detached 6c955d6

Add print mode generation stats

commit
6c955d6a61114aa18a4d3b7f89b71715cd5dfa96
author
patriceckhart · · 2026-08-03 05:25 UTC
parent
2501d48
signature
unsigned
refs
tag: v0.3.31
22 files · +363 -30
Unified
UnifiedSplit
M README.md +4 -0
@@ -189,6 +189,7 @@ The first time you launch a newer zot binary, the TUI shows the GitHub release n
189189 zot # interactive tui
190190 zot "fix the failing test" # tui, pre-filled prompt
191191 zot -p "list all go files" # print final text, exit
192+zot -p --stats stats.json "task" # print final text and write generation stats
192193 zot --json "refactor main.go" # newline-delimited json events, exit
193194 zot --continue # resume the most recent session for this cwd
194195 zot --resume # pick a session to resume
@@ -196,6 +197,8 @@ zot --list-models # show supported models
196197 zot --help
197198 ```
198199
200+Print-mode stats contain `provider`, `model`, `prompt_tokens`, `reasoning_tokens`, `generated_output_tokens`, and `elapsed_ms`. Counts cover all model turns triggered by the prompt, including tool loops. Prompt tokens include cache reads and writes. `reasoning_tokens` is `null` when the provider does not report a separate count; in that case `generated_output_tokens` is the provider's total output count and may include reasoning. Elapsed time covers the agent run, not startup and credential resolution. The file is written only after a successful run.
201+
199202 ## Flags
200203
201204 | Flag | Description |
@@ -208,6 +211,7 @@ zot --help
208211 | `--system-prompt <text>` | Replace the default system prompt for this run (also overrides `$ZOT_HOME/SYSTEM.md`). |
209212 | `--append-system-prompt <text>` | Append text to the system prompt (repeatable). |
210213 | `--reasoning off\|minimum\|low\|medium\|high\|xhigh\|max` | Set thinking level on supported models (default: off). `max` is a separate opt-in tier above `xhigh`. |
214+| `--stats <path>` | With `-p`/`--print`, write generation statistics as JSON. |
211215 | `-c`, `--continue` | Resume the latest session for this cwd. |
212216 | `-r`, `--resume` | Pick a session to resume. |
213217 | `--session <path>` | Resume a specific session file. |
@@ -189,6 +189,7 @@ The first time you launch a newer zot binary, the TUI shows the GitHub release n
189 zot # interactive tui189 zot # interactive tui
190 zot "fix the failing test" # tui, pre-filled prompt190 zot "fix the failing test" # tui, pre-filled prompt
191 zot -p "list all go files" # print final text, exit191 zot -p "list all go files" # print final text, exit
192+zot -p --stats stats.json "task" # print final text and write generation stats
192 zot --json "refactor main.go" # newline-delimited json events, exit193 zot --json "refactor main.go" # newline-delimited json events, exit
193 zot --continue # resume the most recent session for this cwd194 zot --continue # resume the most recent session for this cwd
194 zot --resume # pick a session to resume195 zot --resume # pick a session to resume
@@ -196,6 +197,8 @@ zot --list-models # show supported models
196 zot --help197 zot --help
197 ```198 ```
198 199
200+Print-mode stats contain `provider`, `model`, `prompt_tokens`, `reasoning_tokens`, `generated_output_tokens`, and `elapsed_ms`. Counts cover all model turns triggered by the prompt, including tool loops. Prompt tokens include cache reads and writes. `reasoning_tokens` is `null` when the provider does not report a separate count; in that case `generated_output_tokens` is the provider's total output count and may include reasoning. Elapsed time covers the agent run, not startup and credential resolution. The file is written only after a successful run.
201+
199 ## Flags202 ## Flags
200 203
201 | Flag | Description |204 | Flag | Description |
@@ -208,6 +211,7 @@ zot --help
208 | `--system-prompt <text>` | Replace the default system prompt for this run (also overrides `$ZOT_HOME/SYSTEM.md`). |211 | `--system-prompt <text>` | Replace the default system prompt for this run (also overrides `$ZOT_HOME/SYSTEM.md`). |
209 | `--append-system-prompt <text>` | Append text to the system prompt (repeatable). |212 | `--append-system-prompt <text>` | Append text to the system prompt (repeatable). |
210 | `--reasoning off\|minimum\|low\|medium\|high\|xhigh\|max` | Set thinking level on supported models (default: off). `max` is a separate opt-in tier above `xhigh`. |213 | `--reasoning off\|minimum\|low\|medium\|high\|xhigh\|max` | Set thinking level on supported models (default: off). `max` is a separate opt-in tier above `xhigh`. |
214+| `--stats <path>` | With `-p`/`--print`, write generation statistics as JSON. |
211 | `-c`, `--continue` | Resume the latest session for this cwd. |215 | `-c`, `--continue` | Resume the latest session for this cwd. |
212 | `-r`, `--resume` | Pick a session to resume. |216 | `-r`, `--resume` | Pick a session to resume. |
213 | `--session <path>` | Resume a specific session file. |217 | `--session <path>` | Resume a specific session file. |
M docs/rpc.md +2 -2
@@ -121,7 +121,7 @@ Response data:
121121 "cwd": "/Users/pat/Developer/zot",
122122 "message_count": 12,
123123 "busy": false,
124- "usage": {"input": 1234, "output": 567, "cache_read": 890, "cache_write": 0, "cost_usd": 0.0123}
124+ "usage": {"input": 1234, "output": 567, "reasoning": 123, "cache_read": 890, "cache_write": 0, "cost_usd": 0.0123}
125125 }
126126 ```
127127
@@ -197,7 +197,7 @@ Stream notifications during a `prompt` or `compact`. None carry an `id`.
197197 | `tool_progress` | `id`, `text` | Optional progress line from the tool while it runs |
198198 | `tool_result` | `id`, `is_error`, `content` | Tool finished |
199199 | `assistant_message` | `content`, `time` | Final assistant message after the model turn ends |
200-| `usage` | `input`, `output`, `cache_read`, `cache_write`, `cost_usd`, `cumulative` | Per-turn + cumulative tokens / cost |
200+| `usage` | `input`, `output`, `reasoning`, `cache_read`, `cache_write`, `cost_usd`, `cumulative` | Per-turn + cumulative tokens / cost. `reasoning` is null when unavailable. |
201201 | `turn_end` | `stop`, optional `error` | One model call finished. `stop` is `end_turn`, `tool_use`, `length`, `error`, or `aborted` |
202202 | `done` | (none) | The whole prompt/compact completed (success or error) |
203203 | `error` | `message` | Non-fatal error message |
@@ -121,7 +121,7 @@ Response data:
121 "cwd": "/Users/pat/Developer/zot",121 "cwd": "/Users/pat/Developer/zot",
122 "message_count": 12,122 "message_count": 12,
123 "busy": false,123 "busy": false,
124- "usage": {"input": 1234, "output": 567, "cache_read": 890, "cache_write": 0, "cost_usd": 0.0123}124+ "usage": {"input": 1234, "output": 567, "reasoning": 123, "cache_read": 890, "cache_write": 0, "cost_usd": 0.0123}
125 }125 }
126 ```126 ```
127 127
@@ -197,7 +197,7 @@ Stream notifications during a `prompt` or `compact`. None carry an `id`.
197 | `tool_progress` | `id`, `text` | Optional progress line from the tool while it runs |197 | `tool_progress` | `id`, `text` | Optional progress line from the tool while it runs |
198 | `tool_result` | `id`, `is_error`, `content` | Tool finished |198 | `tool_result` | `id`, `is_error`, `content` | Tool finished |
199 | `assistant_message` | `content`, `time` | Final assistant message after the model turn ends |199 | `assistant_message` | `content`, `time` | Final assistant message after the model turn ends |
200-| `usage` | `input`, `output`, `cache_read`, `cache_write`, `cost_usd`, `cumulative` | Per-turn + cumulative tokens / cost |200+| `usage` | `input`, `output`, `reasoning`, `cache_read`, `cache_write`, `cost_usd`, `cumulative` | Per-turn + cumulative tokens / cost. `reasoning` is null when unavailable. |
201 | `turn_end` | `stop`, optional `error` | One model call finished. `stop` is `end_turn`, `tool_use`, `length`, `error`, or `aborted` |201 | `turn_end` | `stop`, optional `error` | One model call finished. `stop` is `end_turn`, `tool_use`, `length`, `error`, or `aborted` |
202 | `done` | (none) | The whole prompt/compact completed (success or error) |202 | `done` | (none) | The whole prompt/compact completed (success or error) |
203 | `error` | `message` | Non-fatal error message |203 | `error` | `message` | Non-fatal error message |
M packages/agent/args.go +11 -0
@@ -106,6 +106,7 @@ type Args struct {
106106 ListModels bool
107107 Help bool
108108 Version bool
109+ StatsPath string // write print-mode generation statistics as JSON
109110
110111 Prompt string // concatenated positional args
111112
@@ -247,6 +248,12 @@ func ParseArgs(in []string) (Args, error) {
247248 }
248249 t := float32(f)
249250 a.Temperature = &t
251+ case "--stats":
252+ v, err := want(&i, arg)
253+ if err != nil {
254+ return a, err
255+ }
256+ a.StatsPath = v
250257 case "--session":
251258 v, err := want(&i, arg)
252259 if err != nil {
@@ -299,6 +306,9 @@ func ParseArgs(in []string) (Args, error) {
299306 a.Prompt = strings.Join(positional, " ")
300307 }
301308
309+ if a.StatsPath != "" && a.Mode != ModePrint {
310+ return a, fmt.Errorf("--stats requires -p or --print")
311+ }
302312 if a.CWD == "" {
303313 a.CWD, _ = os.Getwd()
304314 }
@@ -412,6 +422,7 @@ func PrintHelp(version string) {
412422 row{"--temperature N", "sampling temperature, 0 to 2 (omit for provider default)"},
413423 )
414424 section("prompt and session flags",
425+ row{"--stats PATH", "write print-mode generation stats as json"},
415426 row{"--system-prompt TEXT", "replace the default system prompt"},
416427 row{"--append-system-prompt TEXT", "append to the system prompt (repeatable)"},
417428 row{"-c, --continue", "continue the most recent session for this cwd"},
@@ -106,6 +106,7 @@ type Args struct {
106 ListModels bool106 ListModels bool
107 Help bool107 Help bool
108 Version bool108 Version bool
109+ StatsPath string // write print-mode generation statistics as JSON
109 110
110 Prompt string // concatenated positional args111 Prompt string // concatenated positional args
111 112
@@ -247,6 +248,12 @@ func ParseArgs(in []string) (Args, error) {
247 }248 }
248 t := float32(f)249 t := float32(f)
249 a.Temperature = &t250 a.Temperature = &t
251+ case "--stats":
252+ v, err := want(&i, arg)
253+ if err != nil {
254+ return a, err
255+ }
256+ a.StatsPath = v
250 case "--session":257 case "--session":
251 v, err := want(&i, arg)258 v, err := want(&i, arg)
252 if err != nil {259 if err != nil {
@@ -299,6 +306,9 @@ func ParseArgs(in []string) (Args, error) {
299 a.Prompt = strings.Join(positional, " ")306 a.Prompt = strings.Join(positional, " ")
300 }307 }
301 308
309+ if a.StatsPath != "" && a.Mode != ModePrint {
310+ return a, fmt.Errorf("--stats requires -p or --print")
311+ }
302 if a.CWD == "" {312 if a.CWD == "" {
303 a.CWD, _ = os.Getwd()313 a.CWD, _ = os.Getwd()
304 }314 }
@@ -412,6 +422,7 @@ func PrintHelp(version string) {
412 row{"--temperature N", "sampling temperature, 0 to 2 (omit for provider default)"},422 row{"--temperature N", "sampling temperature, 0 to 2 (omit for provider default)"},
413 )423 )
414 section("prompt and session flags",424 section("prompt and session flags",
425+ row{"--stats PATH", "write print-mode generation stats as json"},
415 row{"--system-prompt TEXT", "replace the default system prompt"},426 row{"--system-prompt TEXT", "replace the default system prompt"},
416 row{"--append-system-prompt TEXT", "append to the system prompt (repeatable)"},427 row{"--append-system-prompt TEXT", "append to the system prompt (repeatable)"},
417 row{"-c, --continue", "continue the most recent session for this cwd"},428 row{"-c, --continue", "continue the most recent session for this cwd"},
M packages/agent/args_test.go +14 -0
@@ -33,6 +33,20 @@ func TestParseArgsYes(t *testing.T) {
3333 }
3434 }
3535
36+func TestParseArgsStatsRequiresPrintMode(t *testing.T) {
37+ args, err := ParseArgs([]string{"-p", "--stats", "stats.json", "hi"})
38+ if err != nil {
39+ t.Fatal(err)
40+ }
41+ if args.StatsPath != "stats.json" || args.Mode != ModePrint {
42+ t.Fatalf("StatsPath=%q Mode=%q", args.StatsPath, args.Mode)
43+ }
44+
45+ if _, err := ParseArgs([]string{"--stats", "stats.json", "hi"}); err == nil {
46+ t.Fatal("ParseArgs accepted --stats without print mode")
47+ }
48+}
49+
3650 func TestParseArgsStream(t *testing.T) {
3751 args, err := ParseArgs([]string{"--stream", "hi"})
3852 if err != nil {
@@ -33,6 +33,20 @@ func TestParseArgsYes(t *testing.T) {
33 }33 }
34 }34 }
35 35
36+func TestParseArgsStatsRequiresPrintMode(t *testing.T) {
37+ args, err := ParseArgs([]string{"-p", "--stats", "stats.json", "hi"})
38+ if err != nil {
39+ t.Fatal(err)
40+ }
41+ if args.StatsPath != "stats.json" || args.Mode != ModePrint {
42+ t.Fatalf("StatsPath=%q Mode=%q", args.StatsPath, args.Mode)
43+ }
44+
45+ if _, err := ParseArgs([]string{"--stats", "stats.json", "hi"}); err == nil {
46+ t.Fatal("ParseArgs accepted --stats without print mode")
47+ }
48+}
49+
36 func TestParseArgsStream(t *testing.T) {50 func TestParseArgsStream(t *testing.T) {
37 args, err := ParseArgs([]string{"--stream", "hi"})51 args, err := ParseArgs([]string{"--stream", "hi"})
38 if err != nil {52 if err != nil {
M packages/agent/cli.go +49 -2
@@ -330,6 +330,45 @@ func wireNonInteractiveAgentExtHooks(ctx context.Context, ag *core.Agent, extMgr
330330 ag.OnEvent = func(ev core.AgentEvent) { fanoutAgentEvent(extMgr, ev) }
331331 }
332332
333+type printStats struct {
334+ Provider string `json:"provider"`
335+ Model string `json:"model"`
336+ PromptTokens int `json:"prompt_tokens"`
337+ ReasoningTokens *int `json:"reasoning_tokens"`
338+ GeneratedOutputTokens int `json:"generated_output_tokens"`
339+ ElapsedMS int64 `json:"elapsed_ms"`
340+}
341+
342+func writePrintStats(path, providerID, model string, usage provider.Usage, elapsed time.Duration) error {
343+ var reasoning *int
344+ generated := usage.OutputTokens
345+ if usage.ReasoningTokensKnown {
346+ count := usage.ReasoningTokens
347+ reasoning = &count
348+ generated -= count
349+ if generated < 0 {
350+ generated = 0
351+ }
352+ }
353+ stats := printStats{
354+ Provider: providerID,
355+ Model: model,
356+ PromptTokens: usage.InputTokens + usage.CacheReadTokens + usage.CacheWriteTokens,
357+ ReasoningTokens: reasoning,
358+ GeneratedOutputTokens: generated,
359+ ElapsedMS: elapsed.Milliseconds(),
360+ }
361+ data, err := json.MarshalIndent(stats, "", " ")
362+ if err != nil {
363+ return fmt.Errorf("encode stats: %w", err)
364+ }
365+ data = append(data, '\n')
366+ if err := os.WriteFile(path, data, 0o644); err != nil {
367+ return fmt.Errorf("write stats %q: %w", path, err)
368+ }
369+ return nil
370+}
371+
333372 func runPrintMode(ctx context.Context, args Args, version string) error {
334373 if args.NoYolo {
335374 fmt.Fprintln(os.Stderr, "warning: --no-yolo has no effect in print mode (no interactive prompt available); tools will run without confirmation")
@@ -360,9 +399,17 @@ func runPrintMode(ctx context.Context, args Args, version string) error {
360399 return err
361400 }
362401 reloadResourcesAfterStartupPre(ctx, args, extMgr, r.Sandbox, ag)
363- err = modes.RunPrint(ctx, ag, prompt, nil, os.Stdout)
402+ started := time.Now()
403+ usage, err := modes.RunPrint(ctx, ag, prompt, nil, os.Stdout)
404+ elapsed := time.Since(started)
364405 WriteNewTranscript(ag, sess, start)
365- return err
406+ if err != nil {
407+ return err
408+ }
409+ if args.StatsPath != "" {
410+ return writePrintStats(args.StatsPath, r.Provider, r.Model, usage, elapsed)
411+ }
412+ return nil
366413 }
367414
368415 func runStreamMode(ctx context.Context, args Args, version string) error {
@@ -330,6 +330,45 @@ func wireNonInteractiveAgentExtHooks(ctx context.Context, ag *core.Agent, extMgr
330 ag.OnEvent = func(ev core.AgentEvent) { fanoutAgentEvent(extMgr, ev) }330 ag.OnEvent = func(ev core.AgentEvent) { fanoutAgentEvent(extMgr, ev) }
331 }331 }
332 332
333+type printStats struct {
334+ Provider string `json:"provider"`
335+ Model string `json:"model"`
336+ PromptTokens int `json:"prompt_tokens"`
337+ ReasoningTokens *int `json:"reasoning_tokens"`
338+ GeneratedOutputTokens int `json:"generated_output_tokens"`
339+ ElapsedMS int64 `json:"elapsed_ms"`
340+}
341+
342+func writePrintStats(path, providerID, model string, usage provider.Usage, elapsed time.Duration) error {
343+ var reasoning *int
344+ generated := usage.OutputTokens
345+ if usage.ReasoningTokensKnown {
346+ count := usage.ReasoningTokens
347+ reasoning = &count
348+ generated -= count
349+ if generated < 0 {
350+ generated = 0
351+ }
352+ }
353+ stats := printStats{
354+ Provider: providerID,
355+ Model: model,
356+ PromptTokens: usage.InputTokens + usage.CacheReadTokens + usage.CacheWriteTokens,
357+ ReasoningTokens: reasoning,
358+ GeneratedOutputTokens: generated,
359+ ElapsedMS: elapsed.Milliseconds(),
360+ }
361+ data, err := json.MarshalIndent(stats, "", " ")
362+ if err != nil {
363+ return fmt.Errorf("encode stats: %w", err)
364+ }
365+ data = append(data, '\n')
366+ if err := os.WriteFile(path, data, 0o644); err != nil {
367+ return fmt.Errorf("write stats %q: %w", path, err)
368+ }
369+ return nil
370+}
371+
333 func runPrintMode(ctx context.Context, args Args, version string) error {372 func runPrintMode(ctx context.Context, args Args, version string) error {
334 if args.NoYolo {373 if args.NoYolo {
335 fmt.Fprintln(os.Stderr, "warning: --no-yolo has no effect in print mode (no interactive prompt available); tools will run without confirmation")374 fmt.Fprintln(os.Stderr, "warning: --no-yolo has no effect in print mode (no interactive prompt available); tools will run without confirmation")
@@ -360,9 +399,17 @@ func runPrintMode(ctx context.Context, args Args, version string) error {
360 return err399 return err
361 }400 }
362 reloadResourcesAfterStartupPre(ctx, args, extMgr, r.Sandbox, ag)401 reloadResourcesAfterStartupPre(ctx, args, extMgr, r.Sandbox, ag)
363- err = modes.RunPrint(ctx, ag, prompt, nil, os.Stdout)402+ started := time.Now()
403+ usage, err := modes.RunPrint(ctx, ag, prompt, nil, os.Stdout)
404+ elapsed := time.Since(started)
364 WriteNewTranscript(ag, sess, start)405 WriteNewTranscript(ag, sess, start)
365- return err406+ if err != nil {
407+ return err
408+ }
409+ if args.StatsPath != "" {
410+ return writePrintStats(args.StatsPath, r.Provider, r.Model, usage, elapsed)
411+ }
412+ return nil
366 }413 }
367 414
368 func runStreamMode(ctx context.Context, args Args, version string) error {415 func runStreamMode(ctx context.Context, args Args, version string) error {
M packages/agent/modes/json.go +9 -0
@@ -73,12 +73,14 @@ func EventToJSON(ev core.AgentEvent) map[string]any {
7373 case core.EvUsage:
7474 m["input"] = e.Usage.InputTokens
7575 m["output"] = e.Usage.OutputTokens
76+ m["reasoning"] = nullableReasoningTokens(e.Usage)
7677 m["cache_read"] = e.Usage.CacheReadTokens
7778 m["cache_write"] = e.Usage.CacheWriteTokens
7879 m["cost_usd"] = e.Usage.CostUSD
7980 m["cumulative"] = map[string]any{
8081 "input": e.Cumulative.InputTokens,
8182 "output": e.Cumulative.OutputTokens,
83+ "reasoning": nullableReasoningTokens(e.Cumulative),
8284 "cache_read": e.Cumulative.CacheReadTokens,
8385 "cache_write": e.Cumulative.CacheWriteTokens,
8486 "cost_usd": e.Cumulative.CostUSD,
@@ -119,6 +121,13 @@ func ContentToJSON(blocks []provider.Content) []map[string]any {
119121 return out
120122 }
121123
124+func nullableReasoningTokens(usage provider.Usage) any {
125+ if !usage.ReasoningTokensKnown {
126+ return nil
127+ }
128+ return usage.ReasoningTokens
129+}
130+
122131 func jsonString(s string) string {
123132 b, _ := json.Marshal(s)
124133 return string(b)
@@ -73,12 +73,14 @@ func EventToJSON(ev core.AgentEvent) map[string]any {
73 case core.EvUsage:73 case core.EvUsage:
74 m["input"] = e.Usage.InputTokens74 m["input"] = e.Usage.InputTokens
75 m["output"] = e.Usage.OutputTokens75 m["output"] = e.Usage.OutputTokens
76+ m["reasoning"] = nullableReasoningTokens(e.Usage)
76 m["cache_read"] = e.Usage.CacheReadTokens77 m["cache_read"] = e.Usage.CacheReadTokens
77 m["cache_write"] = e.Usage.CacheWriteTokens78 m["cache_write"] = e.Usage.CacheWriteTokens
78 m["cost_usd"] = e.Usage.CostUSD79 m["cost_usd"] = e.Usage.CostUSD
79 m["cumulative"] = map[string]any{80 m["cumulative"] = map[string]any{
80 "input": e.Cumulative.InputTokens,81 "input": e.Cumulative.InputTokens,
81 "output": e.Cumulative.OutputTokens,82 "output": e.Cumulative.OutputTokens,
83+ "reasoning": nullableReasoningTokens(e.Cumulative),
82 "cache_read": e.Cumulative.CacheReadTokens,84 "cache_read": e.Cumulative.CacheReadTokens,
83 "cache_write": e.Cumulative.CacheWriteTokens,85 "cache_write": e.Cumulative.CacheWriteTokens,
84 "cost_usd": e.Cumulative.CostUSD,86 "cost_usd": e.Cumulative.CostUSD,
@@ -119,6 +121,13 @@ func ContentToJSON(blocks []provider.Content) []map[string]any {
119 return out121 return out
120 }122 }
121 123
124+func nullableReasoningTokens(usage provider.Usage) any {
125+ if !usage.ReasoningTokensKnown {
126+ return nil
127+ }
128+ return usage.ReasoningTokens
129+}
130+
122 func jsonString(s string) string {131 func jsonString(s string) string {
123 b, _ := json.Marshal(s)132 b, _ := json.Marshal(s)
124 return string(b)133 return string(b)
M packages/agent/modes/print.go +15 -5
@@ -12,10 +12,13 @@ import (
1212 )
1313
1414 // RunPrint runs the agent to completion and writes only the final
15-// assistant text block to out. Exit code comes from the caller.
16-func RunPrint(ctx context.Context, ag *core.Agent, prompt string, images []provider.ImageBlock, out io.Writer) error {
15+// assistant text block to out. It returns usage for this invocation,
16+// excluding any cumulative usage restored from a session.
17+func RunPrint(ctx context.Context, ag *core.Agent, prompt string, images []provider.ImageBlock, out io.Writer) (provider.Usage, error) {
1718 var finalText strings.Builder
1819 var lastAssistant string
20+ var usage provider.Usage
21+ var haveUsage bool
1922 var runErr error
2023
2124 sink := func(ev core.AgentEvent) {
@@ -34,6 +37,13 @@ func RunPrint(ctx context.Context, ag *core.Agent, prompt string, images []provi
3437 if sb.Len() > 0 {
3538 lastAssistant = sb.String()
3639 }
40+ case core.EvUsage:
41+ if haveUsage {
42+ usage = usage.Add(e.Usage)
43+ } else {
44+ usage = e.Usage
45+ haveUsage = true
46+ }
3747 case core.EvTurnEnd:
3848 if e.Err != nil {
3949 runErr = e.Err
@@ -42,10 +52,10 @@ func RunPrint(ctx context.Context, ag *core.Agent, prompt string, images []provi
4252 }
4353
4454 if err := ag.Prompt(ctx, prompt, images, sink); err != nil {
45- return err
55+ return usage, err
4656 }
4757 if runErr != nil {
48- return runErr
58+ return usage, runErr
4959 }
5060
5161 finalText.WriteString(lastAssistant)
@@ -53,5 +63,5 @@ func RunPrint(ctx context.Context, ag *core.Agent, prompt string, images []provi
5363 finalText.WriteString("\n")
5464 }
5565 _, err := fmt.Fprint(out, finalText.String())
56- return err
66+ return usage, err
5767 }
@@ -12,10 +12,13 @@ import (
12 )12 )
13 13
14 // RunPrint runs the agent to completion and writes only the final14 // RunPrint runs the agent to completion and writes only the final
15-// assistant text block to out. Exit code comes from the caller.15+// assistant text block to out. It returns usage for this invocation,
16-func RunPrint(ctx context.Context, ag *core.Agent, prompt string, images []provider.ImageBlock, out io.Writer) error {16+// excluding any cumulative usage restored from a session.
17+func RunPrint(ctx context.Context, ag *core.Agent, prompt string, images []provider.ImageBlock, out io.Writer) (provider.Usage, error) {
17 var finalText strings.Builder18 var finalText strings.Builder
18 var lastAssistant string19 var lastAssistant string
20+ var usage provider.Usage
21+ var haveUsage bool
19 var runErr error22 var runErr error
20 23
21 sink := func(ev core.AgentEvent) {24 sink := func(ev core.AgentEvent) {
@@ -34,6 +37,13 @@ func RunPrint(ctx context.Context, ag *core.Agent, prompt string, images []provi
34 if sb.Len() > 0 {37 if sb.Len() > 0 {
35 lastAssistant = sb.String()38 lastAssistant = sb.String()
36 }39 }
40+ case core.EvUsage:
41+ if haveUsage {
42+ usage = usage.Add(e.Usage)
43+ } else {
44+ usage = e.Usage
45+ haveUsage = true
46+ }
37 case core.EvTurnEnd:47 case core.EvTurnEnd:
38 if e.Err != nil {48 if e.Err != nil {
39 runErr = e.Err49 runErr = e.Err
@@ -42,10 +52,10 @@ func RunPrint(ctx context.Context, ag *core.Agent, prompt string, images []provi
42 }52 }
43 53
44 if err := ag.Prompt(ctx, prompt, images, sink); err != nil {54 if err := ag.Prompt(ctx, prompt, images, sink); err != nil {
45- return err55+ return usage, err
46 }56 }
47 if runErr != nil {57 if runErr != nil {
48- return runErr58+ return usage, runErr
49 }59 }
50 60
51 finalText.WriteString(lastAssistant)61 finalText.WriteString(lastAssistant)
@@ -53,5 +63,5 @@ func RunPrint(ctx context.Context, ag *core.Agent, prompt string, images []provi
53 finalText.WriteString("\n")63 finalText.WriteString("\n")
54 }64 }
55 _, err := fmt.Fprint(out, finalText.String())65 _, err := fmt.Fprint(out, finalText.String())
56- return err66+ return usage, err
57 }67 }
A packages/agent/modes/print_test.go +50 -0
@@ -0,0 +1,50 @@
1+package modes
2+
3+import (
4+ "bytes"
5+ "context"
6+ "testing"
7+
8+ "github.com/patriceckhart/zot/packages/core"
9+ "github.com/patriceckhart/zot/packages/provider"
10+)
11+
12+type printTestClient struct{}
13+
14+func (printTestClient) Name() string { return "test" }
15+
16+func (printTestClient) Stream(context.Context, provider.Request) (<-chan provider.Event, error) {
17+ events := make(chan provider.Event, 3)
18+ events <- provider.EventStart{Model: "test-model", Provider: "test"}
19+ events <- provider.EventUsage{Usage: provider.Usage{
20+ InputTokens: 12,
21+ OutputTokens: 9,
22+ ReasoningTokens: 6,
23+ ReasoningTokensKnown: true,
24+ }}
25+ events <- provider.EventDone{
26+ Stop: provider.StopEnd,
27+ Message: provider.Message{Role: provider.RoleAssistant, Content: []provider.Content{
28+ provider.TextBlock{Text: "answer"},
29+ }},
30+ }
31+ close(events)
32+ return events, nil
33+}
34+
35+func TestRunPrintReturnsInvocationUsage(t *testing.T) {
36+ agent := core.NewAgent(printTestClient{}, "test-model", "", core.Registry{})
37+ agent.SeedCost(provider.Usage{InputTokens: 100, OutputTokens: 50})
38+ var output bytes.Buffer
39+
40+ usage, err := RunPrint(context.Background(), agent, "question", nil, &output)
41+ if err != nil {
42+ t.Fatal(err)
43+ }
44+ if output.String() != "answer\n" {
45+ t.Fatalf("output = %q", output.String())
46+ }
47+ if usage.InputTokens != 12 || usage.OutputTokens != 9 || usage.ReasoningTokens != 6 || !usage.ReasoningTokensKnown {
48+ t.Fatalf("usage = %+v", usage)
49+ }
50+}
@@ -0,0 +1,50 @@
1+package modes
2+
3+import (
4+ "bytes"
5+ "context"
6+ "testing"
7+
8+ "github.com/patriceckhart/zot/packages/core"
9+ "github.com/patriceckhart/zot/packages/provider"
10+)
11+
12+type printTestClient struct{}
13+
14+func (printTestClient) Name() string { return "test" }
15+
16+func (printTestClient) Stream(context.Context, provider.Request) (<-chan provider.Event, error) {
17+ events := make(chan provider.Event, 3)
18+ events <- provider.EventStart{Model: "test-model", Provider: "test"}
19+ events <- provider.EventUsage{Usage: provider.Usage{
20+ InputTokens: 12,
21+ OutputTokens: 9,
22+ ReasoningTokens: 6,
23+ ReasoningTokensKnown: true,
24+ }}
25+ events <- provider.EventDone{
26+ Stop: provider.StopEnd,
27+ Message: provider.Message{Role: provider.RoleAssistant, Content: []provider.Content{
28+ provider.TextBlock{Text: "answer"},
29+ }},
30+ }
31+ close(events)
32+ return events, nil
33+}
34+
35+func TestRunPrintReturnsInvocationUsage(t *testing.T) {
36+ agent := core.NewAgent(printTestClient{}, "test-model", "", core.Registry{})
37+ agent.SeedCost(provider.Usage{InputTokens: 100, OutputTokens: 50})
38+ var output bytes.Buffer
39+
40+ usage, err := RunPrint(context.Background(), agent, "question", nil, &output)
41+ if err != nil {
42+ t.Fatal(err)
43+ }
44+ if output.String() != "answer\n" {
45+ t.Fatalf("output = %q", output.String())
46+ }
47+ if usage.InputTokens != 12 || usage.OutputTokens != 9 || usage.ReasoningTokens != 6 || !usage.ReasoningTokensKnown {
48+ t.Fatalf("usage = %+v", usage)
49+ }
50+}
A packages/agent/print_stats_test.go +67 -0
@@ -0,0 +1,67 @@
1+package agent
2+
3+import (
4+ "encoding/json"
5+ "os"
6+ "path/filepath"
7+ "testing"
8+ "time"
9+
10+ "github.com/patriceckhart/zot/packages/provider"
11+)
12+
13+func TestWritePrintStats(t *testing.T) {
14+ path := filepath.Join(t.TempDir(), "stats.json")
15+ usage := provider.Usage{
16+ InputTokens: 100,
17+ OutputTokens: 40,
18+ ReasoningTokens: 15,
19+ ReasoningTokensKnown: true,
20+ CacheReadTokens: 20,
21+ CacheWriteTokens: 5,
22+ }
23+ if err := writePrintStats(path, "openai", "gpt-test", usage, 1500*time.Millisecond); err != nil {
24+ t.Fatal(err)
25+ }
26+
27+ data, err := os.ReadFile(path)
28+ if err != nil {
29+ t.Fatal(err)
30+ }
31+ var got printStats
32+ if err := json.Unmarshal(data, &got); err != nil {
33+ t.Fatal(err)
34+ }
35+ if got.Provider != "openai" || got.Model != "gpt-test" {
36+ t.Fatalf("provider/model = %q/%q", got.Provider, got.Model)
37+ }
38+ if got.PromptTokens != 125 || got.ReasoningTokens == nil || *got.ReasoningTokens != 15 {
39+ t.Fatalf("token stats = %+v", got)
40+ }
41+ if got.GeneratedOutputTokens != 25 || got.ElapsedMS != 1500 {
42+ t.Fatalf("output/time stats = %+v", got)
43+ }
44+}
45+
46+func TestWritePrintStatsUsesNullForUnreportedReasoning(t *testing.T) {
47+ path := filepath.Join(t.TempDir(), "stats.json")
48+ usage := provider.Usage{InputTokens: 10, OutputTokens: 7}
49+ if err := writePrintStats(path, "anthropic", "claude-test", usage, time.Second); err != nil {
50+ t.Fatal(err)
51+ }
52+
53+ data, err := os.ReadFile(path)
54+ if err != nil {
55+ t.Fatal(err)
56+ }
57+ var got map[string]any
58+ if err := json.Unmarshal(data, &got); err != nil {
59+ t.Fatal(err)
60+ }
61+ if got["reasoning_tokens"] != nil {
62+ t.Fatalf("reasoning_tokens = %#v, want null", got["reasoning_tokens"])
63+ }
64+ if got["generated_output_tokens"] != float64(7) {
65+ t.Fatalf("generated_output_tokens = %#v", got["generated_output_tokens"])
66+ }
67+}
@@ -0,0 +1,67 @@
1+package agent
2+
3+import (
4+ "encoding/json"
5+ "os"
6+ "path/filepath"
7+ "testing"
8+ "time"
9+
10+ "github.com/patriceckhart/zot/packages/provider"
11+)
12+
13+func TestWritePrintStats(t *testing.T) {
14+ path := filepath.Join(t.TempDir(), "stats.json")
15+ usage := provider.Usage{
16+ InputTokens: 100,
17+ OutputTokens: 40,
18+ ReasoningTokens: 15,
19+ ReasoningTokensKnown: true,
20+ CacheReadTokens: 20,
21+ CacheWriteTokens: 5,
22+ }
23+ if err := writePrintStats(path, "openai", "gpt-test", usage, 1500*time.Millisecond); err != nil {
24+ t.Fatal(err)
25+ }
26+
27+ data, err := os.ReadFile(path)
28+ if err != nil {
29+ t.Fatal(err)
30+ }
31+ var got printStats
32+ if err := json.Unmarshal(data, &got); err != nil {
33+ t.Fatal(err)
34+ }
35+ if got.Provider != "openai" || got.Model != "gpt-test" {
36+ t.Fatalf("provider/model = %q/%q", got.Provider, got.Model)
37+ }
38+ if got.PromptTokens != 125 || got.ReasoningTokens == nil || *got.ReasoningTokens != 15 {
39+ t.Fatalf("token stats = %+v", got)
40+ }
41+ if got.GeneratedOutputTokens != 25 || got.ElapsedMS != 1500 {
42+ t.Fatalf("output/time stats = %+v", got)
43+ }
44+}
45+
46+func TestWritePrintStatsUsesNullForUnreportedReasoning(t *testing.T) {
47+ path := filepath.Join(t.TempDir(), "stats.json")
48+ usage := provider.Usage{InputTokens: 10, OutputTokens: 7}
49+ if err := writePrintStats(path, "anthropic", "claude-test", usage, time.Second); err != nil {
50+ t.Fatal(err)
51+ }
52+
53+ data, err := os.ReadFile(path)
54+ if err != nil {
55+ t.Fatal(err)
56+ }
57+ var got map[string]any
58+ if err := json.Unmarshal(data, &got); err != nil {
59+ t.Fatal(err)
60+ }
61+ if got["reasoning_tokens"] != nil {
62+ t.Fatalf("reasoning_tokens = %#v, want null", got["reasoning_tokens"])
63+ }
64+ if got["generated_output_tokens"] != float64(7) {
65+ t.Fatalf("generated_output_tokens = %#v", got["generated_output_tokens"])
66+ }
67+}
M packages/agent/rpc.go +8 -0
@@ -414,6 +414,7 @@ func (s *rpcServer) snapshotState() map[string]any {
414414 "usage": map[string]any{
415415 "input": cum.InputTokens,
416416 "output": cum.OutputTokens,
417+ "reasoning": rpcReasoningTokens(cum),
417418 "cache_read": cum.CacheReadTokens,
418419 "cache_write": cum.CacheWriteTokens,
419420 "cost_usd": cum.CostUSD,
@@ -421,6 +422,13 @@ func (s *rpcServer) snapshotState() map[string]any {
421422 }
422423 }
423424
425+func rpcReasoningTokens(usage provider.Usage) any {
426+ if !usage.ReasoningTokensKnown {
427+ return nil
428+ }
429+ return usage.ReasoningTokens
430+}
431+
424432 // ---- write helpers (single-line JSON, mutex-guarded) ----
425433
426434 func (s *rpcServer) writeResponse(id, cmd string, data any) {
@@ -414,6 +414,7 @@ func (s *rpcServer) snapshotState() map[string]any {
414 "usage": map[string]any{414 "usage": map[string]any{
415 "input": cum.InputTokens,415 "input": cum.InputTokens,
416 "output": cum.OutputTokens,416 "output": cum.OutputTokens,
417+ "reasoning": rpcReasoningTokens(cum),
417 "cache_read": cum.CacheReadTokens,418 "cache_read": cum.CacheReadTokens,
418 "cache_write": cum.CacheWriteTokens,419 "cache_write": cum.CacheWriteTokens,
419 "cost_usd": cum.CostUSD,420 "cost_usd": cum.CostUSD,
@@ -421,6 +422,13 @@ func (s *rpcServer) snapshotState() map[string]any {
421 }422 }
422 }423 }
423 424
425+func rpcReasoningTokens(usage provider.Usage) any {
426+ if !usage.ReasoningTokensKnown {
427+ return nil
428+ }
429+ return usage.ReasoningTokens
430+}
431+
424 // ---- write helpers (single-line JSON, mutex-guarded) ----432 // ---- write helpers (single-line JSON, mutex-guarded) ----
425 433
426 func (s *rpcServer) writeResponse(id, cmd string, data any) {434 func (s *rpcServer) writeResponse(id, cmd string, data any) {
M packages/agent/sdk/sdk.go +1 -0
@@ -191,6 +191,7 @@ func (r *Runtime) Cost() Usage {
191191 return Usage{
192192 Input: u.InputTokens,
193193 Output: u.OutputTokens,
194+ Reasoning: sdkReasoningTokens(u),
194195 CacheRead: u.CacheReadTokens,
195196 CacheWrite: u.CacheWriteTokens,
196197 CostUSD: u.CostUSD,
@@ -191,6 +191,7 @@ func (r *Runtime) Cost() Usage {
191 return Usage{191 return Usage{
192 Input: u.InputTokens,192 Input: u.InputTokens,
193 Output: u.OutputTokens,193 Output: u.OutputTokens,
194+ Reasoning: sdkReasoningTokens(u),
194 CacheRead: u.CacheReadTokens,195 CacheRead: u.CacheReadTokens,
195 CacheWrite: u.CacheWriteTokens,196 CacheWrite: u.CacheWriteTokens,
196 CostUSD: u.CostUSD,197 CostUSD: u.CostUSD,
M packages/agent/sdk/types.go +11 -0
@@ -85,6 +85,7 @@ type Image struct {
8585 type Usage struct {
8686 Input int `json:"input"`
8787 Output int `json:"output"`
88+ Reasoning *int `json:"reasoning"`
8889 CacheRead int `json:"cache_read"`
8990 CacheWrite int `json:"cache_write"`
9091 CostUSD float64 `json:"cost_usd"`
@@ -117,6 +118,14 @@ type ModelInfo struct {
117118
118119 // ---- internal converters ----
119120
121+func sdkReasoningTokens(usage provider.Usage) *int {
122+ if !usage.ReasoningTokensKnown {
123+ return nil
124+ }
125+ count := usage.ReasoningTokens
126+ return &count
127+}
128+
120129 func toEvent(ev core.AgentEvent) Event {
121130 out := Event{Type: ev.Type()}
122131 switch e := ev.(type) {
@@ -143,6 +152,7 @@ func toEvent(ev core.AgentEvent) Event {
143152 out.Usage = &Usage{
144153 Input: e.Usage.InputTokens,
145154 Output: e.Usage.OutputTokens,
155+ Reasoning: sdkReasoningTokens(e.Usage),
146156 CacheRead: e.Usage.CacheReadTokens,
147157 CacheWrite: e.Usage.CacheWriteTokens,
148158 CostUSD: e.Usage.CostUSD,
@@ -150,6 +160,7 @@ func toEvent(ev core.AgentEvent) Event {
150160 out.Cumulative = &Usage{
151161 Input: e.Cumulative.InputTokens,
152162 Output: e.Cumulative.OutputTokens,
163+ Reasoning: sdkReasoningTokens(e.Cumulative),
153164 CacheRead: e.Cumulative.CacheReadTokens,
154165 CacheWrite: e.Cumulative.CacheWriteTokens,
155166 CostUSD: e.Cumulative.CostUSD,
@@ -85,6 +85,7 @@ type Image struct {
85 type Usage struct {85 type Usage struct {
86 Input int `json:"input"`86 Input int `json:"input"`
87 Output int `json:"output"`87 Output int `json:"output"`
88+ Reasoning *int `json:"reasoning"`
88 CacheRead int `json:"cache_read"`89 CacheRead int `json:"cache_read"`
89 CacheWrite int `json:"cache_write"`90 CacheWrite int `json:"cache_write"`
90 CostUSD float64 `json:"cost_usd"`91 CostUSD float64 `json:"cost_usd"`
@@ -117,6 +118,14 @@ type ModelInfo struct {
117 118
118 // ---- internal converters ----119 // ---- internal converters ----
119 120
121+func sdkReasoningTokens(usage provider.Usage) *int {
122+ if !usage.ReasoningTokensKnown {
123+ return nil
124+ }
125+ count := usage.ReasoningTokens
126+ return &count
127+}
128+
120 func toEvent(ev core.AgentEvent) Event {129 func toEvent(ev core.AgentEvent) Event {
121 out := Event{Type: ev.Type()}130 out := Event{Type: ev.Type()}
122 switch e := ev.(type) {131 switch e := ev.(type) {
@@ -143,6 +152,7 @@ func toEvent(ev core.AgentEvent) Event {
143 out.Usage = &Usage{152 out.Usage = &Usage{
144 Input: e.Usage.InputTokens,153 Input: e.Usage.InputTokens,
145 Output: e.Usage.OutputTokens,154 Output: e.Usage.OutputTokens,
155+ Reasoning: sdkReasoningTokens(e.Usage),
146 CacheRead: e.Usage.CacheReadTokens,156 CacheRead: e.Usage.CacheReadTokens,
147 CacheWrite: e.Usage.CacheWriteTokens,157 CacheWrite: e.Usage.CacheWriteTokens,
148 CostUSD: e.Usage.CostUSD,158 CostUSD: e.Usage.CostUSD,
@@ -150,6 +160,7 @@ func toEvent(ev core.AgentEvent) Event {
150 out.Cumulative = &Usage{160 out.Cumulative = &Usage{
151 Input: e.Cumulative.InputTokens,161 Input: e.Cumulative.InputTokens,
152 Output: e.Cumulative.OutputTokens,162 Output: e.Cumulative.OutputTokens,
163+ Reasoning: sdkReasoningTokens(e.Cumulative),
153 CacheRead: e.Cumulative.CacheReadTokens,164 CacheRead: e.Cumulative.CacheReadTokens,
154 CacheWrite: e.Cumulative.CacheWriteTokens,165 CacheWrite: e.Cumulative.CacheWriteTokens,
155 CostUSD: e.Cumulative.CostUSD,166 CostUSD: e.Cumulative.CostUSD,
M packages/core/agent.go +1 -1
@@ -268,7 +268,7 @@ func (a *Agent) Cost() provider.Usage {
268268 func (a *Agent) SeedCost(u provider.Usage) {
269269 a.mu.Lock()
270270 defer a.mu.Unlock()
271- a.cost.Total = u
271+ a.cost.Seed(u)
272272 }
273273
274274 // LastTurnUsage returns the per-turn usage of the most recent
@@ -268,7 +268,7 @@ func (a *Agent) Cost() provider.Usage {
268 func (a *Agent) SeedCost(u provider.Usage) {268 func (a *Agent) SeedCost(u provider.Usage) {
269 a.mu.Lock()269 a.mu.Lock()
270 defer a.mu.Unlock()270 defer a.mu.Unlock()
271- a.cost.Total = u271+ a.cost.Seed(u)
272 }272 }
273 273
274 // LastTurnUsage returns the per-turn usage of the most recent274 // LastTurnUsage returns the per-turn usage of the most recent
M packages/core/core_test.go +5 -2
@@ -154,11 +154,14 @@ func TestNewSessionAtPathCreatesAtExplicitPath(t *testing.T) {
154154
155155 func TestCostAdd(t *testing.T) {
156156 var c CostTracker
157- c.Add(provider.Usage{InputTokens: 100, OutputTokens: 50, CostUSD: 0.01})
158- c.Add(provider.Usage{InputTokens: 200, OutputTokens: 25, CostUSD: 0.02})
157+ c.Add(provider.Usage{InputTokens: 100, OutputTokens: 50, ReasoningTokens: 20, ReasoningTokensKnown: true, CostUSD: 0.01})
158+ c.Add(provider.Usage{InputTokens: 200, OutputTokens: 25, ReasoningTokens: 10, ReasoningTokensKnown: true, CostUSD: 0.02})
159159 if c.Total.InputTokens != 300 || c.Total.OutputTokens != 75 {
160160 t.Fatalf("got %+v", c.Total)
161161 }
162+ if c.Total.ReasoningTokens != 30 || !c.Total.ReasoningTokensKnown {
163+ t.Fatalf("got reasoning usage %+v", c.Total)
164+ }
162165 if c.Total.CostUSD < 0.0299 || c.Total.CostUSD > 0.0301 {
163166 t.Fatalf("got cost %v", c.Total.CostUSD)
164167 }
@@ -154,11 +154,14 @@ func TestNewSessionAtPathCreatesAtExplicitPath(t *testing.T) {
154 154
155 func TestCostAdd(t *testing.T) {155 func TestCostAdd(t *testing.T) {
156 var c CostTracker156 var c CostTracker
157- c.Add(provider.Usage{InputTokens: 100, OutputTokens: 50, CostUSD: 0.01})157+ c.Add(provider.Usage{InputTokens: 100, OutputTokens: 50, ReasoningTokens: 20, ReasoningTokensKnown: true, CostUSD: 0.01})
158- c.Add(provider.Usage{InputTokens: 200, OutputTokens: 25, CostUSD: 0.02})158+ c.Add(provider.Usage{InputTokens: 200, OutputTokens: 25, ReasoningTokens: 10, ReasoningTokensKnown: true, CostUSD: 0.02})
159 if c.Total.InputTokens != 300 || c.Total.OutputTokens != 75 {159 if c.Total.InputTokens != 300 || c.Total.OutputTokens != 75 {
160 t.Fatalf("got %+v", c.Total)160 t.Fatalf("got %+v", c.Total)
161 }161 }
162+ if c.Total.ReasoningTokens != 30 || !c.Total.ReasoningTokensKnown {
163+ t.Fatalf("got reasoning usage %+v", c.Total)
164+ }
162 if c.Total.CostUSD < 0.0299 || c.Total.CostUSD > 0.0301 {165 if c.Total.CostUSD < 0.0299 || c.Total.CostUSD > 0.0301 {
163 t.Fatalf("got cost %v", c.Total.CostUSD)166 t.Fatalf("got cost %v", c.Total.CostUSD)
164 }167 }
M packages/core/cost.go +13 -1
@@ -12,12 +12,24 @@ import "github.com/patriceckhart/zot/packages/provider"
1212 type CostTracker struct {
1313 Total provider.Usage
1414 LastTurn provider.Usage
15+ hasTotal bool
1516 }
1617
1718 // Add folds u into the running total, records u as the last-turn
1819 // snapshot, and returns the new cumulative value.
1920 func (c *CostTracker) Add(u provider.Usage) provider.Usage {
20- c.Total = c.Total.Add(u)
21+ if c.hasTotal {
22+ c.Total = c.Total.Add(u)
23+ } else {
24+ c.Total = u
25+ c.hasTotal = true
26+ }
2127 c.LastTurn = u
2228 return c.Total
2329 }
30+
31+// Seed replaces the running total with usage restored from a session.
32+func (c *CostTracker) Seed(u provider.Usage) {
33+ c.Total = u
34+ c.hasTotal = u.InputTokens != 0 || u.OutputTokens != 0 || u.CacheReadTokens != 0 || u.CacheWriteTokens != 0 || u.CostUSD != 0
35+}
@@ -12,12 +12,24 @@ import "github.com/patriceckhart/zot/packages/provider"
12 type CostTracker struct {12 type CostTracker struct {
13 Total provider.Usage13 Total provider.Usage
14 LastTurn provider.Usage14 LastTurn provider.Usage
15+ hasTotal bool
15 }16 }
16 17
17 // Add folds u into the running total, records u as the last-turn18 // Add folds u into the running total, records u as the last-turn
18 // snapshot, and returns the new cumulative value.19 // snapshot, and returns the new cumulative value.
19 func (c *CostTracker) Add(u provider.Usage) provider.Usage {20 func (c *CostTracker) Add(u provider.Usage) provider.Usage {
20- c.Total = c.Total.Add(u)21+ if c.hasTotal {
22+ c.Total = c.Total.Add(u)
23+ } else {
24+ c.Total = u
25+ c.hasTotal = true
26+ }
21 c.LastTurn = u27 c.LastTurn = u
22 return c.Total28 return c.Total
23 }29 }
30+
31+// Seed replaces the running total with usage restored from a session.
32+func (c *CostTracker) Seed(u provider.Usage) {
33+ c.Total = u
34+ c.hasTotal = u.InputTokens != 0 || u.OutputTokens != 0 || u.CacheReadTokens != 0 || u.CacheWriteTokens != 0 || u.CostUSD != 0
35+}
M packages/core/session.go +2 -0
@@ -213,6 +213,8 @@ func SessionUsageDetail(path string) (cumulative, lastTurn provider.Usage, err e
213213 // input/output are monotonic totals -> per-turn = delta.
214214 lastTurn.InputTokens = nonNegDelta(cumulative.InputTokens, prevCum.InputTokens)
215215 lastTurn.OutputTokens = nonNegDelta(cumulative.OutputTokens, prevCum.OutputTokens)
216+ lastTurn.ReasoningTokens = nonNegDelta(cumulative.ReasoningTokens, prevCum.ReasoningTokens)
217+ lastTurn.ReasoningTokensKnown = cumulative.ReasoningTokensKnown
216218 // cache_read/write on the final row already represent the last prompt's
217219 // cache hit/creation, not a running total of bytes; use directly.
218220 lastTurn.CacheReadTokens = cumulative.CacheReadTokens - prevCum.CacheReadTokens
@@ -213,6 +213,8 @@ func SessionUsageDetail(path string) (cumulative, lastTurn provider.Usage, err e
213 // input/output are monotonic totals -> per-turn = delta.213 // input/output are monotonic totals -> per-turn = delta.
214 lastTurn.InputTokens = nonNegDelta(cumulative.InputTokens, prevCum.InputTokens)214 lastTurn.InputTokens = nonNegDelta(cumulative.InputTokens, prevCum.InputTokens)
215 lastTurn.OutputTokens = nonNegDelta(cumulative.OutputTokens, prevCum.OutputTokens)215 lastTurn.OutputTokens = nonNegDelta(cumulative.OutputTokens, prevCum.OutputTokens)
216+ lastTurn.ReasoningTokens = nonNegDelta(cumulative.ReasoningTokens, prevCum.ReasoningTokens)
217+ lastTurn.ReasoningTokensKnown = cumulative.ReasoningTokensKnown
216 // cache_read/write on the final row already represent the last prompt's218 // cache_read/write on the final row already represent the last prompt's
217 // cache hit/creation, not a running total of bytes; use directly.219 // cache hit/creation, not a running total of bytes; use directly.
218 lastTurn.CacheReadTokens = cumulative.CacheReadTokens - prevCum.CacheReadTokens220 lastTurn.CacheReadTokens = cumulative.CacheReadTokens - prevCum.CacheReadTokens
M packages/provider/gemini.go +11 -6
@@ -677,11 +677,11 @@ func (c *geminiClient) runStream(ctx context.Context, resp *http.Response, req R
677677 FinishReason string `json:"finishReason"`
678678 } `json:"candidates"`
679679 UsageMetadata *struct {
680- PromptTokenCount int `json:"promptTokenCount"`
681- CandidatesTokenCount int `json:"candidatesTokenCount"`
682- ThoughtsTokenCount int `json:"thoughtsTokenCount"`
683- CachedContentTokenCount int `json:"cachedContentTokenCount"`
684- TotalTokenCount int `json:"totalTokenCount"`
680+ PromptTokenCount int `json:"promptTokenCount"`
681+ CandidatesTokenCount int `json:"candidatesTokenCount"`
682+ ThoughtsTokenCount *int `json:"thoughtsTokenCount"`
683+ CachedContentTokenCount int `json:"cachedContentTokenCount"`
684+ TotalTokenCount int `json:"totalTokenCount"`
685685 } `json:"usageMetadata"`
686686 Error *struct {
687687 Code int `json:"code"`
@@ -753,7 +753,12 @@ func (c *geminiClient) runStream(ctx context.Context, resp *http.Response, req R
753753 input = um.PromptTokenCount
754754 }
755755 usage.InputTokens = input
756- usage.OutputTokens = um.CandidatesTokenCount + um.ThoughtsTokenCount
756+ usage.OutputTokens = um.CandidatesTokenCount
757+ if um.ThoughtsTokenCount != nil {
758+ usage.ReasoningTokens = *um.ThoughtsTokenCount
759+ usage.ReasoningTokensKnown = true
760+ usage.OutputTokens += *um.ThoughtsTokenCount
761+ }
757762 usage.CacheReadTokens = um.CachedContentTokenCount
758763 }
759764 // Promote ToolUse stop when tool calls are present and
@@ -677,11 +677,11 @@ func (c *geminiClient) runStream(ctx context.Context, resp *http.Response, req R
677 FinishReason string `json:"finishReason"`677 FinishReason string `json:"finishReason"`
678 } `json:"candidates"`678 } `json:"candidates"`
679 UsageMetadata *struct {679 UsageMetadata *struct {
680- PromptTokenCount int `json:"promptTokenCount"`680+ PromptTokenCount int `json:"promptTokenCount"`
681- CandidatesTokenCount int `json:"candidatesTokenCount"`681+ CandidatesTokenCount int `json:"candidatesTokenCount"`
682- ThoughtsTokenCount int `json:"thoughtsTokenCount"`682+ ThoughtsTokenCount *int `json:"thoughtsTokenCount"`
683- CachedContentTokenCount int `json:"cachedContentTokenCount"`683+ CachedContentTokenCount int `json:"cachedContentTokenCount"`
684- TotalTokenCount int `json:"totalTokenCount"`684+ TotalTokenCount int `json:"totalTokenCount"`
685 } `json:"usageMetadata"`685 } `json:"usageMetadata"`
686 Error *struct {686 Error *struct {
687 Code int `json:"code"`687 Code int `json:"code"`
@@ -753,7 +753,12 @@ func (c *geminiClient) runStream(ctx context.Context, resp *http.Response, req R
753 input = um.PromptTokenCount753 input = um.PromptTokenCount
754 }754 }
755 usage.InputTokens = input755 usage.InputTokens = input
756- usage.OutputTokens = um.CandidatesTokenCount + um.ThoughtsTokenCount756+ usage.OutputTokens = um.CandidatesTokenCount
757+ if um.ThoughtsTokenCount != nil {
758+ usage.ReasoningTokens = *um.ThoughtsTokenCount
759+ usage.ReasoningTokensKnown = true
760+ usage.OutputTokens += *um.ThoughtsTokenCount
761+ }
757 usage.CacheReadTokens = um.CachedContentTokenCount762 usage.CacheReadTokens = um.CachedContentTokenCount
758 }763 }
759 // Promote ToolUse stop when tool calls are present and764 // Promote ToolUse stop when tool calls are present and
M packages/provider/gemini_test.go +3 -0
@@ -510,6 +510,9 @@ func TestGeminiStreamReasoning(t *testing.T) {
510510 if usage.OutputTokens != 5 { // candidatesTokenCount (2) + thoughtsTokenCount (3)
511511 t.Fatalf("usage output tokens=%d, want 5", usage.OutputTokens)
512512 }
513+ if usage.ReasoningTokens != 3 || !usage.ReasoningTokensKnown {
514+ t.Fatalf("usage reasoning tokens=%d known=%v, want 3 true", usage.ReasoningTokens, usage.ReasoningTokensKnown)
515+ }
513516
514517 // Verify that the final message contains both the ReasoningBlock and the TextBlock
515518 msg := done.Message
@@ -510,6 +510,9 @@ func TestGeminiStreamReasoning(t *testing.T) {
510 if usage.OutputTokens != 5 { // candidatesTokenCount (2) + thoughtsTokenCount (3)510 if usage.OutputTokens != 5 { // candidatesTokenCount (2) + thoughtsTokenCount (3)
511 t.Fatalf("usage output tokens=%d, want 5", usage.OutputTokens)511 t.Fatalf("usage output tokens=%d, want 5", usage.OutputTokens)
512 }512 }
513+ if usage.ReasoningTokens != 3 || !usage.ReasoningTokensKnown {
514+ t.Fatalf("usage reasoning tokens=%d known=%v, want 3 true", usage.ReasoningTokens, usage.ReasoningTokensKnown)
515+ }
513 516
514 // Verify that the final message contains both the ReasoningBlock and the TextBlock517 // Verify that the final message contains both the ReasoningBlock and the TextBlock
515 msg := done.Message518 msg := done.Message
M packages/provider/openai.go +7 -0
@@ -633,6 +633,9 @@ func (c *openaiClient) runStream(ctx context.Context, resp *http.Response, req R
633633 PromptTokensDetails struct {
634634 CachedTokens int `json:"cached_tokens"`
635635 } `json:"prompt_tokens_details"`
636+ CompletionTokensDetails *struct {
637+ ReasoningTokens int `json:"reasoning_tokens"`
638+ } `json:"completion_tokens_details"`
636639 } `json:"usage"`
637640 Error *struct {
638641 Message string `json:"message"`
@@ -655,6 +658,10 @@ func (c *openaiClient) runStream(ctx context.Context, resp *http.Response, req R
655658 }
656659 usage.OutputTokens = chunk.Usage.CompletionTokens
657660 usage.CacheReadTokens = chunk.Usage.PromptTokensDetails.CachedTokens
661+ if details := chunk.Usage.CompletionTokensDetails; details != nil {
662+ usage.ReasoningTokens = details.ReasoningTokens
663+ usage.ReasoningTokensKnown = true
664+ }
658665 }
659666 for _, ch := range chunk.Choices {
660667 if ch.Delta.ReasoningContent != "" {
@@ -633,6 +633,9 @@ func (c *openaiClient) runStream(ctx context.Context, resp *http.Response, req R
633 PromptTokensDetails struct {633 PromptTokensDetails struct {
634 CachedTokens int `json:"cached_tokens"`634 CachedTokens int `json:"cached_tokens"`
635 } `json:"prompt_tokens_details"`635 } `json:"prompt_tokens_details"`
636+ CompletionTokensDetails *struct {
637+ ReasoningTokens int `json:"reasoning_tokens"`
638+ } `json:"completion_tokens_details"`
636 } `json:"usage"`639 } `json:"usage"`
637 Error *struct {640 Error *struct {
638 Message string `json:"message"`641 Message string `json:"message"`
@@ -655,6 +658,10 @@ func (c *openaiClient) runStream(ctx context.Context, resp *http.Response, req R
655 }658 }
656 usage.OutputTokens = chunk.Usage.CompletionTokens659 usage.OutputTokens = chunk.Usage.CompletionTokens
657 usage.CacheReadTokens = chunk.Usage.PromptTokensDetails.CachedTokens660 usage.CacheReadTokens = chunk.Usage.PromptTokensDetails.CachedTokens
661+ if details := chunk.Usage.CompletionTokensDetails; details != nil {
662+ usage.ReasoningTokens = details.ReasoningTokens
663+ usage.ReasoningTokensKnown = true
664+ }
658 }665 }
659 for _, ch := range chunk.Choices {666 for _, ch := range chunk.Choices {
660 if ch.Delta.ReasoningContent != "" {667 if ch.Delta.ReasoningContent != "" {
M packages/provider/openai_codex.go +7 -0
@@ -622,6 +622,9 @@ func (c *codexClient) runStream(ctx context.Context, resp *http.Response, req Re
622622 InputTokensDetails struct {
623623 CachedTokens int `json:"cached_tokens"`
624624 } `json:"input_tokens_details"`
625+ OutputTokensDetails *struct {
626+ ReasoningTokens int `json:"reasoning_tokens"`
627+ } `json:"output_tokens_details"`
625628 } `json:"usage"`
626629 Status string `json:"status"`
627630 } `json:"response"`
@@ -633,6 +636,10 @@ func (c *codexClient) runStream(ctx context.Context, resp *http.Response, req Re
633636 }
634637 usage.OutputTokens = p.Response.Usage.OutputTokens
635638 usage.CacheReadTokens = p.Response.Usage.InputTokensDetails.CachedTokens
639+ if details := p.Response.Usage.OutputTokensDetails; details != nil {
640+ usage.ReasoningTokens = details.ReasoningTokens
641+ usage.ReasoningTokensKnown = true
642+ }
636643
637644 hadTool := false
638645 for _, it := range items {
@@ -622,6 +622,9 @@ func (c *codexClient) runStream(ctx context.Context, resp *http.Response, req Re
622 InputTokensDetails struct {622 InputTokensDetails struct {
623 CachedTokens int `json:"cached_tokens"`623 CachedTokens int `json:"cached_tokens"`
624 } `json:"input_tokens_details"`624 } `json:"input_tokens_details"`
625+ OutputTokensDetails *struct {
626+ ReasoningTokens int `json:"reasoning_tokens"`
627+ } `json:"output_tokens_details"`
625 } `json:"usage"`628 } `json:"usage"`
626 Status string `json:"status"`629 Status string `json:"status"`
627 } `json:"response"`630 } `json:"response"`
@@ -633,6 +636,10 @@ func (c *codexClient) runStream(ctx context.Context, resp *http.Response, req Re
633 }636 }
634 usage.OutputTokens = p.Response.Usage.OutputTokens637 usage.OutputTokens = p.Response.Usage.OutputTokens
635 usage.CacheReadTokens = p.Response.Usage.InputTokensDetails.CachedTokens638 usage.CacheReadTokens = p.Response.Usage.InputTokensDetails.CachedTokens
639+ if details := p.Response.Usage.OutputTokensDetails; details != nil {
640+ usage.ReasoningTokens = details.ReasoningTokens
641+ usage.ReasoningTokensKnown = true
642+ }
636 643
637 hadTool := false644 hadTool := false
638 for _, it := range items {645 for _, it := range items {
M packages/provider/provider.go +16 -11
@@ -156,21 +156,26 @@ func activatedToolNames(messages []Message) map[string]bool {
156156
157157 // Usage aggregates token counts and cost for a turn.
158158 type Usage struct {
159- InputTokens int `json:"input_tokens"`
160- OutputTokens int `json:"output_tokens"`
161- CacheReadTokens int `json:"cache_read_tokens"`
162- CacheWriteTokens int `json:"cache_write_tokens"`
163- CostUSD float64 `json:"cost_usd"`
159+ InputTokens int `json:"input_tokens"`
160+ OutputTokens int `json:"output_tokens"`
161+ ReasoningTokens int `json:"reasoning_tokens"`
162+ ReasoningTokensKnown bool `json:"reasoning_tokens_known,omitempty"`
163+ CacheReadTokens int `json:"cache_read_tokens"`
164+ CacheWriteTokens int `json:"cache_write_tokens"`
165+ CostUSD float64 `json:"cost_usd"`
164166 }
165167
166-// Add returns u plus v.
168+// Add returns u plus v. A reasoning total is known only when both
169+// component usage reports include a separate reasoning-token count.
167170 func (u Usage) Add(v Usage) Usage {
168171 return Usage{
169- InputTokens: u.InputTokens + v.InputTokens,
170- OutputTokens: u.OutputTokens + v.OutputTokens,
171- CacheReadTokens: u.CacheReadTokens + v.CacheReadTokens,
172- CacheWriteTokens: u.CacheWriteTokens + v.CacheWriteTokens,
173- CostUSD: u.CostUSD + v.CostUSD,
172+ InputTokens: u.InputTokens + v.InputTokens,
173+ OutputTokens: u.OutputTokens + v.OutputTokens,
174+ ReasoningTokens: u.ReasoningTokens + v.ReasoningTokens,
175+ ReasoningTokensKnown: u.ReasoningTokensKnown && v.ReasoningTokensKnown,
176+ CacheReadTokens: u.CacheReadTokens + v.CacheReadTokens,
177+ CacheWriteTokens: u.CacheWriteTokens + v.CacheWriteTokens,
178+ CostUSD: u.CostUSD + v.CostUSD,
174179 }
175180 }
176181
@@ -156,21 +156,26 @@ func activatedToolNames(messages []Message) map[string]bool {
156 156
157 // Usage aggregates token counts and cost for a turn.157 // Usage aggregates token counts and cost for a turn.
158 type Usage struct {158 type Usage struct {
159- InputTokens int `json:"input_tokens"`159+ InputTokens int `json:"input_tokens"`
160- OutputTokens int `json:"output_tokens"`160+ OutputTokens int `json:"output_tokens"`
161- CacheReadTokens int `json:"cache_read_tokens"`161+ ReasoningTokens int `json:"reasoning_tokens"`
162- CacheWriteTokens int `json:"cache_write_tokens"`162+ ReasoningTokensKnown bool `json:"reasoning_tokens_known,omitempty"`
163- CostUSD float64 `json:"cost_usd"`163+ CacheReadTokens int `json:"cache_read_tokens"`
164+ CacheWriteTokens int `json:"cache_write_tokens"`
165+ CostUSD float64 `json:"cost_usd"`
164 }166 }
165 167
166-// Add returns u plus v.168+// Add returns u plus v. A reasoning total is known only when both
169+// component usage reports include a separate reasoning-token count.
167 func (u Usage) Add(v Usage) Usage {170 func (u Usage) Add(v Usage) Usage {
168 return Usage{171 return Usage{
169- InputTokens: u.InputTokens + v.InputTokens,172+ InputTokens: u.InputTokens + v.InputTokens,
170- OutputTokens: u.OutputTokens + v.OutputTokens,173+ OutputTokens: u.OutputTokens + v.OutputTokens,
171- CacheReadTokens: u.CacheReadTokens + v.CacheReadTokens,174+ ReasoningTokens: u.ReasoningTokens + v.ReasoningTokens,
172- CacheWriteTokens: u.CacheWriteTokens + v.CacheWriteTokens,175+ ReasoningTokensKnown: u.ReasoningTokensKnown && v.ReasoningTokensKnown,
173- CostUSD: u.CostUSD + v.CostUSD,176+ CacheReadTokens: u.CacheReadTokens + v.CacheReadTokens,
177+ CacheWriteTokens: u.CacheWriteTokens + v.CacheWriteTokens,
178+ CostUSD: u.CostUSD + v.CostUSD,
174 }179 }
175 }180 }
176 181
A packages/provider/reasoning_usage_test.go +57 -0
@@ -0,0 +1,57 @@
1+package provider
2+
3+import (
4+ "context"
5+ "fmt"
6+ "net/http"
7+ "net/http/httptest"
8+ "testing"
9+)
10+
11+func TestOpenAIReportsReasoningTokens(t *testing.T) {
12+ srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
13+ w.Header().Set("content-type", "text/event-stream")
14+ fmt.Fprintln(w, `data: {"choices":[{"delta":{"content":"ok"},"finish_reason":"stop"}],"usage":{"prompt_tokens":12,"completion_tokens":9,"completion_tokens_details":{"reasoning_tokens":6}}}`)
15+ fmt.Fprintln(w)
16+ fmt.Fprintln(w, "data: [DONE]")
17+ fmt.Fprintln(w)
18+ }))
19+ defer srv.Close()
20+
21+ events, err := NewOpenAI("test", srv.URL).Stream(context.Background(), Request{Model: "gpt-test"})
22+ if err != nil {
23+ t.Fatal(err)
24+ }
25+ usage := usageFromEvents(events)
26+ if usage.OutputTokens != 9 || usage.ReasoningTokens != 6 || !usage.ReasoningTokensKnown {
27+ t.Fatalf("usage = %+v", usage)
28+ }
29+}
30+
31+func TestOpenAIResponsesReportsReasoningTokens(t *testing.T) {
32+ srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
33+ w.Header().Set("content-type", "text/event-stream")
34+ fmt.Fprintln(w, `data: {"type":"response.completed","response":{"status":"completed","usage":{"input_tokens":12,"output_tokens":9,"output_tokens_details":{"reasoning_tokens":6}}}}`)
35+ fmt.Fprintln(w)
36+ }))
37+ defer srv.Close()
38+
39+ events, err := NewOpenAICodex("test", "account", srv.URL).Stream(context.Background(), Request{Model: "gpt-5.5"})
40+ if err != nil {
41+ t.Fatal(err)
42+ }
43+ usage := usageFromEvents(events)
44+ if usage.OutputTokens != 9 || usage.ReasoningTokens != 6 || !usage.ReasoningTokensKnown {
45+ t.Fatalf("usage = %+v", usage)
46+ }
47+}
48+
49+func usageFromEvents(events <-chan Event) Usage {
50+ var usage Usage
51+ for event := range events {
52+ if e, ok := event.(EventUsage); ok {
53+ usage = e.Usage
54+ }
55+ }
56+ return usage
57+}
@@ -0,0 +1,57 @@
1+package provider
2+
3+import (
4+ "context"
5+ "fmt"
6+ "net/http"
7+ "net/http/httptest"
8+ "testing"
9+)
10+
11+func TestOpenAIReportsReasoningTokens(t *testing.T) {
12+ srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
13+ w.Header().Set("content-type", "text/event-stream")
14+ fmt.Fprintln(w, `data: {"choices":[{"delta":{"content":"ok"},"finish_reason":"stop"}],"usage":{"prompt_tokens":12,"completion_tokens":9,"completion_tokens_details":{"reasoning_tokens":6}}}`)
15+ fmt.Fprintln(w)
16+ fmt.Fprintln(w, "data: [DONE]")
17+ fmt.Fprintln(w)
18+ }))
19+ defer srv.Close()
20+
21+ events, err := NewOpenAI("test", srv.URL).Stream(context.Background(), Request{Model: "gpt-test"})
22+ if err != nil {
23+ t.Fatal(err)
24+ }
25+ usage := usageFromEvents(events)
26+ if usage.OutputTokens != 9 || usage.ReasoningTokens != 6 || !usage.ReasoningTokensKnown {
27+ t.Fatalf("usage = %+v", usage)
28+ }
29+}
30+
31+func TestOpenAIResponsesReportsReasoningTokens(t *testing.T) {
32+ srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
33+ w.Header().Set("content-type", "text/event-stream")
34+ fmt.Fprintln(w, `data: {"type":"response.completed","response":{"status":"completed","usage":{"input_tokens":12,"output_tokens":9,"output_tokens_details":{"reasoning_tokens":6}}}}`)
35+ fmt.Fprintln(w)
36+ }))
37+ defer srv.Close()
38+
39+ events, err := NewOpenAICodex("test", "account", srv.URL).Stream(context.Background(), Request{Model: "gpt-5.5"})
40+ if err != nil {
41+ t.Fatal(err)
42+ }
43+ usage := usageFromEvents(events)
44+ if usage.OutputTokens != 9 || usage.ReasoningTokens != 6 || !usage.ReasoningTokensKnown {
45+ t.Fatalf("usage = %+v", usage)
46+ }
47+}
48+
49+func usageFromEvents(events <-chan Event) Usage {
50+ var usage Usage
51+ for event := range events {
52+ if e, ok := event.(EventUsage); ok {
53+ usage = e.Usage
54+ }
55+ }
56+ return usage
57+}

Keyboard shortcuts

Single-key shortcuts are on. ctrl k and esc work whatever this says.

ctrl k :
go to a section, branch, tag, file or repository
?
this list
j k
move the row cursor down / up
enter
open the row under the cursor; on Files, a directory opens or closes in place
↑ ↓ ← → home end
in the file tree (focus on a row): previous / next row, close / open a directory or go to its parent / first entry, first / last row. They never open a file
l h
on Files: open a directory or a file / close a directory or go up to its parent, then to the parent page. Elsewhere: into the side panel (on a merge request, its files) / back to the content; there j k move between its controls
g g G
first / last row
/
filter the list; on Files, find a file anywhere in the repository (esc clears, enter opens the best match)
b
switch branch or tag
g s g f g c g b g m g i g r g a
tabs: summary, files, commits, branches & tags, merge requests, issues, releases, actions (when available)
g h
home
w
wide mode: use the whole width of the screen, or not (remembered)
y
copy a permalink pinned to the commit
d u x
diff: split / unified / collapse or expand all
e E
diff: the whole file in view / every file, with the lines the hunks hide
n
merge request: next note in the diff
esc
close a dialog, leave a field

Branches 1

Tags 476

↑↓ moveenter openesc close