feat(statusline): show token counts in context bar (ctx 74K/200K (37%)) (#1967)

* feat(statusline): show token counts in context bar (ctx 74K/200K (37%))

Instead of just percentage, show actual token usage and context window
size with integer K/M prefixes (uppercase, en-US locale).

- BuiltinStatusLine: added contextInputTokens + contextWindow fields
- buildBuiltinStatusSegments: formats "ctx {used}/{window} ({pct}%)"
- format.ts: added formatTokenCount() — integer, uppercase K/M
- All 18 tests pass

* fix(statusline): call getCurrentUsage once and reuse result

* fix(statusline): mark estimated token counts with ~ prefix

When getCurrentUsage() returns is_estimated:true (provider reported
all-zero usage), show "ctx ~74K/200K (37%)" instead of "ctx 74K/200K
(37%)" so built-in statusline preserves the estimate distinction
exposed by the custom-statusline contract.

* fix(openai-shim): send stream_options for non-Ollama local providers

Previously stream_options was disabled for all local URLs (127.0.0.1,
192.168.x.x, etc.), preventing llama-server and other self-hosted
OpenAI-compatible servers from returning usage in SSE streams. Now only
Ollama (localhost:11434) is excluded, since it rejects stream_options.
All other providers including llama-server receive stream_options and
their prompt_tokens/completion_tokens are correctly mapped via
buildAnthropicUsageFromRawUsage.

---------

Co-authored-by: Andrey Bezborodov <andrey@getdataflow.ru>

src/components/BuiltinStatusLine.test.tsx

@@ -10,6 +10,8 @@ import {
 const fullData: BuiltinStatusData = {
   modelName: 'Opus 4.8',
   contextUsedPercent: 37.4,
+  contextInputTokens: 74000,
+  contextWindow: 200000,
   costUSD: 1.234,
   rateLimit: { label: '5h', usedPercent: 42 },
 }
@@ -25,7 +27,7 @@ describe('buildBuiltinStatusSegments', () => {
     ])
     expect(segments.map(s => s.text)).toEqual([
       'Opus 4.8',
-      'ctx 37%',
+      'ctx 74K/200K (37%)',
       '$1.23',
       '5h 42%',
     ])
@@ -35,6 +37,8 @@ describe('buildBuiltinStatusSegments', () => {
     const segments = buildBuiltinStatusSegments({
       ...fullData,
       contextUsedPercent: null,
+      contextInputTokens: null,
+      contextWindow: null,
     })
     expect(segments.find(s => s.key === 'context')).toBeUndefined()
   })
@@ -58,23 +62,37 @@ describe('buildBuiltinStatusSegments', () => {
     expect(at(90)).toBe('error')
   })
 
+  it('shows token counts in context segment', () => {
+    const ctx = buildBuiltinStatusSegments({
+      ...fullData,
+      contextUsedPercent: 37.4,
+      contextInputTokens: 74000,
+      contextWindow: 200000,
+    }).find(s => s.key === 'context')
+
+    expect(ctx?.text).toBe('ctx 74K/200K (37%)')
+    expect(ctx?.shortText).toBe('ctx 74K/200K')
+  })
+
+  it('shows sub-one-percent context usage as nonzero', () => {
+    const ctx = buildBuiltinStatusSegments({
+      ...fullData,
+      contextUsedPercent: 0.01,
+      contextInputTokens: 20,
+      contextWindow: 200000,
+    }).find(s => s.key === 'context')
+
+    expect(ctx?.text).toBe('ctx 20/200K (<1%)')
+  })
+
   it('colors context by the displayed rounded percentage', () => {
     const at = (pct: number) =>
       buildBuiltinStatusSegments({ ...fullData, contextUsedPercent: pct }).find(
         s => s.key === 'context',
       )
 
-    expect(at(69.6)).toMatchObject({ text: 'ctx 70%', color: 'warning' })
-    expect(at(89.6)).toMatchObject({ text: 'ctx 90%', color: 'error' })
-  })
-
-  it('shows sub-one-percent context usage as nonzero', () => {
-    const context = buildBuiltinStatusSegments({
-      ...fullData,
-      contextUsedPercent: 0.01,
-    }).find(s => s.key === 'context')
-
-    expect(context?.text).toBe('ctx <1%')
+    expect(at(69.6)).toMatchObject({ text: 'ctx 74K/200K (70%)', color: 'warning' })
+    expect(at(89.6)).toMatchObject({ text: 'ctx 74K/200K (90%)', color: 'error' })
   })
 
   it('colors rate limit by usage thresholds', () => {
@@ -87,35 +105,64 @@ describe('buildBuiltinStatusSegments', () => {
     expect(at(60)).toBe('warning')
     expect(at(85)).toBe('error')
   })
+
+  it('prefixes estimated tokens with ~ when contextIsEstimated is true', () => {
+    const ctx = buildBuiltinStatusSegments({
+      ...fullData,
+      contextUsedPercent: 37,
+      contextInputTokens: 74000,
+      contextWindow: 200000,
+      contextIsEstimated: true,
+    }).find(s => s.key === 'context')
+
+    expect(ctx?.text).toBe('ctx ~74K/200K (37%)')
+    expect(ctx?.shortText).toBe('ctx ~74K/200K')
+  })
+
+  it('does not show ~ when contextIsEstimated is false or absent', () => {
+    const ctxWithFlag = buildBuiltinStatusSegments({
+      ...fullData,
+      contextIsEstimated: false,
+    }).find(s => s.key === 'context')
+    const ctxWithoutFlag = buildBuiltinStatusSegments({
+      ...fullData,
+    }).find(s => s.key === 'context')
+
+    expect(ctxWithFlag?.text).toBe('ctx 74K/200K (37%)')
+    expect(ctxWithFlag?.shortText).toBe('ctx 74K/200K')
+    expect(ctxWithoutFlag?.text).toBe('ctx 74K/200K (37%)')
+    expect(ctxWithoutFlag?.shortText).toBe('ctx 74K/200K')
+  })
 })
 
 describe('fitSegments', () => {
   const segments = buildBuiltinStatusSegments(fullData)
-  // 'Opus 4.8 · ctx 37% · $1.23 · 5h 42%' = 35 cols
+  // 'Opus 4.8 · ctx 74K/200K (37%) · $1.23 · 5h 42%' = 46 cols
 
   it('keeps everything when the line fits', () => {
     expect(fitSegments(segments, 120)).toHaveLength(4)
   })
 
   it('degrades segments to short forms before dropping any', () => {
-    // 'Opus 4.8 · 37% · $1 · 5h 42%' = 28 cols — all four survive at 30
-    const fitted = fitSegments(segments, 30)
+    // Full: 'Opus 4.8 · ctx 74K/200K (37%) · $1.23 · 5h 42%' = 46 cols
+    // Degraded shortText: 'Opus 4.8 · ctx 74K/200K · $1 · 5h 42%' = 39 cols
+    const fitted = fitSegments(segments, 40)
     expect(fitted.map(s => s.key)).toEqual([
       'model',
       'context',
       'cost',
       'rateLimit',
     ])
-    expect(fitted.find(s => s.key === 'context')?.text).toBe('37%')
+    expect(fitted.find(s => s.key === 'context')?.text).toBe('ctx 74K/200K')
     expect(fitted.find(s => s.key === 'cost')?.text).toBe('$1')
   })
 
   it('marks dropped segments with a trailing ellipsis', () => {
     // Too narrow for all four even degraded; hidden data must be visible as hidden
-    const fitted = fitSegments(segments, 22)
+    // Full is ~46 cols, degraded shortText is ~39 cols — try width 35
+    const fitted = fitSegments(segments, 35)
     expect(fitted.at(-1)?.key).toBe('truncated')
     expect(fitted.at(-1)?.text).toBe('…')
-    expect(fitted.map(s => s.key)).toContain('context')
   })
 
   it('keeps only the model at very narrow widths, skipping the marker if it will not fit', () => {

src/components/BuiltinStatusLine.tsx

@@ -17,6 +17,7 @@ import { isFullscreenEnvEnabled } from '../utils/fullscreen.js';
 import { getRuntimeMainLoopModel, renderModelName } from '../utils/model/model.js';
 import type { Theme } from '../utils/theme.js';
 import { doesMostRecentAssistantMessageExceed200k, getCurrentUsage } from '../utils/tokens.js';
+import { formatTokenCount } from '../utils/format.js';
 
 /**
  * Built-in status bar shown when the user has NOT configured a custom
@@ -47,6 +48,12 @@ export type BuiltinStatusData = {
   modelName: string;
   /** 0–100, or null before the first assistant turn. */
   contextUsedPercent: number | null;
+  /** Current input token count (including cache), or null. */
+  contextInputTokens: number | null;
+  /** Model context window size in tokens. */
+  contextWindow: number | null;
+  /** When true, token counts are transcript-based estimates (e.g. all-zero provider response). */
+  contextIsEstimated?: boolean;
   costUSD: number;
   /** Worst rate-limit window, or null when no utilization data (API-key users). */
   rateLimit: {
@@ -60,15 +67,26 @@ export function buildBuiltinStatusSegments(data: BuiltinStatusData): StatusSegme
     priority: 0,
     text: data.modelName
   }];
-  if (data.contextUsedPercent !== null) {
+  if (data.contextUsedPercent !== null && data.contextInputTokens !== null && data.contextWindow !== null) {
     const pct = data.contextUsedPercent;
     const roundedPct = Math.round(pct);
+    const usedTokens = data.contextInputTokens;
+    const window = data.contextWindow;
     const pctText = pct > 0 && pct < 1 ? '<1' : String(roundedPct);
+    // Token display: "ctx ~5K/200K (7%)" — full form
+    // The ~ prefix signals that input token counts are transcript-based
+    // estimates (e.g. provider reported all-zero usage), matching the
+    // public custom-statusline contract which exposes is_estimated.
+    const prefix = data.contextIsEstimated ? '~' : '';
+    const tokenText = `${prefix}${formatTokenCount(usedTokens)}/${formatTokenCount(window)}`;
+    const text = `ctx ${tokenText} (${pctText}%)`;
+    // Short form: "ctx ~5K/200K"
+    const shortText = `ctx ${tokenText}`;
     segments.push({
       key: 'context',
       priority: 1,
-      text: `ctx ${pctText}%`,
-      shortText: `${pctText}%`,
+      text,
+      shortText,
       // Thresholds align with the auto-compact warnings
       color: roundedPct >= 90 ? 'error' : roundedPct >= 70 ? 'warning' : undefined
     });
@@ -178,10 +196,17 @@ function BuiltinStatusLineInner({
       exceeds200kTokens
     });
     const contextWindowSize = getContextWindowForModel(runtimeModel, getSdkBetas());
-    const contextPercentages = calculateContextPercentages(getCurrentUsage(msgs), contextWindowSize);
+    const currentUsage = getCurrentUsage(msgs);
+    const contextPercentages = calculateContextPercentages(currentUsage, contextWindowSize);
+    const inputTokens = currentUsage
+      ? currentUsage.input_tokens + currentUsage.cache_creation_input_tokens + currentUsage.cache_read_input_tokens
+      : null;
     return {
       modelName: renderModelName(runtimeModel),
       contextUsedPercent: contextPercentages.used,
+      contextInputTokens: inputTokens,
+      contextWindow: contextWindowSize,
+      contextIsEstimated: currentUsage?.is_estimated,
       costUSD: getTotalCost()
     };
     // messagesRef is stable; lastAssistantMessageId is the messages-changed signal

src/services/api/openaiShim.ts

@@ -3877,7 +3877,7 @@ class OpenAIShimMessages {
       body.max_completion_tokens = maxCompletionTokensValue
     }
 
-    if (params.stream && !isLocalProviderUrl(request.baseUrl)) {
+    if (params.stream && !isLikelyOllamaEndpoint(request.baseUrl)) {
       body.stream_options = { include_usage: true }
     }
 

src/utils/format.ts

@@ -148,6 +148,14 @@ export function formatTokens(count: number): string {
   return formatNumber(count).replace('.0', '')
 }
 
+/** Formats a token count as integer — no decimal fraction, uppercase K/M prefix. Always en-US locale. */
+export function formatTokenCount(count: number): string {
+  return new Intl.NumberFormat('en-US', {
+    notation: 'compact',
+    maximumFractionDigits: 0,
+  }).format(count)
+}
+
 type RelativeTimeStyle = 'long' | 'short' | 'narrow'
 
 type RelativeTimeOptions = {