feat(agent-conversation): Show aggregate metrics and costs

packages/cli/src/commands/agent-conversation/view.ts

@@ -136,15 +136,18 @@ export const viewCommand = buildCommand({
     }
     const org = resolved.org;
 
-    const { spans, truncated, title } = await withProgress(
+    const { spans, stats, truncated, title } = await withProgress(
       {
         message: "Fetching conversation spans...",
         json: flags.json,
       },
       () => getConversationSpans(org, conversationId),
     );
 
-    const result = buildTranscriptResult(conversationId, org, spans, title);
+    const result = buildTranscriptResult(conversationId, org, spans, {
+      stats,
+      title,
+    });
     result.truncated = truncated;
     yield new CommandOutput<TranscriptResult>(result);
   },

packages/cli/src/lib/api/conversations.ts

@@ -20,6 +20,7 @@ import {
   type AgentConversationSpan,
   type ConversationListItem,
   ConversationListItemSchema,
+  type ConversationStats,
 } from "../../types/conversation.js";
 
 import { logger } from "../logger.js";
@@ -126,6 +127,7 @@ export async function getConversationSpans(
   } = {},
 ): Promise<{
   spans: AgentConversationSpan[];
+  stats?: ConversationStats;
   truncated: boolean;
   title: string | null;
 }> {
@@ -140,6 +142,7 @@ export async function getConversationSpans(
   }
 
   const spans: AgentConversationSpan[] = [];
+  let stats: ConversationStats | undefined;
   let title: string | null = null;
   let cursor: string | undefined;
 
@@ -156,6 +159,7 @@ export async function getConversationSpans(
       );
 
     if (page === 0) {
+      stats = data.stats;
       title = data.title;
     }
     spans.push(...data.spans);
@@ -173,5 +177,5 @@ export async function getConversationSpans(
     );
   }
 
-  return { spans, truncated, title };
+  return { spans, stats, truncated, title };
 }

packages/cli/src/lib/formatters/conversation.ts

@@ -8,6 +8,7 @@
 import type {
   AgentConversationSpan,
   ConversationListItem,
+  ConversationStats,
 } from "../../types/conversation.js";
 import {
   colorTag,
@@ -37,10 +38,20 @@ function formatTimestamp(epochSeconds: number): string {
   return new Date(epochSeconds * 1000).toLocaleString();
 }
 
+const USD_FORMAT = new Intl.NumberFormat("en-US", {
+  style: "currency",
+  currency: "USD",
+  maximumFractionDigits: 6,
+});
+
+function formatCost(cost: number): string {
+  return USD_FORMAT.format(cost);
+}
+
 /**
  * Minimum terminal width (columns) to show the full conversation table.
  * Below this, Tools/Errs/User are dropped so the core ID/Title/Started/
- * Tokens/First Input columns don't truncate aggressively.
+ * Tokens/Cost/First Input columns don't truncate aggressively.
  */
 const WIDE_TABLE_MIN_TERM_WIDTH = 100;
 
@@ -71,6 +82,12 @@ const TOKENS_COLUMN: Column<ConversationListItem> = {
   align: "right",
 };
 
+const COST_COLUMN: Column<ConversationListItem> = {
+  header: "Cost",
+  value: (c) => formatCost(c.totalCost),
+  align: "right",
+};
+
 const TOOLS_COLUMN: Column<ConversationListItem> = {
   header: "Tools",
   value: (c) => String(c.toolCalls),
@@ -109,6 +126,7 @@ function selectConversationColumns(): Column<ConversationListItem>[] {
       TITLE_COLUMN,
       STARTED_COLUMN,
       TOKENS_COLUMN,
+      COST_COLUMN,
       TOOLS_COLUMN,
       ERRS_COLUMN,
       USER_COLUMN,
@@ -120,6 +138,7 @@ function selectConversationColumns(): Column<ConversationListItem>[] {
     TITLE_COLUMN,
     STARTED_COLUMN,
     TOKENS_COLUMN,
+    COST_COLUMN,
     FIRST_INPUT_COLUMN,
   ];
 }
@@ -440,7 +459,13 @@ export type TranscriptResult = {
   /** Stored conversation title, or null when none has been titled yet. */
   title: string | null;
   turns: ConversationTurn[];
+  errors?: number;
+  llmCalls?: number;
+  toolCalls?: number;
+  toolErrors?: number;
+  totalCost?: number;
   totalTokens: number;
+  traceIds: string[];
   spanCount: number;
   projects: string[];
   startTimestamp: number;
@@ -468,6 +493,21 @@ export function formatTranscriptResult(result: TranscriptResult): string {
     ["Spans", String(result.spanCount)],
     ["Tokens", String(result.totalTokens)],
   ];
+  if (result.totalCost !== undefined) {
+    rows.push(["Cost", formatCost(result.totalCost)]);
+  }
+  if (result.llmCalls !== undefined) {
+    rows.push(["LLM Calls", String(result.llmCalls)]);
+  }
+  if (result.toolCalls !== undefined) {
+    rows.push(["Tool Calls", String(result.toolCalls)]);
+  }
+  if (result.errors !== undefined) {
+    rows.push(["Errors", String(result.errors)]);
+  }
+  if (result.toolErrors !== undefined) {
+    rows.push(["Tool Errors", String(result.toolErrors)]);
+  }
   if (result.title) {
     rows.unshift(["Title", escapeMarkdownCell(result.title)]);
   }
@@ -494,17 +534,30 @@ export function buildTranscriptResult(
   conversationId: string,
   org: string,
   spans: AgentConversationSpan[],
-  title: string | null = null,
+  details: { stats?: ConversationStats; title?: string | null } = {},
 ): TranscriptResult {
+  const { stats, title = null } = details;
   const turns = extractTurns(spans);
+  const summary = stats
+    ? {
+        errors: stats.errors,
+        llmCalls: stats.llmCalls,
+        toolCalls: stats.toolCalls,
+        toolErrors: stats.toolErrors,
+        totalCost: stats.totalCost,
+      }
+    : {};
   return {
     conversationId,
     org,
     title,
     turns,
-    // Sum tokens per turn (ai_client spans only). Summing every span would
-    // double-count parent invoke_agent spans that aggregate child usage.
-    totalTokens: turns.reduce((sum, t) => sum + t.totalTokens, 0),
+    ...summary,
+    // Prefer server totals. They cover the whole conversation without
+    // double-counting parent agent spans.
+    totalTokens:
+      stats?.totalTokens ?? turns.reduce((sum, t) => sum + t.totalTokens, 0),
+    traceIds: [...new Set(spans.map((s) => s.trace))].sort(),
     spanCount: spans.length,
     projects: [...new Set(spans.map((s) => s.project))].sort(),
     startTimestamp:

packages/cli/src/types/conversation.ts

@@ -96,6 +96,17 @@ export type AgentConversationSpan = InferOutput<
   typeof AgentConversationSpanSchema
 >;
 
+export const ConversationStatsSchema = looseObject({
+  errors: number(),
+  llmCalls: number(),
+  toolCalls: number(),
+  toolErrors: number(),
+  totalCost: number(),
+  totalTokens: number(),
+});
+
+export type ConversationStats = InferOutput<typeof ConversationStatsSchema>;
+
 /**
  * Conversation details envelope returned by
  * `GET /organizations/{org}/agents/conversations/{conversationId}/`.
@@ -109,6 +120,8 @@ export const AgentConversationDetailsSchema = looseObject({
   conversationId: string(),
   title: nullable(string()),
   spans: array(AgentConversationSpanSchema),
+  /** Added after the original envelope; optional for older self-hosted servers. */
+  stats: optional(ConversationStatsSchema),
 });
 
 export type AgentConversationDetails = InferOutput<

packages/cli/src/types/index.ts

@@ -24,11 +24,13 @@ export type {
   AgentConversationDetails,
   AgentConversationSpan,
   ConversationListItem,
+  ConversationStats,
 } from "./conversation.js";
 export {
   AgentConversationDetailsSchema,
   AgentConversationSpanSchema,
   ConversationListItemSchema,
+  ConversationStatsSchema,
 } from "./conversation.js";
 // Dashboard types
 export type {

packages/cli/test/commands/agent-conversation/view.test.ts

@@ -318,6 +318,14 @@ describe("viewCommand.func", () => {
   test("yields CommandOutput with transcript result (JSON)", async () => {
     getConversationSpansSpy.mockResolvedValue({
       spans: sampleSpans,
+      stats: {
+        errors: 1,
+        llmCalls: 2,
+        toolCalls: 3,
+        toolErrors: 1,
+        totalCost: 0.0042,
+        totalTokens: 150,
+      },
       truncated: false,
       title: null,
     });
@@ -330,7 +338,16 @@ describe("viewCommand.func", () => {
     const parsed = JSON.parse(output);
     expect(parsed.conversationId).toBe(CONVERSATION_ID);
     expect(parsed.org).toBe(ORG);
-    expect(parsed.spanCount).toBe(2);
+    expect(parsed).toMatchObject({
+      errors: 1,
+      llmCalls: 2,
+      toolCalls: 3,
+      toolErrors: 1,
+      totalCost: 0.0042,
+      totalTokens: 150,
+      traceIds: ["00112233445566778899aabbccddeeff"],
+      spanCount: 2,
+    });
     expect(parsed.turns).toBeDefined();
     expect(Array.isArray(parsed.turns)).toBe(true);
   });

packages/cli/test/lib/api/conversations.test.ts

@@ -19,6 +19,7 @@ import {
   listConversations,
 } from "../../../src/lib/api/conversations.js";
 import { MAX_PAGINATION_PAGES } from "../../../src/lib/api/infrastructure.js";
+import type { ConversationStats } from "../../../src/types/conversation.js";
 import { mockFetch, useTestConfigDir } from "../../helpers.js";
 
 // ============================================================================
@@ -311,11 +312,13 @@ describe("getConversationSpans", () => {
   function makeEnvelope(
     spans: ReturnType<typeof makeSpan>[],
     title: string | null = "Test conversation",
+    stats?: ConversationStats,
   ) {
     return {
       conversationId: CONV_ID,
       title,
       spans,
+      ...(stats ? { stats } : {}),
     };
   }
 
@@ -340,15 +343,26 @@ describe("getConversationSpans", () => {
     expect(url).toContain("statsPeriod=30d");
   });
 
-  test("returns spans from a single page envelope", async () => {
+  test("returns spans and stats from a single page envelope", async () => {
     const spans = [makeSpan("span-1-aabb1122")];
-    mockOk(makeEnvelope(spans, "Refund flow"));
+    mockOk(
+      makeEnvelope(spans, "Refund flow", {
+        errors: 1,
+        llmCalls: 1,
+        toolCalls: 2,
+        toolErrors: 1,
+        totalCost: 0.0042,
+        totalTokens: 100,
+      }),
+    );
 
     const result = await getConversationSpans(ORG, CONV_ID);
 
     expect(result.spans).toHaveLength(1);
     expect(result.spans[0].span_id).toBe("span-1-aabb1122");
     expect(result.title).toBe("Refund flow");
+    expect(result.stats?.totalCost).toBe(0.0042);
+    expect(result.stats?.totalTokens).toBe(100);
     expect(result.truncated).toBe(false);
   });