json:"reasoning_effort,omitempty"`
		179	+	ServiceTier string `json:"service_tier,omitempty"`
179	}	180		}
180		181
181	// ---- request building ----	182		// ---- request building ----
182		183
183	func (c openaiClient) buildRequest(req Request) (oaiRequest, error) {	184		func (c openaiClient) buildRequest(req Request) (oaiRequest, error) {
		185	+	if err := ValidateFastMode(c.Name(), req.FastMode); err != nil {
		186	+	return nil, err
		187	+	}
184	// The OpenAI wire client is shared by many providers, so prefer its own	188		// The OpenAI wire client is shared by many providers, so prefer its own
185	// provider namespace before falling back to a provider-agnostic lookup.	189		// provider namespace before falling back to a provider-agnostic lookup.
186	m, err := FindModel(c.Name(), req.Model)	190		m, err := FindModel(c.Name(), req.Model)

@@ -203,6 +207,9 @@ func (c openaiClient) buildRequest(req Request) (oaiRequest, error) {
203	StreamOptions: &oaiStreamOptions{IncludeUsage: true},	207		StreamOptions: &oaiStreamOptions{IncludeUsage: true},
204	Temperature: req.Temperature,	208		Temperature: req.Temperature,
205	}	209		}
		210	+	if req.FastMode {
		211	+	out.ServiceTier = fastModeServiceTier
		212	+	}
206		213
207	maxTok := req.MaxTokens	214		maxTok := req.MaxTokens
208	if maxTok <= 0 {	215		if maxTok <= 0 {

M packages/provider/openai_codex.go +11 -0


@@ -163,6 +163,7 @@ type codexRequest struct {
163	163		Include []string `json:"include,omitempty"`
164	164		Reasoning *codexReasoningConfig `json:"reasoning,omitempty"`
165	165		PromptCacheKey string `json:"prompt_cache_key,omitempty"`
	166	+	ServiceTier string `json:"service_tier,omitempty"`
166	167		}
167	168
168	169		// ---- Request building ----

@@ -180,6 +181,13 @@ func (c *codexClient) findModel(id string) (Model, error) {
180	181		}
181	182
182	183		func (c codexClient) buildRequest(req Request) (codexRequest, error) {
	184	+	providerName := c.providerName
	185	+	if providerName == "" {
	186	+	providerName = "openai-codex"
	187	+	}
	188	+	if err := ValidateFastMode(providerName, req.FastMode); err != nil {
	189	+	return nil, err
	190	+	}
183	191		m, err := c.findModel(req.Model)
184	192		if err != nil {
185	193		return nil, err

@@ -194,6 +202,9 @@ func (c codexClient) buildRequest(req Request) (codexRequest, error) {
194	202		ParallelToolCalls: true,
195	203		Include: []string{"reasoning.encrypted_content"},
196	204		}
	205	+	if req.FastMode {
	206	+	body.ServiceTier = fastModeServiceTier
	207	+	}
197	208		if m.Reasoning {
198	209		if effort := OpenAICodexReasoningEffort(reasoning, req.Model); effort != "" {
199	210		body.Reasoning = &codexReasoningConfig{Effort: effort}


@@ -163,6 +163,7 @@ type codexRequest struct {
163	Include []string `json:"include,omitempty"`	163		Include []string `json:"include,omitempty"`
164	Reasoning *codexReasoningConfig `json:"reasoning,omitempty"`	164		Reasoning *codexReasoningConfig `json:"reasoning,omitempty"`
165	PromptCacheKey string `json:"prompt_cache_key,omitempty"`	165		PromptCacheKey string `json:"prompt_cache_key,omitempty"`
		166	+	ServiceTier string `json:"service_tier,omitempty"`
166	}	167		}
167		168
168	// ---- Request building ----	169		// ---- Request building ----

@@ -180,6 +181,13 @@ func (c *codexClient) findModel(id string) (Model, error) {
180	}	181		}
181		182
182	func (c codexClient) buildRequest(req Request) (codexRequest, error) {	183		func (c codexClient) buildRequest(req Request) (codexRequest, error) {
		184	+	providerName := c.providerName
		185	+	if providerName == "" {
		186	+	providerName = "openai-codex"
		187	+	}
		188	+	if err := ValidateFastMode(providerName, req.FastMode); err != nil {
		189	+	return nil, err
		190	+	}
183	m, err := c.findModel(req.Model)	191		m, err := c.findModel(req.Model)
184	if err != nil {	192		if err != nil {
185	return nil, err	193		return nil, err

@@ -194,6 +202,9 @@ func (c codexClient) buildRequest(req Request) (codexRequest, error) {
194	ParallelToolCalls: true,	202		ParallelToolCalls: true,
195	Include: []string{"reasoning.encrypted_content"},	203		Include: []string{"reasoning.encrypted_content"},
196	}	204		}
		205	+	if req.FastMode {
		206	+	body.ServiceTier = fastModeServiceTier
		207	+	}
197	if m.Reasoning {	208		if m.Reasoning {
198	if effort := OpenAICodexReasoningEffort(reasoning, req.Model); effort != "" {	209		if effort := OpenAICodexReasoningEffort(reasoning, req.Model); effort != "" {
199	body.Reasoning = &codexReasoningConfig{Effort: effort}	210		body.Reasoning = &codexReasoningConfig{Effort: effort}

M packages/provider/provider.go +30 -0


@@ -8,6 +8,8 @@ package provider
8	8		import (
9	9		"context"
10	10		"encoding/json"
	11	+	"fmt"
	12	+	"strings"
11	13		"time"
12	14		)
13	15

@@ -255,6 +257,34 @@ type Request struct {
255	257		// "max". Empty disables reasoning. The max tier is sent natively only to
256	258		// models that support it and clamped elsewhere.
257	259		Reasoning string
	260	+	// FastMode requests OpenAI's fast service tier. Providers that do not
	261	+	// implement the OpenAI service-tier contract reject the request.
	262	+	FastMode bool
	263	+	}
	264	+
	265	+	// OpenAI's current Fast mode is represented as the priority service tier on
	266	+	// the wire, including the ChatGPT/Codex Responses endpoint.
	267	+	const fastModeServiceTier = "priority"
	268	+
	269	+	// SupportsFastMode reports whether a provider uses an OpenAI request
	270	+	// contract that supports the fast service tier.
	271	+	func SupportsFastMode(providerName string) bool {
	272	+	switch strings.ToLower(strings.TrimSpace(providerName)) {
	273	+	case "openai", "openai-codex", "openai-responses":
	274	+	return true
	275	+	default:
	276	+	return false
	277	+	}
	278	+	}
	279	+
	280	+	// ValidateFastMode rejects fast-mode requests before a provider performs
	281	+	// any network or credential work. Keep this check at the provider boundary
	282	+	// so direct Client users get the same contract as the core agent loop.
	283	+	func ValidateFastMode(providerName string, enabled bool) error {
	284	+	if !enabled \|\| SupportsFastMode(providerName) {
	285	+	return nil
	286	+	}
	287	+	return fmt.Errorf("fast mode is only supported for OpenAI providers, not %q", providerName)
258	288		}
259	289
260	290		// Client is an LLM streaming client.


@@ -8,6 +8,8 @@ package provider
8	import (	8		import (
9	"context"	9		"context"
10	"encoding/json"	10		"encoding/json"
		11	+	"fmt"
		12	+	"strings"
11	"time"	13		"time"
12	)	14		)
13		15

@@ -255,6 +257,34 @@ type Request struct {
255	// "max". Empty disables reasoning. The max tier is sent natively only to	257		// "max". Empty disables reasoning. The max tier is sent natively only to
256	// models that support it and clamped elsewhere.	258		// models that support it and clamped elsewhere.
257	Reasoning string	259		Reasoning string
		260	+	// FastMode requests OpenAI's fast service tier. Providers that do not
		261	+	// implement the OpenAI service-tier contract reject the request.
		262	+	FastMode bool
		263	+	}
		264	+
		265	+	// OpenAI's current Fast mode is represented as the priority service tier on
		266	+	// the wire, including the ChatGPT/Codex Responses endpoint.
		267	+	const fastModeServiceTier = "priority"
		268	+
		269	+	// SupportsFastMode reports whether a provider uses an OpenAI request
		270	+	// contract that supports the fast service tier.
		271	+	func SupportsFastMode(providerName string) bool {
		272	+	switch strings.ToLower(strings.TrimSpace(providerName)) {
		273	+	case "openai", "openai-codex", "openai-responses":
		274	+	return true
		275	+	default:
		276	+	return false
		277	+	}
		278	+	}
		279	+
		280	+	// ValidateFastMode rejects fast-mode requests before a provider performs
		281	+	// any network or credential work. Keep this check at the provider boundary
		282	+	// so direct Client users get the same contract as the core agent loop.
		283	+	func ValidateFastMode(providerName string, enabled bool) error {
		284	+	if !enabled \|\| SupportsFastMode(providerName) {
		285	+	return nil
		286	+	}
		287	+	return fmt.Errorf("fast mode is only supported for OpenAI providers, not %q", providerName)
258	}	288		}
259		289
260	// Client is an LLM streaming client.	290		// Client is an LLM streaming client.

M packages/provider/refreshing.go +3 -0


@@ -41,6 +41,9 @@ func (c *RefreshingClient) Name() string {
41	41		}
42	42
43	43		func (c *RefreshingClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {
	44	+	if err := ValidateFastMode(c.Name(), req.FastMode); err != nil {
	45	+	return nil, err
	46	+	}
44	47		if c.refresh != nil {
45	48		if newToken, err := c.refresh(ctx); err == nil && newToken != "" {
46	49		c.mu.Lock()


@@ -41,6 +41,9 @@ func (c *RefreshingClient) Name() string {
41	}	41		}
42		42
43	func (c *RefreshingClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {	43		func (c *RefreshingClient) Stream(ctx context.Context, req Request) (<-chan Event, error) {
		44	+	if err := ValidateFastMode(c.Name(), req.FastMode); err != nil {
		45	+	return nil, err
		46	+	}
44	if c.refresh != nil {	47		if c.refresh != nil {
45	if newToken, err := c.refresh(ctx); err == nil && newToken != "" {	48		if newToken, err := c.refresh(ctx); err == nil && newToken != "" {
46	c.mu.Lock()	49		c.mu.Lock()