feat: integrate upstream provider and proxy improvements

commit: da49ca5bf7fb146a6814b4fd8d7427a9dab548ac
author: brice · 2mo ago · 2026-08-10 13:50 UTC
committer: brice · 2mo ago · 2026-08-11 07:21 UTC
parent: 59c2524
signature: GPG VerifiedCD88F9E564E0102646FB6077A1A07C7C7C3E349D · arch-gpg Verified for the committer email

11 files · +271 -8

M README.md +12 -0


@@ -151,6 +151,18 @@ Ponytail coding mode is enabled by default, including when an existing `config.j
151	151
152	152		A selected subagent profile with `systemPromptMode: replace` supplies the child's base identity even when `--system-prompt` was provided for the parent run. Global append addenda, including enabled Ponytail guidance, still follow their normal inclusion rules.
153	153
	154	+	### HTTP proxy
	155	+
	156	+	To route zut-managed HTTP and HTTPS requests through one proxy, add `http_proxy` to `$ZUT_HOME/config.json`:
	157	+
	158	+	```json
	159	+	{
	160	+	"http_proxy": "http://127.0.0.1:7890"
	161	+	}
	162	+	```
	163	+
	164	+	The setting is applied at startup to both HTTP and HTTPS traffic. Existing `HTTP_PROXY`, `HTTPS_PROXY`, `http_proxy`, and `https_proxy` environment variables take precedence for their corresponding protocol. `NO_PROXY` and `no_proxy` continue to control bypasses. Restart zut after changing the config file. If the URL contains proxy credentials, prefer protected environment variables because `config.json` is not a credential store.
	165	+
154	166		## Persistent instructions (AGENTS.md)
155	167
156	168		Use `AGENTS.md` to give zut standing instructions that layer on top of the default system prompt, without replacing it. This is the friendliest way to shape behavior (for example, taming local models that jump straight to code edits) because it adds guidance rather than taking over the base identity the way `SYSTEM.md` does.


@@ -151,6 +151,18 @@ Ponytail coding mode is enabled by default, including when an existing `config.j
151		151
152	A selected subagent profile with `systemPromptMode: replace` supplies the child's base identity even when `--system-prompt` was provided for the parent run. Global append addenda, including enabled Ponytail guidance, still follow their normal inclusion rules.	152		A selected subagent profile with `systemPromptMode: replace` supplies the child's base identity even when `--system-prompt` was provided for the parent run. Global append addenda, including enabled Ponytail guidance, still follow their normal inclusion rules.
153		153
		154	+	### HTTP proxy
		155	+
		156	+	To route zut-managed HTTP and HTTPS requests through one proxy, add `http_proxy` to `$ZUT_HOME/config.json`:
		157	+
		158	+	```json
		159	+	{
		160	+	"http_proxy": "http://127.0.0.1:7890"
		161	+	}
		162	+	```
		163	+
		164	+	The setting is applied at startup to both HTTP and HTTPS traffic. Existing `HTTP_PROXY`, `HTTPS_PROXY`, `http_proxy`, and `https_proxy` environment variables take precedence for their corresponding protocol. `NO_PROXY` and `no_proxy` continue to control bypasses. Restart zut after changing the config file. If the URL contains proxy credentials, prefer protected environment variables because `config.json` is not a credential store.
		165	+
154	## Persistent instructions (AGENTS.md)	166		## Persistent instructions (AGENTS.md)
155		167
156	Use `AGENTS.md` to give zut standing instructions that layer on top of the default system prompt, without replacing it. This is the friendliest way to shape behavior (for example, taming local models that jump straight to code edits) because it adds guidance rather than taking over the base identity the way `SYSTEM.md` does.	168		Use `AGENTS.md` to give zut standing instructions that layer on top of the default system prompt, without replacing it. This is the friendliest way to shape behavior (for example, taming local models that jump straight to code edits) because it adds guidance rather than taking over the base identity the way `SYSTEM.md` does.

M docs/providers.md +14 -0


@@ -4,6 +4,20 @@ zut ships with built-in providers and a model catalog. You can select models
4	4		with `/model`, list them with `zut --list-models`, and add private models in
5	5		`$ZUT_HOME/models.json`.
6	6
	7	+	## HTTP proxy
	8	+
	9	+	Set one global proxy for zut-managed HTTP and HTTPS traffic with the `http_proxy` key in `$ZUT_HOME/config.json`:
	10	+
	11	+	```json
	12	+	{
	13	+	"http_proxy": "http://127.0.0.1:7890"
	14	+	}
	15	+	```
	16	+
	17	+	zut applies this setting at startup. Existing `HTTP_PROXY`, `HTTPS_PROXY`, `http_proxy`, and `https_proxy` environment variables take precedence for their corresponding protocol. Standard `NO_PROXY` and `no_proxy` bypass lists remain effective. Restart zut after changing the setting.
	18	+
	19	+	`config.json` is not a credential store. If the proxy URL contains a username or password, prefer setting the proxy environment variables in a protected shell or service configuration rather than saving those credentials in `config.json`.
	20	+
7	21		## Login methods
8	22
9	23		Use `/login` in interactive mode. Type in either provider picker to filter the list by provider ID or display name.


@@ -4,6 +4,20 @@ zut ships with built-in providers and a model catalog. You can select models
4	with `/model`, list them with `zut --list-models`, and add private models in	4		with `/model`, list them with `zut --list-models`, and add private models in
5	`$ZUT_HOME/models.json`.	5		`$ZUT_HOME/models.json`.
6		6
		7	+	## HTTP proxy
		8	+
		9	+	Set one global proxy for zut-managed HTTP and HTTPS traffic with the `http_proxy` key in `$ZUT_HOME/config.json`:
		10	+
		11	+	```json
		12	+	{
		13	+	"http_proxy": "http://127.0.0.1:7890"
		14	+	}
		15	+	```
		16	+
		17	+	zut applies this setting at startup. Existing `HTTP_PROXY`, `HTTPS_PROXY`, `http_proxy`, and `https_proxy` environment variables take precedence for their corresponding protocol. Standard `NO_PROXY` and `no_proxy` bypass lists remain effective. Restart zut after changing the setting.
		18	+
		19	+	`config.json` is not a credential store. If the proxy URL contains a username or password, prefer setting the proxy environment variables in a protected shell or service configuration rather than saving those credentials in `config.json`.
		20	+
7	## Login methods	21		## Login methods
8		22
9	Use `/login` in interactive mode. Type in either provider picker to filter the list by provider ID or display name.	23		Use `/login` in interactive mode. Type in either provider picker to filter the list by provider ID or display name.

M packages/agent/cli.go +4 -0


@@ -765,6 +765,10 @@ func fanoutAgentEvent(mgr *extensions.Manager, ev core.AgentEvent) {
765	765
766	766		// Run is the top-level entrypoint for the zut binary.
767	767		func Run(rawArgs []string, version string) error {
	768	+	// Apply network configuration before any subcommand can make an HTTP
	769	+	// request. Standard proxy environment variables retain precedence.
	770	+	applyConfiguredHTTPProxy()
	771	+
768	772		// Extension installs can invoke external build and clone processes.
769	773		// Give those processes a cancellation context without changing signal
770	774		// handling for the interactive modes below.


@@ -765,6 +765,10 @@ func fanoutAgentEvent(mgr *extensions.Manager, ev core.AgentEvent) {
765		765
766	// Run is the top-level entrypoint for the zut binary.	766		// Run is the top-level entrypoint for the zut binary.
767	func Run(rawArgs []string, version string) error {	767		func Run(rawArgs []string, version string) error {
		768	+	// Apply network configuration before any subcommand can make an HTTP
		769	+	// request. Standard proxy environment variables retain precedence.
		770	+	applyConfiguredHTTPProxy()
		771	+
768	// Extension installs can invoke external build and clone processes.	772		// Extension installs can invoke external build and clone processes.
769	// Give those processes a cancellation context without changing signal	773		// Give those processes a cancellation context without changing signal
770	// handling for the interactive modes below.	774		// handling for the interactive modes below.

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


@@ -181,6 +181,10 @@ type Config struct {
181	181		// Insecure skips TLS verification for custom inference endpoints.
182	182		Insecure bool `json:"insecure,omitempty"`
183	183
	184	+	// HTTPProxy is a global proxy URL used for HTTP and HTTPS requests when
	185	+	// the corresponding standard proxy environment variable is not already set.
	186	+	HTTPProxy string `json:"http_proxy,omitempty"`
	187	+
184	188		// LastChangelogShown is the version whose release-notes
185	189		// dialog the user has already seen. When the running binary's
186	190		// version differs, the next interactive run shows the

@@ -332,6 +336,26 @@ func SaveConfig(c Config) error {
332	336		return os.WriteFile(ConfigPath(), b, 0o644)
333	337		}
334	338
	339	+	// applyConfiguredHTTPProxy makes the persisted proxy available to Go's
	340	+	// standard HTTP transport and to child processes. Explicit environment
	341	+	// variables take precedence over config.json.
	342	+	func applyConfiguredHTTPProxy() {
	343	+	cfg, err := LoadConfig()
	344	+	if err != nil {
	345	+	return
	346	+	}
	347	+	proxy := strings.TrimSpace(cfg.HTTPProxy)
	348	+	if proxy == "" {
	349	+	return
	350	+	}
	351	+	if os.Getenv("HTTP_PROXY") == "" && os.Getenv("http_proxy") == "" {
	352	+	_ = os.Setenv("HTTP_PROXY", proxy)
	353	+	}
	354	+	if os.Getenv("HTTPS_PROXY") == "" && os.Getenv("https_proxy") == "" {
	355	+	_ = os.Setenv("HTTPS_PROXY", proxy)
	356	+	}
	357	+	}
	358	+
335	359		// AuthStoreFor returns the auth.Store backed by AuthPath().
336	360		func AuthStoreFor() *auth.Store { return auth.NewStore(AuthPath()) }
337	361


@@ -181,6 +181,10 @@ type Config struct {
181	// Insecure skips TLS verification for custom inference endpoints.	181		// Insecure skips TLS verification for custom inference endpoints.
182	Insecure bool `json:"insecure,omitempty"`	182		Insecure bool `json:"insecure,omitempty"`
183		183
		184	+	// HTTPProxy is a global proxy URL used for HTTP and HTTPS requests when
		185	+	// the corresponding standard proxy environment variable is not already set.
		186	+	HTTPProxy string `json:"http_proxy,omitempty"`
		187	+
184	// LastChangelogShown is the version whose release-notes	188		// LastChangelogShown is the version whose release-notes
185	// dialog the user has already seen. When the running binary's	189		// dialog the user has already seen. When the running binary's
186	// version differs, the next interactive run shows the	190		// version differs, the next interactive run shows the

@@ -332,6 +336,26 @@ func SaveConfig(c Config) error {
332	return os.WriteFile(ConfigPath(), b, 0o644)	336		return os.WriteFile(ConfigPath(), b, 0o644)
333	}	337		}
334		338
		339	+	// applyConfiguredHTTPProxy makes the persisted proxy available to Go's
		340	+	// standard HTTP transport and to child processes. Explicit environment
		341	+	// variables take precedence over config.json.
		342	+	func applyConfiguredHTTPProxy() {
		343	+	cfg, err := LoadConfig()
		344	+	if err != nil {
		345	+	return
		346	+	}
		347	+	proxy := strings.TrimSpace(cfg.HTTPProxy)
		348	+	if proxy == "" {
		349	+	return
		350	+	}
		351	+	if os.Getenv("HTTP_PROXY") == "" && os.Getenv("http_proxy") == "" {
		352	+	_ = os.Setenv("HTTP_PROXY", proxy)
		353	+	}
		354	+	if os.Getenv("HTTPS_PROXY") == "" && os.Getenv("https_proxy") == "" {
		355	+	_ = os.Setenv("HTTPS_PROXY", proxy)
		356	+	}
		357	+	}
		358	+
335	// AuthStoreFor returns the auth.Store backed by AuthPath().	359		// AuthStoreFor returns the auth.Store backed by AuthPath().
336	func AuthStoreFor() *auth.Store { return auth.NewStore(AuthPath()) }	360		func AuthStoreFor() *auth.Store { return auth.NewStore(AuthPath()) }
337		361

A packages/agent/config_proxy_test.go +92 -0

@@ -0,0 +1,92 @@
	1	+	package agent
	2	+
	3	+	import (
	4	+	"os"
	5	+	"testing"
	6	+	)
	7	+
	8	+	func TestApplyConfiguredHTTPProxy(t *testing.T) {
	9	+	t.Setenv("ZUT_HOME", t.TempDir())
	10	+	for _, key := range []string{"HTTP_PROXY", "HTTPS_PROXY", "http_proxy", "https_proxy"} {
	11	+	t.Setenv(key, "")
	12	+	}
	13	+	if err := SaveConfig(Config{HTTPProxy: " http://proxy.example:8080 "}); err != nil {
	14	+	t.Fatal(err)
	15	+	}
	16	+
	17	+	applyConfiguredHTTPProxy()
	18	+
	19	+	if got := getenvForTest("HTTP_PROXY"); got != "http://proxy.example:8080" {
	20	+	t.Fatalf("HTTP_PROXY = %q", got)
	21	+	}
	22	+	if got := getenvForTest("HTTPS_PROXY"); got != "http://proxy.example:8080" {
	23	+	t.Fatalf("HTTPS_PROXY = %q", got)
	24	+	}
	25	+	}
	26	+
	27	+	func TestApplyConfiguredHTTPProxyPreservesUppercaseEnvironment(t *testing.T) {
	28	+	t.Setenv("ZUT_HOME", t.TempDir())
	29	+	// Set lowercase first because environment variable names are
	30	+	// case-insensitive on Windows.
	31	+	t.Setenv("http_proxy", "")
	32	+	t.Setenv("https_proxy", "")
	33	+	t.Setenv("HTTP_PROXY", "http://http-env.example:8080")
	34	+	t.Setenv("HTTPS_PROXY", "http://https-env.example:8080")
	35	+	if err := SaveConfig(Config{HTTPProxy: "http://config.example:8080"}); err != nil {
	36	+	t.Fatal(err)
	37	+	}
	38	+
	39	+	applyConfiguredHTTPProxy()
	40	+
	41	+	if got := getenvForTest("HTTP_PROXY"); got != "http://http-env.example:8080" {
	42	+	t.Fatalf("HTTP_PROXY = %q", got)
	43	+	}
	44	+	if got := getenvForTest("HTTPS_PROXY"); got != "http://https-env.example:8080" {
	45	+	t.Fatalf("HTTPS_PROXY = %q", got)
	46	+	}
	47	+	}
	48	+
	49	+	func TestApplyConfiguredHTTPProxyPreservesLowercaseEnvironment(t *testing.T) {
	50	+	t.Setenv("ZUT_HOME", t.TempDir())
	51	+	// Set uppercase first because environment variable names are
	52	+	// case-insensitive on Windows.
	53	+	t.Setenv("HTTP_PROXY", "")
	54	+	t.Setenv("HTTPS_PROXY", "")
	55	+	t.Setenv("http_proxy", "http://http-env.example:8080")
	56	+	t.Setenv("https_proxy", "http://https-env.example:8080")
	57	+	if err := SaveConfig(Config{HTTPProxy: "http://config.example:8080"}); err != nil {
	58	+	t.Fatal(err)
	59	+	}
	60	+
	61	+	applyConfiguredHTTPProxy()
	62	+
	63	+	if got := getenvForTest("http_proxy"); got != "http://http-env.example:8080" {
	64	+	t.Fatalf("http_proxy = %q", got)
	65	+	}
	66	+	if got := getenvForTest("https_proxy"); got != "http://https-env.example:8080" {
	67	+	t.Fatalf("https_proxy = %q", got)
	68	+	}
	69	+	}
	70	+
	71	+	func TestApplyConfiguredHTTPProxyIgnoresEmptySetting(t *testing.T) {
	72	+	t.Setenv("ZUT_HOME", t.TempDir())
	73	+	for _, key := range []string{"HTTP_PROXY", "HTTPS_PROXY", "http_proxy", "https_proxy"} {
	74	+	t.Setenv(key, "")
	75	+	}
	76	+	if err := SaveConfig(Config{HTTPProxy: " "}); err != nil {
	77	+	t.Fatal(err)
	78	+	}
	79	+
	80	+	applyConfiguredHTTPProxy()
	81	+
	82	+	if got := getenvForTest("HTTP_PROXY"); got != "" {
	83	+	t.Fatalf("HTTP_PROXY = %q", got)
	84	+	}
	85	+	if got := getenvForTest("HTTPS_PROXY"); got != "" {
	86	+	t.Fatalf("HTTPS_PROXY = %q", got)
	87	+	}
	88	+	}
	89	+
	90	+	func getenvForTest(key string) string {
	91	+	return os.Getenv(key)
	92	+	}

@@ -0,0 +1,92 @@
	1	+	package agent
	2	+
	3	+	import (
	4	+	"os"
	5	+	"testing"
	6	+	)
	7	+
	8	+	func TestApplyConfiguredHTTPProxy(t *testing.T) {
	9	+	t.Setenv("ZUT_HOME", t.TempDir())
	10	+	for _, key := range []string{"HTTP_PROXY", "HTTPS_PROXY", "http_proxy", "https_proxy"} {
	11	+	t.Setenv(key, "")
	12	+	}
	13	+	if err := SaveConfig(Config{HTTPProxy: " http://proxy.example:8080 "}); err != nil {
	14	+	t.Fatal(err)
	15	+	}
	16	+
	17	+	applyConfiguredHTTPProxy()
	18	+
	19	+	if got := getenvForTest("HTTP_PROXY"); got != "http://proxy.example:8080" {
	20	+	t.Fatalf("HTTP_PROXY = %q", got)
	21	+	}
	22	+	if got := getenvForTest("HTTPS_PROXY"); got != "http://proxy.example:8080" {
	23	+	t.Fatalf("HTTPS_PROXY = %q", got)
	24	+	}
	25	+	}
	26	+
	27	+	func TestApplyConfiguredHTTPProxyPreservesUppercaseEnvironment(t *testing.T) {
	28	+	t.Setenv("ZUT_HOME", t.TempDir())
	29	+	// Set lowercase first because environment variable names are
	30	+	// case-insensitive on Windows.
	31	+	t.Setenv("http_proxy", "")
	32	+	t.Setenv("https_proxy", "")
	33	+	t.Setenv("HTTP_PROXY", "http://http-env.example:8080")
	34	+	t.Setenv("HTTPS_PROXY", "http://https-env.example:8080")
	35	+	if err := SaveConfig(Config{HTTPProxy: "http://config.example:8080"}); err != nil {
	36	+	t.Fatal(err)
	37	+	}
	38	+
	39	+	applyConfiguredHTTPProxy()
	40	+
	41	+	if got := getenvForTest("HTTP_PROXY"); got != "http://http-env.example:8080" {
	42	+	t.Fatalf("HTTP_PROXY = %q", got)
	43	+	}
	44	+	if got := getenvForTest("HTTPS_PROXY"); got != "http://https-env.example:8080" {
	45	+	t.Fatalf("HTTPS_PROXY = %q", got)
	46	+	}
	47	+	}
	48	+
	49	+	func TestApplyConfiguredHTTPProxyPreservesLowercaseEnvironment(t *testing.T) {
	50	+	t.Setenv("ZUT_HOME", t.TempDir())
	51	+	// Set uppercase first because environment variable names are
	52	+	// case-insensitive on Windows.
	53	+	t.Setenv("HTTP_PROXY", "")
	54	+	t.Setenv("HTTPS_PROXY", "")
	55	+	t.Setenv("http_proxy", "http://http-env.example:8080")
	56	+	t.Setenv("https_proxy", "http://https-env.example:8080")
	57	+	if err := SaveConfig(Config{HTTPProxy: "http://config.example:8080"}); err != nil {
	58	+	t.Fatal(err)
	59	+	}
	60	+
	61	+	applyConfiguredHTTPProxy()
	62	+
	63	+	if got := getenvForTest("http_proxy"); got != "http://http-env.example:8080" {
	64	+	t.Fatalf("http_proxy = %q", got)
	65	+	}
	66	+	if got := getenvForTest("https_proxy"); got != "http://https-env.example:8080" {
	67	+	t.Fatalf("https_proxy = %q", got)
	68	+	}
	69	+	}
	70	+
	71	+	func TestApplyConfiguredHTTPProxyIgnoresEmptySetting(t *testing.T) {
	72	+	t.Setenv("ZUT_HOME", t.TempDir())
	73	+	for _, key := range []string{"HTTP_PROXY", "HTTPS_PROXY", "http_proxy", "https_proxy"} {
	74	+	t.Setenv(key, "")
	75	+	}
	76	+	if err := SaveConfig(Config{HTTPProxy: " "}); err != nil {
	77	+	t.Fatal(err)
	78	+	}
	79	+
	80	+	applyConfiguredHTTPProxy()
	81	+
	82	+	if got := getenvForTest("HTTP_PROXY"); got != "" {
	83	+	t.Fatalf("HTTP_PROXY = %q", got)
	84	+	}
	85	+	if got := getenvForTest("HTTPS_PROXY"); got != "" {
	86	+	t.Fatalf("HTTPS_PROXY = %q", got)
	87	+	}
	88	+	}
	89	+
	90	+	func getenvForTest(key string) string {
	91	+	return os.Getenv(key)
	92	+	}

M packages/provider/catalog_builtin.go +1 -0


@@ -308,6 +308,7 @@ var builtinCatalog = []Model{
308	308		{Provider: "opencode-go", ID: "glm-5", DisplayName: "GLM-5", ContextWindow: 202752, MaxOutput: 32768, Reasoning: true, PriceInput: 1, PriceOutput: 3.2, PriceCacheRead: 0.2, BaseURL: "https://opencode.ai/zen/go/v1"},
309	309		{Provider: "opencode-go", ID: "glm-5.1", DisplayName: "GLM-5.1", ContextWindow: 202752, MaxOutput: 32768, Reasoning: true, PriceInput: 1.4, PriceOutput: 4.4, PriceCacheRead: 0.26, BaseURL: "https://opencode.ai/zen/go/v1"},
310	310		{Provider: "opencode-go", ID: "glm-5.2", DisplayName: "GLM-5.2", ContextWindow: 1024000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.91, PriceOutput: 2.86, PriceCacheRead: 0.169, BaseURL: "https://opencode.ai/zen/go/v1"},
	311	+	{Provider: "opencode-go", ID: "gpt-5.6-luna", DisplayName: "GPT-5.6 Luna", API: APIResponses, ContextWindow: 1050000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.2, PriceOutput: 1.2, PriceCacheRead: 0.02, PriceCacheWrite: 0.25, PriceTierInputTokens: 272000, PriceInputAbove: 0.4, PriceOutputAbove: 1.8, PriceCacheReadAbove: 0.04, PriceCacheWriteAbove: 0.5, BaseURL: "https://opencode.ai/zen/go/v1"},
311	312		{Provider: "opencode-go", ID: "grok-4.5", DisplayName: "Grok 4.5", ContextWindow: 500000, MaxOutput: 500000, Reasoning: true, PriceInput: 2, PriceOutput: 6, PriceCacheRead: 0.3, BaseURL: "https://opencode.ai/zen/go/v1"},
312	313		{Provider: "opencode-go", ID: "kimi-k2.5", DisplayName: "Kimi K2.5", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.6, PriceOutput: 3, PriceCacheRead: 0.1, BaseURL: "https://opencode.ai/zen/go/v1"},
313	314		{Provider: "opencode-go", ID: "kimi-k2.6", DisplayName: "Kimi K2.6", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.95, PriceOutput: 4, PriceCacheRead: 0.16, BaseURL: "https://opencode.ai/zen/go/v1"},


@@ -308,6 +308,7 @@ var builtinCatalog = []Model{
308	{Provider: "opencode-go", ID: "glm-5", DisplayName: "GLM-5", ContextWindow: 202752, MaxOutput: 32768, Reasoning: true, PriceInput: 1, PriceOutput: 3.2, PriceCacheRead: 0.2, BaseURL: "https://opencode.ai/zen/go/v1"},	308		{Provider: "opencode-go", ID: "glm-5", DisplayName: "GLM-5", ContextWindow: 202752, MaxOutput: 32768, Reasoning: true, PriceInput: 1, PriceOutput: 3.2, PriceCacheRead: 0.2, BaseURL: "https://opencode.ai/zen/go/v1"},
309	{Provider: "opencode-go", ID: "glm-5.1", DisplayName: "GLM-5.1", ContextWindow: 202752, MaxOutput: 32768, Reasoning: true, PriceInput: 1.4, PriceOutput: 4.4, PriceCacheRead: 0.26, BaseURL: "https://opencode.ai/zen/go/v1"},	309		{Provider: "opencode-go", ID: "glm-5.1", DisplayName: "GLM-5.1", ContextWindow: 202752, MaxOutput: 32768, Reasoning: true, PriceInput: 1.4, PriceOutput: 4.4, PriceCacheRead: 0.26, BaseURL: "https://opencode.ai/zen/go/v1"},
310	{Provider: "opencode-go", ID: "glm-5.2", DisplayName: "GLM-5.2", ContextWindow: 1024000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.91, PriceOutput: 2.86, PriceCacheRead: 0.169, BaseURL: "https://opencode.ai/zen/go/v1"},	310		{Provider: "opencode-go", ID: "glm-5.2", DisplayName: "GLM-5.2", ContextWindow: 1024000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.91, PriceOutput: 2.86, PriceCacheRead: 0.169, BaseURL: "https://opencode.ai/zen/go/v1"},
		311	+	{Provider: "opencode-go", ID: "gpt-5.6-luna", DisplayName: "GPT-5.6 Luna", API: APIResponses, ContextWindow: 1050000, MaxOutput: 128000, Reasoning: true, PriceInput: 0.2, PriceOutput: 1.2, PriceCacheRead: 0.02, PriceCacheWrite: 0.25, PriceTierInputTokens: 272000, PriceInputAbove: 0.4, PriceOutputAbove: 1.8, PriceCacheReadAbove: 0.04, PriceCacheWriteAbove: 0.5, BaseURL: "https://opencode.ai/zen/go/v1"},
311	{Provider: "opencode-go", ID: "grok-4.5", DisplayName: "Grok 4.5", ContextWindow: 500000, MaxOutput: 500000, Reasoning: true, PriceInput: 2, PriceOutput: 6, PriceCacheRead: 0.3, BaseURL: "https://opencode.ai/zen/go/v1"},	312		{Provider: "opencode-go", ID: "grok-4.5", DisplayName: "Grok 4.5", ContextWindow: 500000, MaxOutput: 500000, Reasoning: true, PriceInput: 2, PriceOutput: 6, PriceCacheRead: 0.3, BaseURL: "https://opencode.ai/zen/go/v1"},
312	{Provider: "opencode-go", ID: "kimi-k2.5", DisplayName: "Kimi K2.5", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.6, PriceOutput: 3, PriceCacheRead: 0.1, BaseURL: "https://opencode.ai/zen/go/v1"},	313		{Provider: "opencode-go", ID: "kimi-k2.5", DisplayName: "Kimi K2.5", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.6, PriceOutput: 3, PriceCacheRead: 0.1, BaseURL: "https://opencode.ai/zen/go/v1"},
313	{Provider: "opencode-go", ID: "kimi-k2.6", DisplayName: "Kimi K2.6", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.95, PriceOutput: 4, PriceCacheRead: 0.16, BaseURL: "https://opencode.ai/zen/go/v1"},	314		{Provider: "opencode-go", ID: "kimi-k2.6", DisplayName: "Kimi K2.6", ContextWindow: 262144, MaxOutput: 65536, Reasoning: true, PriceInput: 0.95, PriceOutput: 4, PriceCacheRead: 0.16, BaseURL: "https://opencode.ai/zen/go/v1"},

M packages/provider/extra_providers.go +7 -2


@@ -124,9 +124,14 @@ func NewOpenCode(apiKey, baseURL string) Client {
124	124		return NewOpenAICompat("opencode", apiKey, baseURL, "https://opencode.ai/zen/v1")
125	125		}
126	126
127		-	// NewOpenCodeGo is the opencode-go variant.
	127	+	// NewOpenCodeGo routes each model through the wire API exposed by OpenCode Go.
128	128		func NewOpenCodeGo(apiKey, baseURL string) Client {
129		-	return NewOpenAICompat("opencode-go", apiKey, baseURL, "https://opencode.ai/zen/go/v1")
	129	+	const defaultBaseURL = "https://opencode.ai/zen/go/v1"
	130	+	return NewModelRouter("opencode-go",
	131	+	NewOpenAICompat("opencode-go", apiKey, baseURL, defaultBaseURL),
	132	+	map[string]Client{
	133	+	APIResponses: NewOpenAIResponsesNamed(apiKey, firstNonEmptyString(baseURL, defaultBaseURL), "opencode-go"),
	134	+	})
130	135		}
131	136
132	137		// NewMinimaxOpenAI is the OpenAI-completions flavor of MiniMax, in case

 	return NewOpenAICompat("opencode", apiKey, baseURL, "https://opencode.ai/zen/v1")
 }
-// NewOpenCodeGo is the opencode-go variant.
+// NewOpenCodeGo routes each model through the wire API exposed by OpenCode Go.
 func NewOpenCodeGo(apiKey, baseURL string) Client {
-	return NewOpenAICompat("opencode-go", apiKey, baseURL, "https://opencode.ai/zen/go/v1")
+	const defaultBaseURL = "https://opencode.ai/zen/go/v1"
+	return NewModelRouter("opencode-go",
+		NewOpenAICompat("opencode-go", apiKey, baseURL, defaultBaseURL),
+		map[string]Client{
+			APIResponses: NewOpenAIResponsesNamed(apiKey, firstNonEmptyString(baseURL, defaultBaseURL), "opencode-go"),
+		})
 }
 // NewMinimaxOpenAI is the OpenAI-completions flavor of MiniMax, in case

M packages/provider/llamacpp.go +16 -2


@@ -47,6 +47,10 @@ type LlamaCPPModel struct {
47	47		Size int64 `json:"size,omitempty"`
48	48		FileType string `json:"ftype,omitempty"`
49	49		} `json:"meta,omitempty"`
	50	+	// Progress is emitted at the model's top level by current llama.cpp
	51	+	// routers. Status.Progress supports versions that follow the documented
	52	+	// nested shape.
	53	+	Progress map[string]LlamaCPPBytes `json:"progress,omitempty"`
50	54		}
51	55
52	56		// LlamaCPPBytes describes byte progress for one downloaded file.

@@ -204,7 +208,13 @@ func (c *LlamaCPPClient) watch(ctx context.Context, events chan<- llamaCPPEvent)
204	208		if c.APIKey != "" {
205	209		req.Header.Set("authorization", "Bearer "+c.APIKey)
206	210		}
207		-	resp, err := c.HTTP.Do(req)
	211	+	// http.Client.Timeout covers the complete response body lifetime, which is
	212	+	// appropriate for ordinary management requests but would terminate this
	213	+	// long-lived SSE stream. Preserve the configured transport and other client
	214	+	// behavior while letting ctx control the stream lifetime.
	215	+	streamClient := *c.HTTP
	216	+	streamClient.Timeout = 0
	217	+	resp, err := streamClient.Do(req)
208	218		if err != nil {
209	219		return err
210	220		}

@@ -408,7 +418,11 @@ func (c *LlamaCPPClient) DownloadAndWait(ctx context.Context, model string, upda
408	418		}
409	419		if entry.Status.Value == "downloading" {
410	420		sawDownload = true
411		-	if progress, ok := downloadProgress(entry.Status.Progress); ok {
	421	+	files := entry.Status.Progress
	422	+	if len(files) == 0 {
	423	+	files = entry.Progress
	424	+	}
	425	+	if progress, ok := downloadProgress(files); ok {
412	426		update(progress)
413	427		}
414	428		} else if sawDownload \|\| polls >= 2 {

 		Size         int64  `json:"size,omitempty"`
 		FileType     string `json:"ftype,omitempty"`
 	} `json:"meta,omitempty"`
+	// Progress is emitted at the model's top level by current llama.cpp
+	// routers. Status.Progress supports versions that follow the documented
+	// nested shape.
+	Progress map[string]LlamaCPPBytes `json:"progress,omitempty"`
 }
 // LlamaCPPBytes describes byte progress for one downloaded file.
 	if c.APIKey != "" {
 		req.Header.Set("authorization", "Bearer "+c.APIKey)
 	}
-	resp, err := c.HTTP.Do(req)
+	// http.Client.Timeout covers the complete response body lifetime, which is
+	// appropriate for ordinary management requests but would terminate this
+	// long-lived SSE stream. Preserve the configured transport and other client
+	// behavior while letting ctx control the stream lifetime.
+	streamClient := *c.HTTP
+	streamClient.Timeout = 0
+	resp, err := streamClient.Do(req)
 	if err != nil {
 		return err
 	}
 				}
 				if entry.Status.Value == "downloading" {
 					sawDownload = true
-					if progress, ok := downloadProgress(entry.Status.Progress); ok {
+					files := entry.Status.Progress
+					if len(files) == 0 {
+						files = entry.Progress
+					}
+					if progress, ok := downloadProgress(files); ok {
 						update(progress)
 					}
 				} else if sawDownload || polls >= 2 {

M packages/provider/llamacpp_test.go +56 -0


@@ -32,6 +32,62 @@ func TestLlamaCPPLoadProgressStages(t *testing.T) {
32	32		}
33	33		}
34	34
	35	+	func TestLlamaCPPWatchOutlivesRequestTimeout(t *testing.T) {
	36	+	requestStarted := make(chan struct{})
	37	+	releaseEvent := make(chan struct{})
	38	+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
	39	+	w.Header().Set("content-type", "text/event-stream")
	40	+	w.WriteHeader(http.StatusOK)
	41	+	w.(http.Flusher).Flush()
	42	+	close(requestStarted)
	43	+	select {
	44	+	case <-releaseEvent:
	45	+	fmt.Fprint(w, "data: {\"model\":\"test\",\"event\":\"download_progress\",\"data\":{}}\n\n")
	46	+	w.(http.Flusher).Flush()
	47	+	case <-r.Context().Done():
	48	+	}
	49	+	}))
	50	+	defer server.Close()
	51	+
	52	+	client, err := NewLlamaCPPClient(server.URL, "")
	53	+	if err != nil {
	54	+	t.Fatal(err)
	55	+	}
	56	+	client.HTTP.Timeout = 20 * time.Millisecond
	57	+	ctx, cancel := context.WithCancel(context.Background())
	58	+	defer cancel()
	59	+	events := make(chan llamaCPPEvent, 1)
	60	+	watchDone := make(chan error, 1)
	61	+	go func() { watchDone <- client.watch(ctx, events) }()
	62	+	<-requestStarted
	63	+
	64	+	select {
	65	+	case err := <-watchDone:
	66	+	t.Fatalf("watch stopped at ordinary request timeout: %v", err)
	67	+	case <-time.After(3 * client.HTTP.Timeout):
	68	+	}
	69	+	close(releaseEvent)
	70	+	select {
	71	+	case event := <-events:
	72	+	if event.Model != "test" \|\| event.Event != "download_progress" {
	73	+	t.Fatalf("event = %#v", event)
	74	+	}
	75	+	case <-time.After(time.Second):
	76	+	t.Fatal("timed out waiting for delayed SSE event")
	77	+	}
	78	+	}
	79	+
	80	+	func TestLlamaCPPModelAcceptsTopLevelDownloadProgress(t *testing.T) {
	81	+	var model LlamaCPPModel
	82	+	if err := json.Unmarshal([]byte(`{"id":"test","status":{"value":"downloading"},"progress":{"model.gguf":{"done":512,"total":1024}}}`), &model); err != nil {
	83	+	t.Fatal(err)
	84	+	}
	85	+	progress, ok := downloadProgress(model.Progress)
	86	+	if !ok \|\| progress.Ratio != 0.5 {
	87	+	t.Fatalf("progress = %#v, ok = %v", progress, ok)
	88	+	}
	89	+	}
	90	+
35	91		func TestLlamaCPPClientLoadAndDownloadProgress(t *testing.T) {
36	92		var mu sync.Mutex
37	93		status := "unloaded"


@@ -32,6 +32,62 @@ func TestLlamaCPPLoadProgressStages(t *testing.T) {
32	}	32		}
33	}	33		}
34		34
		35	+	func TestLlamaCPPWatchOutlivesRequestTimeout(t *testing.T) {
		36	+	requestStarted := make(chan struct{})
		37	+	releaseEvent := make(chan struct{})
		38	+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		39	+	w.Header().Set("content-type", "text/event-stream")
		40	+	w.WriteHeader(http.StatusOK)
		41	+	w.(http.Flusher).Flush()
		42	+	close(requestStarted)
		43	+	select {
		44	+	case <-releaseEvent:
		45	+	fmt.Fprint(w, "data: {\"model\":\"test\",\"event\":\"download_progress\",\"data\":{}}\n\n")
		46	+	w.(http.Flusher).Flush()
		47	+	case <-r.Context().Done():
		48	+	}
		49	+	}))
		50	+	defer server.Close()
		51	+
		52	+	client, err := NewLlamaCPPClient(server.URL, "")
		53	+	if err != nil {
		54	+	t.Fatal(err)
		55	+	}
		56	+	client.HTTP.Timeout = 20 * time.Millisecond
		57	+	ctx, cancel := context.WithCancel(context.Background())
		58	+	defer cancel()
		59	+	events := make(chan llamaCPPEvent, 1)
		60	+	watchDone := make(chan error, 1)
		61	+	go func() { watchDone <- client.watch(ctx, events) }()
		62	+	<-requestStarted
		63	+
		64	+	select {
		65	+	case err := <-watchDone:
		66	+	t.Fatalf("watch stopped at ordinary request timeout: %v", err)
		67	+	case <-time.After(3 * client.HTTP.Timeout):
		68	+	}
		69	+	close(releaseEvent)
		70	+	select {
		71	+	case event := <-events:
		72	+	if event.Model != "test" \|\| event.Event != "download_progress" {
		73	+	t.Fatalf("event = %#v", event)
		74	+	}
		75	+	case <-time.After(time.Second):
		76	+	t.Fatal("timed out waiting for delayed SSE event")
		77	+	}
		78	+	}
		79	+
		80	+	func TestLlamaCPPModelAcceptsTopLevelDownloadProgress(t *testing.T) {
		81	+	var model LlamaCPPModel
		82	+	if err := json.Unmarshal([]byte(`{"id":"test","status":{"value":"downloading"},"progress":{"model.gguf":{"done":512,"total":1024}}}`), &model); err != nil {
		83	+	t.Fatal(err)
		84	+	}
		85	+	progress, ok := downloadProgress(model.Progress)
		86	+	if !ok \|\| progress.Ratio != 0.5 {
		87	+	t.Fatalf("progress = %#v, ok = %v", progress, ok)
		88	+	}
		89	+	}
		90	+
35	func TestLlamaCPPClientLoadAndDownloadProgress(t *testing.T) {	91		func TestLlamaCPPClientLoadAndDownloadProgress(t *testing.T) {
36	var mu sync.Mutex	92		var mu sync.Mutex
37	status := "unloaded"	93		status := "unloaded"

M packages/provider/model_router_test.go +31 -0


@@ -48,3 +48,34 @@ func TestModelRouterRejectsMissingAPIClient(t *testing.T) {
48	48		t.Fatal("expected missing API client error")
49	49		}
50	50		}
	51	+
	52	+	func TestOpenCodeGoRoutesLunaToResponses(t *testing.T) {
	53	+	router := NewOpenCodeGo("token", "https://example.com/go/v1").(*modelRouter)
	54	+	if got := router.fallback.(*openaiClient).baseURL; got != "https://example.com/go/v1" {
	55	+	t.Fatalf("Completions base URL = %q", got)
	56	+	}
	57	+	responses := router.byAPI[APIResponses].(renamedClient).inner.(codexClient)
	58	+	if got := responses.baseURL; got != "https://example.com/go/v1/responses" {
	59	+	t.Fatalf("Responses base URL = %q", got)
	60	+	}
	61	+
	62	+	completionsCapture := &routeCaptureClient{name: "opencode-go"}
	63	+	responsesCapture := &routeCaptureClient{name: "opencode-go"}
	64	+	router.fallback = completionsCapture
	65	+	router.byAPI[APIResponses] = responsesCapture
	66	+	for _, model := range []string{"gpt-5.6-luna", "kimi-k3"} {
	67	+	stream, err := router.Stream(context.Background(), Request{Model: model})
	68	+	if err != nil {
	69	+	t.Fatal(err)
	70	+	}
	71	+	for range stream {
	72	+	}
	73	+	}
	74	+
	75	+	if len(responsesCapture.models) != 1 \|\| responsesCapture.models[0] != "gpt-5.6-luna" {
	76	+	t.Fatalf("Responses models = %v", responsesCapture.models)
	77	+	}
	78	+	if len(completionsCapture.models) != 1 \|\| completionsCapture.models[0] != "kimi-k3" {
	79	+	t.Fatalf("Completions models = %v", completionsCapture.models)
	80	+	}
	81	+	}


@@ -48,3 +48,34 @@ func TestModelRouterRejectsMissingAPIClient(t *testing.T) {
48	t.Fatal("expected missing API client error")	48		t.Fatal("expected missing API client error")
49	}	49		}
50	}	50		}
		51	+
		52	+	func TestOpenCodeGoRoutesLunaToResponses(t *testing.T) {
		53	+	router := NewOpenCodeGo("token", "https://example.com/go/v1").(*modelRouter)
		54	+	if got := router.fallback.(*openaiClient).baseURL; got != "https://example.com/go/v1" {
		55	+	t.Fatalf("Completions base URL = %q", got)
		56	+	}
		57	+	responses := router.byAPI[APIResponses].(renamedClient).inner.(codexClient)
		58	+	if got := responses.baseURL; got != "https://example.com/go/v1/responses" {
		59	+	t.Fatalf("Responses base URL = %q", got)
		60	+	}
		61	+
		62	+	completionsCapture := &routeCaptureClient{name: "opencode-go"}
		63	+	responsesCapture := &routeCaptureClient{name: "opencode-go"}
		64	+	router.fallback = completionsCapture
		65	+	router.byAPI[APIResponses] = responsesCapture
		66	+	for _, model := range []string{"gpt-5.6-luna", "kimi-k3"} {
		67	+	stream, err := router.Stream(context.Background(), Request{Model: model})
		68	+	if err != nil {
		69	+	t.Fatal(err)
		70	+	}
		71	+	for range stream {
		72	+	}
		73	+	}
		74	+
		75	+	if len(responsesCapture.models) != 1 \|\| responsesCapture.models[0] != "gpt-5.6-luna" {
		76	+	t.Fatalf("Responses models = %v", responsesCapture.models)
		77	+	}
		78	+	if len(completionsCapture.models) != 1 \|\| completionsCapture.models[0] != "kimi-k3" {
		79	+	t.Fatalf("Completions models = %v", completionsCapture.models)
		80	+	}
		81	+	}

M packages/provider/provider_test.go +14 -4


@@ -649,10 +649,8 @@ func TestGPT56CatalogEntries(t *testing.T) {
649	649		}
650	650		}
651	651
652		-	// TestOpenCodeGoCatalog pins the Grok 4.5 and Kimi K3 entries added to
653		-	// the OpenCode Go provider. Both are listed in the vendor's current Go
654		-	// model lineup (https://opencode.ai/docs/go) and were missing from the
655		-	// baked-in catalog.
	652	+	// TestOpenCodeGoCatalog pins entries added from the vendor's current Go
	653	+	// model lineup (https://opencode.ai/docs/go).
656	654		func TestOpenCodeGoCatalog(t *testing.T) {
657	655		cases := []struct {
658	656		id string

@@ -662,6 +660,7 @@ func TestOpenCodeGoCatalog(t *testing.T) {
662	660		priceOut float64
663	661		cacheRead float64
664	662		}{
	663	+	{"gpt-5.6-luna", 1050000, 128000, 0.2, 1.2, 0.02},
665	664		{"grok-4.5", 500000, 500000, 2, 6, 0.3},
666	665		{"kimi-k3", 1048576, 131072, 3, 15, 0.3},
667	666		}

@@ -680,6 +679,17 @@ func TestOpenCodeGoCatalog(t *testing.T) {
680	679		t.Fatalf("opencode-go/%s baseURL: %s", tc.id, m.BaseURL)
681	680		}
682	681		}
	682	+
	683	+	luna, err := FindModel("opencode-go", "gpt-5.6-luna")
	684	+	if err != nil {
	685	+	t.Fatal(err)
	686	+	}
	687	+	if luna.API != APIResponses \|\| luna.PriceCacheWrite != 0.25 {
	688	+	t.Fatalf("opencode-go/gpt-5.6-luna API or cache-write price: %+v", luna)
	689	+	}
	690	+	if luna.PriceTierInputTokens != 272000 \|\| luna.PriceInputAbove != 0.4 \|\| luna.PriceOutputAbove != 1.8 \|\| luna.PriceCacheReadAbove != 0.04 \|\| luna.PriceCacheWriteAbove != 0.5 {
	691	+	t.Fatalf("opencode-go/gpt-5.6-luna long-context prices: %+v", luna)
	692	+	}
683	693		}
684	694
685	695		// TestOpenAIOmitsZeroMaxTokens guards against sending max_tokens: 0 for

 	}
 }
-// TestOpenCodeGoCatalog pins the Grok 4.5 and Kimi K3 entries added to
+// TestOpenCodeGoCatalog pins entries added from the vendor's current Go
-// the OpenCode Go provider. Both are listed in the vendor's current Go
+// model lineup (https://opencode.ai/docs/go).
-// model lineup (https://opencode.ai/docs/go) and were missing from the
-// baked-in catalog.
 func TestOpenCodeGoCatalog(t *testing.T) {
 	cases := []struct {
 		id        string
 		priceOut  float64
 		cacheRead float64
 	}{
+		{"gpt-5.6-luna", 1050000, 128000, 0.2, 1.2, 0.02},
 		{"grok-4.5", 500000, 500000, 2, 6, 0.3},
 		{"kimi-k3", 1048576, 131072, 3, 15, 0.3},
 	}
 			t.Fatalf("opencode-go/%s baseURL: %s", tc.id, m.BaseURL)
 		}
 	}
+	luna, err := FindModel("opencode-go", "gpt-5.6-luna")
+	if err != nil {
+		t.Fatal(err)
+	}
+	if luna.API != APIResponses || luna.PriceCacheWrite != 0.25 {
+		t.Fatalf("opencode-go/gpt-5.6-luna API or cache-write price: %+v", luna)
+	}
+	if luna.PriceTierInputTokens != 272000 || luna.PriceInputAbove != 0.4 || luna.PriceOutputAbove != 1.8 || luna.PriceCacheReadAbove != 0.04 || luna.PriceCacheWriteAbove != 0.5 {
+		t.Fatalf("opencode-go/gpt-5.6-luna long-context prices: %+v", luna)
+	}
 }
 // TestOpenAIOmitsZeroMaxTokens guards against sending max_tokens: 0 for