BuzzbeeSCD/openchamber
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(ui): stop the context meter from counting every internal round-trip

Browse Source 
The token breakdown of an assistant message accumulates across every API
round-trip inside the turn: each tool call re-reads the whole cached
prompt, so input/cache.read add up to several times the context window.
Every context-usage surface summed those fields, which is why the meter
could read 330% of a 1M window whose real fill was 232,872 tokens
(23.3%), and why reopening an older session jumps the readout (#2562).

The server reports the final round-trip's window as tokens.total
(optional in the message schema; opencode 1.18.18 returns it, verified
against its live /session/:id/message API). Prefer it everywhere the
window fill is displayed and fall back to summing only when the server
did not send it: contextTokensFromBreakdown in tokenUtils now owns that
rule, and the context store extractor, sync store getter, work status
panel, context sidebar, VS Code layout, mini chat, and mobile metadata
all use it instead of their own inline sums.

Fixes #2562
This commit is contained in:
dibanez
2026-08-15 21:46:49 +02:00
parent e3094ee676
commit 9e1a9b59b1
11 changed files with 156 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files
packages/ui/src/components/chat/work-status/contextUsage.test.ts
+22 -2
View File
@@ -14,8 +14,8 @@ describe('computeContextUsage', () => {
});
test('reports the latest turn rather than a sum across turns', () => {
// Each assistant turn reports the whole window it saw, so adding them up
// would report several times the real fill.
// A turn's tokens describe that turn's window, so adding turns up would
// report several times the real fill.
const usage = computeContextUsage(
[
assistant({ input: 400, output: 0, reasoning: 0 }, 'old'),
@@ -61,4 +61,24 @@ describe('computeContextUsage', () => {
const usage = computeContextUsage([assistant({ input: 10 })], 100);
expect(usage?.totalTokens).toBe(10);
});
test('prefers the server-reported total over summing round-trip fields', () => {
// Real payload from opencode 1.18.18: ~14 tool-call round-trips accumulated
// cache.read to 3.29M while the 1M window really held 232,872. Summing
// rendered 330.6%; the reported total renders the real 23.3%.
const usage = computeContextUsage(
[assistant({ total: 232_872, input: 0, output: 14_523, reasoning: 0, cache: { read: 3_291_956, write: 0 } })],
1_000_000,
);
expect(usage?.totalTokens).toBe(232_872);
expect(usage?.percent.toFixed(4)).toBe('23.2872');
});
test('selects a message whose only signal is the reported total', () => {
const usage = computeContextUsage(
[assistant({ total: 5_000, input: 0, output: 0, reasoning: 0 })],
100_000,
);
expect(usage?.totalTokens).toBe(5_000);
});
});
packages/ui/src/components/chat/work-status/contextUsage.ts
+9 -11
View File
@@ -13,7 +13,11 @@
* global read to race with.
*/
import { contextTokensFromBreakdown } from '@/stores/utils/tokenUtils';
type MessageTokens = {
/** Server-reported window of the turn's final round-trip; absent on older servers. */
total?: number;
input?: number;
output?: number;
reasoning?: number;
@@ -37,18 +41,12 @@ type WorkStatusContextUsage = {
/** The store's own fallback when a model exposes no context limit. */
export const DEFAULT_CONTEXT_LIMIT = 200_000;
const sumTokens = (tokens: MessageTokens): number => (
(tokens.input ?? 0)
+ (tokens.output ?? 0)
+ (tokens.reasoning ?? 0)
+ (tokens.cache?.read ?? 0)
+ (tokens.cache?.write ?? 0)
);
/**
* Usage from the newest assistant message that reported a non-zero token count.
* Each assistant turn reports the whole window it saw, so the latest one is the
* current fill — not a sum across turns.
* The latest turn describes the current fill — not a sum across turns. Within
* a turn, the server-reported `total` is the final round-trip's window;
* summing the breakdown fields instead overstates multi-step turns, whose
* input/cache fields accumulate across round-trips.
*/
export const computeContextUsage = (
messages: readonly MessageLike[],
@@ -60,7 +58,7 @@ export const computeContextUsage = (
const message = messages[index];
if (message?.role !== 'assistant' || !message.tokens) continue;
const totalTokens = sumTokens(message.tokens);
const totalTokens = contextTokensFromBreakdown(message.tokens);
if (totalTokens <= 0) continue;
const limit = contextLimit > 0 ? contextLimit : DEFAULT_CONTEXT_LIMIT;
packages/ui/src/components/layout/ContextSidebarTab.tsx
+6 -1
View File
@@ -92,6 +92,7 @@ const extractTokenBreakdown = (message: SessionMessage): TokenBreakdown => {
}
const breakdown = source as {
total?: unknown;
input?: unknown;
output?: unknown;
reasoning?: unknown;
@@ -103,6 +104,10 @@ const extractTokenBreakdown = (message: SessionMessage): TokenBreakdown => {
const reasoning = toNonNegativeNumber(breakdown.reasoning);
const cacheRead = toNonNegativeNumber(breakdown.cache?.read);
const cacheWrite = toNonNegativeNumber(breakdown.cache?.write);
// Multi-step turns accumulate the fields across API round-trips (every tool
// call re-reads the whole cached prompt), so summing them overstates the
// window. The server-reported total is the final round-trip's window.
const reportedTotal = toNonNegativeNumber(breakdown.total);
return {
input,
@@ -110,7 +115,7 @@ const extractTokenBreakdown = (message: SessionMessage): TokenBreakdown => {
reasoning,
cacheRead,
cacheWrite,
total: input + output + reasoning + cacheRead + cacheWrite,
total: reportedTotal > 0 ? reportedTotal : input + output + reasoning + cacheRead + cacheWrite,
};
};
packages/ui/src/components/layout/VSCodeLayout.tsx
+3 -2
View File
@@ -8,6 +8,7 @@ import { useViewportStore } from '@/sync/viewport-store';
import { useSessions, useDirectorySync, useSessionMessages, useSessionMessagesResolved } from '@/sync/sync-context';
import { useConfigStore } from '@/stores/useConfigStore';
import { resolveGlobalSessionDirectory, useGlobalSessionsStore } from '@/stores/useGlobalSessionsStore';
import { contextTokensFromBreakdown } from '@/stores/utils/tokenUtils';
import { ContextUsageDisplay } from '@/components/ui/ContextUsageDisplay';
import { McpDropdown } from '@/components/mcp/McpDropdown';
import { ArchiveAllDropdown } from '@/components/session/ArchiveAllDropdown';
@@ -702,7 +703,7 @@ const VSCodeHeader: React.FC<VSCodeHeaderProps> = ({ title, showBack, onBack, on
}
if (!lastTokens && message.tokens) {
const total = message.tokens.input + message.tokens.output + message.tokens.reasoning + (message.tokens.cache?.read ?? 0) + (message.tokens.cache?.write ?? 0);
const total = contextTokensFromBreakdown(message.tokens);
if (total > 0) {
lastTokens = message.tokens;
lastMessageId = (currentSessionMessages[i] as { id?: string }).id;
@@ -730,7 +731,7 @@ const VSCodeHeader: React.FC<VSCodeHeaderProps> = ({ title, showBack, onBack, on
}
const lastTokens = headerMessageSummary.lastTokens;
const totalTokens = lastTokens.input + lastTokens.output + lastTokens.reasoning + (lastTokens.cache?.read ?? 0) + (lastTokens.cache?.write ?? 0);
const totalTokens = contextTokensFromBreakdown(lastTokens);
const thresholdLimit = contextLimit > 0 ? contextLimit : 200000;
const percentage = contextLimit > 0 ? Math.round((totalTokens / contextLimit) * 100) : 0;
const normalizedOutput = outputLimit > 0 ? Math.round((lastTokens.output / outputLimit) * 100) : undefined;
packages/ui/src/components/mini-chat/MiniChatLayout.tsx
+4 -3
View File
@@ -18,6 +18,7 @@ import { useGitBranchLabel, useGitStore } from '@/stores/useGitStore';
import { useConfigStore } from '@/stores/useConfigStore';
import { Icon } from "@/components/icon/Icon";
import { useRuntimeAPIs } from '@/hooks/useRuntimeAPIs';
import { contextTokensFromBreakdown } from '@/stores/utils/tokenUtils';
import type { SessionContextUsage } from '@/stores/types/sessionTypes';
type MiniChatMode = 'session' | 'draft';
@@ -157,7 +158,7 @@ const MiniChatHeader: React.FC<{ mode: MiniChatMode }> = ({ mode }) => {
return null;
}
type AssistantTokens = { input: number; output: number; reasoning: number; cache: { read: number; write: number } };
type AssistantTokens = { total?: number; input: number; output: number; reasoning: number; cache: { read: number; write: number } };
let lastTokens: AssistantTokens | undefined;
let lastMessageId: string | undefined;
@@ -166,7 +167,7 @@ const MiniChatHeader: React.FC<{ mode: MiniChatMode }> = ({ mode }) => {
if (message.role !== 'assistant') continue;
const tokens = (message as { tokens?: AssistantTokens }).tokens;
if (!tokens) continue;
const total = tokens.input + tokens.output + tokens.reasoning + (tokens.cache?.read ?? 0) + (tokens.cache?.write ?? 0);
const total = contextTokensFromBreakdown(tokens);
if (total > 0) {
lastTokens = tokens;
lastMessageId = message.id;
@@ -178,7 +179,7 @@ const MiniChatHeader: React.FC<{ mode: MiniChatMode }> = ({ mode }) => {
return null;
}
const totalTokens = lastTokens.input + lastTokens.output + lastTokens.reasoning + (lastTokens.cache?.read ?? 0) + (lastTokens.cache?.write ?? 0);
const totalTokens = contextTokensFromBreakdown(lastTokens);
const thresholdLimit = contextLimit > 0 ? contextLimit : 200000;
const percentage = contextLimit > 0 ? Math.round((totalTokens / contextLimit) * 100) : 0;
const normalizedOutput = outputLimit > 0 ? Math.round((lastTokens.output / outputLimit) * 100) : undefined;