feat(extension): show session cost in the context popover

apps/extension/entrypoints/sidepanel/agent-chat-atoms.ts

@@ -20,6 +20,10 @@ export const contextUsageAtomFamily = atomFamily((_conversationId: string) =>
   atom<ContextUsage | undefined>()
 );
 
+// Per-conversation running session spend (USD) from streamed usage.cost (in-memory only).
+// Same lifecycle as drafts/usage: kept across close and compaction, evicted on delete and sign-out.
+export const sessionCostAtomFamily = atomFamily((_conversationId: string) => atom(0));
+
 // Event id of the assistant message currently streaming content for a conversation.
 // Cleared when that message's stream ends (even if later tool rounds continue).
 export const streamingMessageIdAtomFamily = atomFamily((_conversationId: string) =>
@@ -30,10 +34,11 @@ export const streamingMessageIdAtomFamily = atomFamily((_conversationId: string)
 export const runningConversationIdsAtom = atom<readonly string[]>([]);
 export const compactingConversationIdsAtom = atom<readonly string[]>([]);
 
-// Evict a single conversation's in-memory atoms (draft + context usage + streaming id).
+// Evict a single conversation's in-memory atoms (draft + context usage + session cost + streaming id).
 export const evictConversationAtoms = (conversationId: string): void => {
   draftAtomFamily.remove(conversationId);
   contextUsageAtomFamily.remove(conversationId);
+  sessionCostAtomFamily.remove(conversationId);
   streamingMessageIdAtomFamily.remove(conversationId);
 };
 
@@ -47,6 +52,7 @@ export const clearPerConversationAtoms = (): void => {
   const ids = new Set([
     ...draftAtomFamily.getParams(),
     ...contextUsageAtomFamily.getParams(),
+    ...sessionCostAtomFamily.getParams(),
     ...streamingMessageIdAtomFamily.getParams(),
   ]);
   for (const id of ids) {

apps/extension/entrypoints/sidepanel/agent-chat-panel.tsx

@@ -10,6 +10,7 @@ import {
   evictConversationAtoms,
   remoteMcpStoreAtom,
   runningConversationIdsAtom,
+  sessionCostAtomFamily,
   streamingMessageIdAtomFamily,
 } from './agent-chat-atoms';
 import {
@@ -24,6 +25,7 @@ import {
   compactConversationEvents,
   hasCompactableHistory,
 } from '@/src/shared/agent-context-compaction';
+import type { TurnUsage } from '@/src/shared/agent-llm-turn-runner-core';
 import { defaultMode } from '@/src/shared/agent-chat-placeholder';
 import { getKiloApiBaseUrl } from '@/src/shared/auth';
 import type { StoredAuth } from '@/src/shared/auth';
@@ -46,6 +48,7 @@ import { AgentFooterControls } from './agent-footer-controls';
 import { ContextDonut } from './context-donut';
 import { runDangerousLlmTurn, runSafeLlmTurn } from './agent-turn-runners';
 import { AUTO_COMPACT_RATIO, getContextRatio } from '@/src/shared/context-usage';
+import { addSessionCost } from '@/src/shared/session-cost';
 import { useTabDebugger } from './use-tab-debugger';
 import { ConversationList } from './conversation-list';
 import { ConversationTabs } from './conversation-tabs';
@@ -162,6 +165,7 @@ export const AgentChatPanel = ({
   const isCompacting = compactingConversationIds.includes(activeConversationId);
   const activeUsage = useAtomValue(contextUsageAtomFamily(activeConversationId));
   const activePromptTokens = activeUsage?.promptTokens ?? 0;
+  const activeSessionCostUsd = useAtomValue(sessionCostAtomFamily(activeConversationId));
   const streamingMessageId = useAtomValue(streamingMessageIdAtomFamily(activeConversationId));
   const contextLength = selectedModel?.contextLength;
 
@@ -246,10 +250,12 @@ export const AgentChatPanel = ({
           void compactActiveConversation();
         }}
         promptTokens={activePromptTokens}
+        sessionCostUsd={activeSessionCostUsd}
       />
     ),
     [
       activePromptTokens,
+      activeSessionCostUsd,
       canCompactActive,
       compactActiveConversation,
       contextLength,
@@ -471,10 +477,15 @@ export const AgentChatPanel = ({
       }
     };
     let currentRunHasUsage = false;
-    const updateRunUsage = (usage: { promptTokens: number }): void => {
+    const updateRunUsage = (usage: TurnUsage): void => {
       if (isCurrentRun()) {
         currentRunHasUsage = true;
         store.set(contextUsageAtomFamily(conversationId), { promptTokens: usage.promptTokens });
+        const previousCost = store.get(sessionCostAtomFamily(conversationId));
+        store.set(
+          sessionCostAtomFamily(conversationId),
+          addSessionCost(previousCost, usage.costUsd)
+        );
       }
     };
 

apps/extension/entrypoints/sidepanel/context-donut.tsx

@@ -1,6 +1,7 @@
 import { useRef } from 'react';
 import type { JSX } from 'react';
 import { formatContextSummary, getContextRatio, getContextTone } from '@/src/shared/context-usage';
+import { formatSessionCost } from '@/src/shared/session-cost';
 
 const toneStroke: Record<'danger' | 'safe' | 'warn', string> = {
   danger: '#f87171',
@@ -16,18 +17,21 @@ export const ContextDonut = ({
   contextLength,
   onCompact,
   promptTokens,
+  sessionCostUsd,
 }: {
   canCompact: boolean;
   contextLength: number | undefined;
   onCompact: () => void;
   promptTokens: number;
+  sessionCostUsd: number;
 }): JSX.Element => {
   const detailsRef = useRef<HTMLDetailsElement>(null);
   const ratio = getContextRatio(promptTokens, contextLength);
   const stroke = ratio === undefined ? '#52525b' : toneStroke[getContextTone(ratio)];
   const dash = ratio === undefined ? 0 : ratio * CIRCUMFERENCE;
   const summary = formatContextSummary(promptTokens, contextLength);
   const label = `Context usage: ${summary}`;
+  const sessionCostLabel = formatSessionCost(sessionCostUsd);
 
   return (
     <details className="relative shrink-0" ref={detailsRef}>
@@ -54,6 +58,7 @@ export const ContextDonut = ({
       <div className="absolute bottom-10 right-0 z-20 w-56 rounded-md border border-zinc-800 bg-zinc-950 p-3 text-xs text-zinc-300 shadow-lg shadow-zinc-950/60">
         <p className="font-medium text-zinc-100">Context</p>
         <p className="mt-1 text-zinc-400">{summary}</p>
+        <p className="mt-1 text-zinc-400">Session cost {sessionCostLabel}</p>
         <button
           className="mt-3 h-7 w-full rounded-md bg-[#EDFF00] px-2 text-xs font-semibold text-zinc-950 outline-none transition hover:bg-[#d9ea00] focus:ring-2 focus:ring-[#EDFF00]/50 disabled:cursor-not-allowed disabled:bg-zinc-800 disabled:text-zinc-500"
           disabled={!canCompact}

apps/extension/src/shared/agent-chat-atoms.test.ts

@@ -9,37 +9,49 @@ import {
   draftAtomFamily,
   evictConversationAtoms,
   runningConversationIdsAtom,
+  sessionCostAtomFamily,
   streamingMessageIdAtomFamily,
 } from '@/entrypoints/sidepanel/agent-chat-atoms';
 
 describe('per-conversation atom eviction', () => {
-  it('evictConversationAtoms resets draft, usage, and streaming id for a conversation id', () => {
+  it('evictConversationAtoms resets draft, usage, session cost, and streaming id for a conversation id', () => {
     const store = getDefaultStore();
     store.set(draftAtomFamily('conversation-1'), 'hello');
     store.set(contextUsageAtomFamily('conversation-1'), { promptTokens: 42 });
+    store.set(sessionCostAtomFamily('conversation-1'), 0.0123);
     store.set(streamingMessageIdAtomFamily('conversation-1'), 'msg-streaming');
 
     evictConversationAtoms('conversation-1');
 
     // A fresh atom (post-remove) starts from its initial value.
     expect(store.get(draftAtomFamily('conversation-1'))).toBe('');
     expect(store.get(contextUsageAtomFamily('conversation-1'))).toBeUndefined();
+    expect(store.get(sessionCostAtomFamily('conversation-1'))).toBe(0);
     expect(store.get(streamingMessageIdAtomFamily('conversation-1'))).toBeUndefined();
   });
 
-  it('clearPerConversationAtoms wipes all drafts, usage, streaming ids, and run-state on sign-out', () => {
+  it('clearPerConversationAtoms wipes all drafts, usage, session cost, and streaming ids on sign-out', () => {
     const store = getDefaultStore();
     store.set(draftAtomFamily('conversation-1'), 'prev account draft');
     store.set(contextUsageAtomFamily('conversation-2'), { promptTokens: 999 });
+    store.set(sessionCostAtomFamily('conversation-3'), 1.5);
     store.set(streamingMessageIdAtomFamily('conversation-3'), 'msg-only-streaming');
-    store.set(runningConversationIdsAtom, ['conversation-1']);
-    store.set(compactingConversationIdsAtom, ['conversation-2']);
 
     clearPerConversationAtoms();
 
     expect(store.get(draftAtomFamily('conversation-1'))).toBe('');
     expect(store.get(contextUsageAtomFamily('conversation-2'))).toBeUndefined();
+    expect(store.get(sessionCostAtomFamily('conversation-3'))).toBe(0);
     expect(store.get(streamingMessageIdAtomFamily('conversation-3'))).toBeUndefined();
+  });
+
+  it('clearPerConversationAtoms clears run-state on sign-out', () => {
+    const store = getDefaultStore();
+    store.set(runningConversationIdsAtom, ['conversation-1']);
+    store.set(compactingConversationIdsAtom, ['conversation-2']);
+
+    clearPerConversationAtoms();
+
     expect(store.get(runningConversationIdsAtom)).toStrictEqual([]);
     expect(store.get(compactingConversationIdsAtom)).toStrictEqual([]);
   });

apps/extension/src/shared/agent-llm-turn-runner-core.test.ts

@@ -14,10 +14,12 @@ function* createGatewayResponses(): Generator<Response, Response> {
   yield streamResponse([
     'data: {"choices":[{"delta":{"content":"Reading"}}]}\n\n',
     'data: {"choices":[{"delta":{"tool_calls":[{"index":0,"id":"call_snapshot","type":"function","function":{"name":"get_page_snapshot","arguments":"{}"}}]}}]}\n\n',
+    'data: {"choices":[],"usage":{"prompt_tokens":100,"completion_tokens":5,"total_tokens":105,"cost":0.0007}}\n\n',
     'data: [DONE]\n\n',
   ]);
   yield streamResponse([
     'data: {"choices":[{"delta":{"content":"Done."}}]}\n\n',
+    'data: {"choices":[],"usage":{"prompt_tokens":200,"completion_tokens":3,"total_tokens":203,"cost":0.001}}\n\n',
     'data: [DONE]\n\n',
   ]);
   return streamResponse(['data: [DONE]\n\n']);
@@ -63,7 +65,7 @@ describe('agent LLM turn runner core', () => {
     const fetch: FetchLike = () =>
       streamResponse([
         'data: {"choices":[{"delta":{"content":"Done."}}]}\n\n',
-        'data: {"choices":[],"usage":{"completion_tokens":5,"prompt_tokens":999,"total_tokens":1004}}\n\n',
+        'data: {"choices":[],"usage":{"completion_tokens":5,"prompt_tokens":999,"total_tokens":1004,"cost":0.0123}}\n\n',
         'data: [DONE]\n\n',
       ]);
 
@@ -88,13 +90,15 @@ describe('agent LLM turn runner core', () => {
     });
 
     expect(usageCalls).toContainEqual({
+      costUsd: 0.0123,
       promptTokens: 999,
     });
   });
 
   it('streams, runs tools, and continues with tool results', async () => {
     const appendedEvents: AgentConversationEvent[] = [];
     const updatedMessages: string[] = [];
+    const usageCalls: unknown[] = [];
     const fetchCalls: unknown[] = [];
     const responses = createGatewayResponses();
     const fetch: FetchLike = (_input, init) => {
@@ -115,6 +119,7 @@ describe('agent LLM turn runner core', () => {
       maxToolRounds: 4,
       model: 'anthropic/claude-sonnet-4',
       noResponseMessage: 'The model did not return a response.',
+      onUsage: usage => usageCalls.push(usage),
       signal: undefined,
       toToolCallEvents: (toolCalls: KiloGatewayToolCallRequest[]) =>
         toolCalls.map(toolCall =>
@@ -152,6 +157,10 @@ describe('agent LLM turn runner core', () => {
       { role: 'assistant', text: 'Done.', type: 'message' },
     ]);
     expect(fetchCalls).toHaveLength(2);
+    expect(usageCalls).toStrictEqual([
+      { costUsd: 0.0007, promptTokens: 100 },
+      { costUsd: 0.001, promptTokens: 200 },
+    ]);
   });
 
   it('allows twenty tool rounds before asking the user to continue', async () => {

apps/extension/src/shared/agent-llm-turn-runner-core.ts

@@ -10,6 +10,7 @@ import type { EvalTabResult } from './tab-debugger';
 type ToolCallEvent = Extract<AgentConversationEvent, { readonly type: 'tool-call' }>;
 
 export interface TurnUsage {
+  readonly costUsd?: number;
   readonly promptTokens: number;
 }
 

apps/extension/src/shared/kilo-gateway-chat-client.ts

@@ -56,6 +56,7 @@ export interface KiloGatewayChatCompletion {
   readonly reasoningDetails?: readonly unknown[];
   readonly toolCalls: KiloGatewayToolCallRequest[];
   readonly usage?: {
+    readonly costUsd?: number;
     readonly promptTokens: number;
   };
 }

apps/extension/src/shared/kilo-gateway-chat-stream-client.test.ts

@@ -583,4 +583,49 @@ describe('kilo gateway chat stream client', () => {
       promptTokens: 1200,
     });
   });
+
+  it('carries costUsd when the usage chunk includes cost', () => {
+    const sse = [
+      'data: {"choices":[{"delta":{"content":"hi"}}]}\n\n',
+      'data: {"choices":[],"usage":{"prompt_tokens":1200,"completion_tokens":34,"total_tokens":1234,"cost":0.0123}}\n\n',
+      'data: [DONE]\n\n',
+    ].join('');
+
+    const completion = parseKiloGatewayChatCompletionStream(sse, () => {});
+
+    expect(completion.usage).toStrictEqual({
+      costUsd: 0.0123,
+      promptTokens: 1200,
+    });
+  });
+
+  it('omits costUsd when the usage chunk has no cost field', () => {
+    const sse = [
+      'data: {"choices":[{"delta":{"content":"hi"}}]}\n\n',
+      'data: {"choices":[],"usage":{"prompt_tokens":800,"completion_tokens":10,"total_tokens":810}}\n\n',
+      'data: [DONE]\n\n',
+    ].join('');
+
+    const completion = parseKiloGatewayChatCompletionStream(sse, () => {});
+
+    expect(completion.usage).toStrictEqual({
+      promptTokens: 800,
+    });
+    expect(completion.usage).not.toHaveProperty('costUsd');
+  });
+
+  it('parses prompt_tokens when cost is null and omits costUsd', () => {
+    const sse = [
+      'data: {"choices":[{"delta":{"content":"hi"}}]}\n\n',
+      'data: {"choices":[],"usage":{"prompt_tokens":500,"completion_tokens":2,"total_tokens":502,"cost":null}}\n\n',
+      'data: [DONE]\n\n',
+    ].join('');
+
+    const completion = parseKiloGatewayChatCompletionStream(sse, () => {});
+
+    expect(completion.usage).toStrictEqual({
+      promptTokens: 500,
+    });
+    expect(completion.usage).not.toHaveProperty('costUsd');
+  });
 });

apps/extension/src/shared/kilo-gateway-chat-stream-client.ts

@@ -87,8 +87,9 @@ const streamingToolCallDeltaSchema = z.object({
   id: z.string().optional(),
   index: z.number(),
 });
-// Only prompt_tokens is consumed (the context-usage ratio); other usage fields are ignored.
+// Prompt_tokens drives the context-usage ratio; cost is optional session spend (USD).
 const usageSchema = z.object({
+  cost: z.number().nullish(),
   prompt_tokens: z.number(),
 });
 const streamDataSchema = z.object({
@@ -261,7 +262,11 @@ const applyStreamingData = (
   }
 
   if (parsed.data.usage !== undefined && parsed.data.usage !== null) {
-    accumulator.usage = { promptTokens: parsed.data.usage.prompt_tokens };
+    const { cost, prompt_tokens: promptTokens } = parsed.data.usage;
+    accumulator.usage = {
+      promptTokens,
+      ...(typeof cost === 'number' ? { costUsd: cost } : {}),
+    };
   }
 
   const choice = parsed.data.choices.at(0);

apps/extension/src/shared/session-cost.test.ts

@@ -0,0 +1,31 @@
+import { describe, expect, it } from 'vitest';
+import { addSessionCost, formatSessionCost } from './session-cost';
+
+describe('session cost helpers', () => {
+  it('formats zero and non-positive values as $0.0000', () => {
+    expect(formatSessionCost(0)).toBe('$0.0000');
+    expect(formatSessionCost(-0.01)).toBe('$0.0000');
+    expect(formatSessionCost(Number.NaN)).toBe('$0.0000');
+    expect(formatSessionCost(Number.POSITIVE_INFINITY)).toBe('$0.0000');
+  });
+
+  it('formats positive costs to four decimal places', () => {
+    expect(formatSessionCost(0.0123)).toBe('$0.0123');
+    expect(formatSessionCost(1.234)).toBe('$1.2340');
+    expect(formatSessionCost(5e-5)).toBe('$0.0001');
+  });
+
+  it('adds finite non-negative costs', () => {
+    expect(addSessionCost(0.0123, 0.0007)).toBeCloseTo(0.013);
+    expect(addSessionCost(0.013, 0.001)).toBeCloseTo(0.014);
+    expect(addSessionCost(0, 0)).toBe(0);
+  });
+
+  it('leaves the previous total unchanged for missing, non-finite, or negative cost', () => {
+    const usageWithoutCost: { costUsd?: number } = {};
+    expect(addSessionCost(0.5, usageWithoutCost.costUsd)).toBe(0.5);
+    expect(addSessionCost(0.5, Number.NaN)).toBe(0.5);
+    expect(addSessionCost(0.5, Number.POSITIVE_INFINITY)).toBe(0.5);
+    expect(addSessionCost(0.5, -0.01)).toBe(0.5);
+  });
+});

apps/extension/src/shared/session-cost.ts

@@ -0,0 +1,20 @@
+/** Format session spend like mobile `formatCost`: always `$X.XXXX`. */
+export const formatSessionCost = (cost: number): string => {
+  if (!Number.isFinite(cost) || cost <= 0) {
+    return '$0.0000';
+  }
+
+  return `$${cost.toFixed(4)}`;
+};
+
+/**
+ * Add a completion's USD cost to a running session total.
+ * Missing, non-finite, or negative values leave the previous total unchanged.
+ */
+export const addSessionCost = (previous: number, costUsd: number | undefined): number => {
+  if (costUsd === undefined || !Number.isFinite(costUsd) || costUsd < 0) {
+    return previous;
+  }
+
+  return previous + costUsd;
+};