@amodalai/runtime 0.1.14 → 0.1.16
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/src/agent/agent-runner.d.ts +20 -0
- package/dist/src/agent/agent-runner.js +165 -1
- package/dist/src/agent/agent-runner.js.map +1 -1
- package/dist/src/agent/config-watcher.d.ts +2 -2
- package/dist/src/agent/config-watcher.js +28 -15
- package/dist/src/agent/config-watcher.js.map +1 -1
- package/dist/src/agent/config-watcher.test.js +13 -4
- package/dist/src/agent/config-watcher.test.js.map +1 -1
- package/dist/src/agent/eval-store.d.ts +50 -0
- package/dist/src/agent/eval-store.js +137 -0
- package/dist/src/agent/eval-store.js.map +1 -0
- package/dist/src/agent/local-server.d.ts +1 -1
- package/dist/src/agent/local-server.js +44 -26
- package/dist/src/agent/local-server.js.map +1 -1
- package/dist/src/agent/local-server.test.js +2 -1
- package/dist/src/agent/local-server.test.js.map +1 -1
- package/dist/src/agent/proactive/proactive-runner.d.ts +3 -3
- package/dist/src/agent/proactive/proactive-runner.js +3 -3
- package/dist/src/agent/proactive/proactive-runner.js.map +1 -1
- package/dist/src/agent/proactive/proactive-runner.test.js +2 -2
- package/dist/src/agent/proactive/proactive-runner.test.js.map +1 -1
- package/dist/src/agent/routes/admin-chat.d.ts +2 -2
- package/dist/src/agent/routes/admin-chat.js +2 -2
- package/dist/src/agent/routes/admin-chat.js.map +1 -1
- package/dist/src/agent/routes/chat.d.ts +6 -6
- package/dist/src/agent/routes/chat.js +3 -3
- package/dist/src/agent/routes/chat.js.map +1 -1
- package/dist/src/agent/routes/chat.test.js +7 -6
- package/dist/src/agent/routes/chat.test.js.map +1 -1
- package/dist/src/agent/routes/evals.d.ts +16 -0
- package/dist/src/agent/routes/evals.js +383 -0
- package/dist/src/agent/routes/evals.js.map +1 -0
- package/dist/src/agent/routes/inspect.d.ts +2 -2
- package/dist/src/agent/routes/inspect.js +11 -22
- package/dist/src/agent/routes/inspect.js.map +1 -1
- package/dist/src/agent/routes/inspect.test.js +11 -16
- package/dist/src/agent/routes/inspect.test.js.map +1 -1
- package/dist/src/agent/routes/task.d.ts +2 -2
- package/dist/src/agent/routes/task.js +4 -4
- package/dist/src/agent/routes/task.js.map +1 -1
- package/dist/src/agent/routes/task.test.js.map +1 -1
- package/dist/src/agent/session-store.d.ts +2 -2
- package/dist/src/agent/session-store.js +9 -6
- package/dist/src/agent/session-store.js.map +1 -1
- package/dist/src/agent/snapshot-server.js +10 -2
- package/dist/src/agent/snapshot-server.js.map +1 -1
- package/dist/src/agent/{agent-runner.test.d.ts → write-repo-file.test.d.ts} +1 -1
- package/dist/src/agent/write-repo-file.test.js +270 -0
- package/dist/src/agent/write-repo-file.test.js.map +1 -0
- package/dist/src/index.d.ts +4 -1
- package/dist/src/index.js +3 -1
- package/dist/src/index.js.map +1 -1
- package/dist/src/routes/chat-stream.js +2 -1
- package/dist/src/routes/chat-stream.js.map +1 -1
- package/dist/src/server.js +0 -2
- package/dist/src/server.js.map +1 -1
- package/dist/src/server.test.js +37 -6
- package/dist/src/server.test.js.map +1 -1
- package/dist/src/session/custom-tool-adapter.d.ts +74 -0
- package/dist/src/session/custom-tool-adapter.js +149 -0
- package/dist/src/session/custom-tool-adapter.js.map +1 -0
- package/dist/src/session/session-manager.d.ts +67 -3
- package/dist/src/session/session-manager.js +313 -7
- package/dist/src/session/session-manager.js.map +1 -1
- package/dist/src/session/session-manager.test.js +63 -6
- package/dist/src/session/session-manager.test.js.map +1 -1
- package/dist/src/session/session-runner.js +12 -0
- package/dist/src/session/session-runner.js.map +1 -1
- package/dist/src/types.d.ts +1 -0
- package/dist/tsconfig.tsbuildinfo +1 -1
- package/package.json +2 -2
- package/dist/src/agent/agent-runner.test.js +0 -552
- package/dist/src/agent/agent-runner.test.js.map +0 -1
- package/dist/src/agent/session-manager.d.ts +0 -88
- package/dist/src/agent/session-manager.js +0 -341
- package/dist/src/agent/session-manager.js.map +0 -1
- package/dist/src/agent/session-manager.test.d.ts +0 -6
- package/dist/src/agent/session-manager.test.js +0 -145
- package/dist/src/agent/session-manager.test.js.map +0 -1
|
@@ -5,26 +5,26 @@
|
|
|
5
5
|
*/
|
|
6
6
|
import { Router } from 'express';
|
|
7
7
|
import type { Request, Response } from 'express';
|
|
8
|
-
import type {
|
|
9
|
-
import type {
|
|
8
|
+
import type { ManagedSession } from '../../session/session-manager.js';
|
|
9
|
+
import type { SessionManager } from '../../session/session-manager.js';
|
|
10
10
|
/**
|
|
11
11
|
* Optional hook for customizing session creation.
|
|
12
12
|
* When provided, called instead of sessionManager.create().
|
|
13
13
|
* Receives the Express request and response (for auth context access).
|
|
14
14
|
*/
|
|
15
|
-
export type SessionCreator = (req: Request, res: Response, appId: string, appToken?: string) => Promise<
|
|
15
|
+
export type SessionCreator = (req: Request, res: Response, appId: string, appToken?: string) => Promise<ManagedSession>;
|
|
16
16
|
/**
|
|
17
17
|
* Optional hook for hydrating a session from external storage.
|
|
18
18
|
* Called when session_id is provided but not found in memory.
|
|
19
19
|
*/
|
|
20
|
-
export type SessionHydrator = (req: Request, res: Response, sessionId: string, appId: string) => Promise<
|
|
20
|
+
export type SessionHydrator = (req: Request, res: Response, sessionId: string, appId: string) => Promise<ManagedSession | null>;
|
|
21
21
|
/**
|
|
22
22
|
* Optional hook called after each agent turn completes.
|
|
23
23
|
* Used by hosted runtime to persist conversation history.
|
|
24
24
|
*/
|
|
25
|
-
export type TurnCompleteHandler = (session:
|
|
25
|
+
export type TurnCompleteHandler = (session: ManagedSession, req: Request, res: Response) => void;
|
|
26
26
|
export interface ChatRouterOptions {
|
|
27
|
-
sessionManager:
|
|
27
|
+
sessionManager: SessionManager;
|
|
28
28
|
/** Optional session creator hook — used by hosted runtime to inject app config from platform API */
|
|
29
29
|
sessionCreator?: SessionCreator;
|
|
30
30
|
/** Optional session hydrator — called when session_id not found in memory (e.g., load from platform API) */
|
|
@@ -5,7 +5,7 @@
|
|
|
5
5
|
*/
|
|
6
6
|
import { Router } from 'express';
|
|
7
7
|
import { AgentChatRequestSchema } from '../agent-types.js';
|
|
8
|
-
import {
|
|
8
|
+
import { streamMessage } from '../../session/session-runner.js';
|
|
9
9
|
import { SSEEventType } from '../../types.js';
|
|
10
10
|
export function createChatRouter(options) {
|
|
11
11
|
const router = Router();
|
|
@@ -42,7 +42,7 @@ export function createChatRouter(options) {
|
|
|
42
42
|
try {
|
|
43
43
|
session = options.sessionCreator
|
|
44
44
|
? await options.sessionCreator(req, res, app_id, app_token)
|
|
45
|
-
: await options.sessionManager.create(app_id
|
|
45
|
+
: await options.sessionManager.create(app_id);
|
|
46
46
|
}
|
|
47
47
|
catch (err) {
|
|
48
48
|
const errMsg = err instanceof Error ? err.message : String(err);
|
|
@@ -58,7 +58,7 @@ export function createChatRouter(options) {
|
|
|
58
58
|
const controller = new AbortController();
|
|
59
59
|
res.on('close', () => controller.abort());
|
|
60
60
|
try {
|
|
61
|
-
for await (const event of
|
|
61
|
+
for await (const event of streamMessage(session, message, controller.signal, undefined, options.sessionManager)) {
|
|
62
62
|
if (controller.signal.aborted)
|
|
63
63
|
break;
|
|
64
64
|
writeSSE(res, event);
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"chat.js","sourceRoot":"","sources":["../../../../src/agent/routes/chat.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,EAAC,MAAM,EAAC,MAAM,SAAS,CAAC;AAE/B,OAAO,EAAC,sBAAsB,EAAC,MAAM,mBAAmB,CAAC;AAGzD,OAAO,EAAC,
|
|
1
|
+
{"version":3,"file":"chat.js","sourceRoot":"","sources":["../../../../src/agent/routes/chat.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,EAAC,MAAM,EAAC,MAAM,SAAS,CAAC;AAE/B,OAAO,EAAC,sBAAsB,EAAC,MAAM,mBAAmB,CAAC;AAGzD,OAAO,EAAC,aAAa,EAAC,MAAM,iCAAiC,CAAC;AAC9D,OAAO,EAAC,YAAY,EAAC,MAAM,gBAAgB,CAAC;AAgC5C,MAAM,UAAU,gBAAgB,CAAC,OAA0B;IACzD,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC;IAExB,MAAM,CAAC,IAAI,CAAC,OAAO,EAAE,KAAK,EAAE,GAAY,EAAE,GAAa,EAAE,EAAE;QACzD,MAAM,MAAM,GAAG,sBAAsB,CAAC,SAAS,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QAC1D,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;YACpB,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC;gBACnB,KAAK,EAAE;oBACL,IAAI,EAAE,kBAAkB;oBACxB,OAAO,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO;iBAC9B;aACF,CAAC,CAAC;YACH,OAAO;QACT,CAAC;QAED,MAAM,EAAC,OAAO,EAAE,UAAU,EAAE,MAAM,EAAE,SAAS,EAAC,GAAG,MAAM,CAAC,IAAI,CAAC;QAE7D,cAAc;QACd,GAAG,CAAC,SAAS,CAAC,GAAG,EAAE;YACjB,cAAc,EAAE,mBAAmB;YACnC,eAAe,EAAE,UAAU;YAC3B,YAAY,EAAE,YAAY;SAC3B,CAAC,CAAC;QAEH,wBAAwB;QACxB,IAAI,OAAO,GAAG,UAAU,CAAC,CAAC,CAAC,OAAO,CAAC,cAAc,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;QAE9E,yCAAyC;QACzC,IAAI,CAAC,OAAO,IAAI,UAAU,IAAI,OAAO,CAAC,eAAe,EAAE,CAAC;YACtD,IAAI,CAAC;gBACH,OAAO,GAAG,MAAM,OAAO,CAAC,eAAe,CAAC,GAAG,EAAE,GAAG,EAAE,UAAU,EAAE,MAAM,CAAC,IAAI,SAAS,CAAC;YACrF,CAAC;YAAC,MAAM,CAAC;gBACP,wDAAwD;YAC1D,CAAC;QACH,CAAC;QAED,IAAI,CAAC,OAAO,EAAE,CAAC;YACb,IAAI,CAAC;gBACH,OAAO,GAAG,OAAO,CAAC,cAAc;oBAC9B,CAAC,CAAC,MAAM,OAAO,CAAC,cAAc,CAAC,GAAG,EAAE,GAAG,EAAE,MAAM,EAAE,SAAS,CAAC;oBAC3D,CAAC,CAAC,MAAM,OAAO,CAAC,cAAc,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;YAClD,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,MAAM,MAAM,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;gBAChE,QAAQ,CAAC,GAAG,EAAE,EAAC,IAAI,EAAE,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,EAAE,EAAE,EAAC,CAAC,CAAC;gBAC5E,QAAQ,CAAC,GAAG,EAAE,EAAC,IAAI,EAAE,YAAY,CAAC,IAAI,EAAE,SAAS,EAAE,EAAE,EAAE,EAAC,CAAC,CAAC;gBAC1D,GAAG,CAAC,GAAG,EAAE,CAAC;gBACV,OAAO;YACT,CAAC;QACH,CAAC;QAED,kBAAkB;QAClB,QAAQ,CAAC,GAAG,EAAE,EAAC,IAAI,EAAE,YAAY,CAAC,IAAI,EAAE,UAAU,EAAE,OAAO,CAAC,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE,EAAC,CAAC,CAAC;QAElF,mFAAmF;QACnF,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;QACzC,GAAG,CAAC,EAAE,CAAC,OAAO,EAAE,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,CAAC,CAAC;QAE1C,IAAI,CAAC;YACH,IAAI,KAAK,EAAE,MAAM,KAAK,IAAI,aAAa,CAAC,OAAO,EAAE,OAAO,EAAE,UAAU,CAAC,MAAM,EAAE,SAAS,EAAE,OAAO,CAAC,cAAc,CAAC,EAAE,CAAC;gBAChH,IAAI,UAAU,CAAC,MAAM,CAAC,OAAO;oBAAE,MAAM;gBACrC,QAAQ,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;YACvB,CAAC;QACH,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,IAAI,CAAC,UAAU,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;gBAC/B,MAAM,MAAM,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;gBAChE,QAAQ,CAAC,GAAG,EAAE,EAAC,IAAI,EAAE,YAAY,CAAC,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,EAAE,EAAE,EAAC,CAAC,CAAC;gBAC5E,QAAQ,CAAC,GAAG,EAAE,EAAC,IAAI,EAAE,YAAY,CAAC,IAAI,EAAE,SAAS,EAAE,EAAE,EAAE,EAAC,CAAC,CAAC;YAC5D,CAAC;QACH,CAAC;QAED,sDAAsD;QACtD,IAAI,OAAO,CAAC,cAAc,EAAE,CAAC;YAC3B,OAAO,CAAC,cAAc,CAAC,OAAO,EAAE,GAAG,EAAE,GAAG,CAAC,CAAC;QAC5C,CAAC;QAED,GAAG,CAAC,GAAG,EAAE,CAAC;IACZ,CAAC,CAAC,CAAC;IAEH,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,SAAS,QAAQ,CAAC,GAAa,EAAE,KAAe;IAC9C,GAAG,CAAC,KAAK,CAAC,SAAS,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC;AAClD,CAAC;AAED,SAAS,EAAE;IACT,OAAO,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;AAClC,CAAC"}
|
|
@@ -7,9 +7,9 @@ import { describe, it, expect, vi, beforeEach } from 'vitest';
|
|
|
7
7
|
import express from 'express';
|
|
8
8
|
import request from 'supertest';
|
|
9
9
|
import { createChatRouter } from './chat.js';
|
|
10
|
-
// Mock
|
|
11
|
-
vi.mock('
|
|
12
|
-
|
|
10
|
+
// Mock session-runner
|
|
11
|
+
vi.mock('../../session/session-runner.js', () => ({
|
|
12
|
+
streamMessage: vi.fn(async function* () {
|
|
13
13
|
yield { type: 'text_delta', content: 'Hello!', timestamp: new Date().toISOString() };
|
|
14
14
|
yield { type: 'done', timestamp: new Date().toISOString() };
|
|
15
15
|
}),
|
|
@@ -75,7 +75,7 @@ describe('repo-chat route', () => {
|
|
|
75
75
|
await request(app)
|
|
76
76
|
.post('/chat')
|
|
77
77
|
.send({ message: 'hello', app_id: 'tenant-1' });
|
|
78
|
-
expect(sessionManager.create).toHaveBeenCalledWith('tenant-1'
|
|
78
|
+
expect(sessionManager.create).toHaveBeenCalledWith('tenant-1');
|
|
79
79
|
});
|
|
80
80
|
it('should reuse existing session if session_id provided', async () => {
|
|
81
81
|
const app = createTestApp(sessionManager);
|
|
@@ -104,12 +104,13 @@ describe('repo-chat route', () => {
|
|
|
104
104
|
.send({ message: 'hello', app_id: 'tenant-1' });
|
|
105
105
|
expect(res.text).toContain('"type":"error"');
|
|
106
106
|
});
|
|
107
|
-
it('should
|
|
107
|
+
it('should create session with app_id as role', async () => {
|
|
108
108
|
const app = createTestApp(sessionManager);
|
|
109
109
|
await request(app)
|
|
110
110
|
.post('/chat')
|
|
111
111
|
.send({ message: 'hello', app_id: 'tenant-1', app_token: 'tok-123' });
|
|
112
|
-
|
|
112
|
+
// app_id is passed as role to SessionManager.create()
|
|
113
|
+
expect(sessionManager.create).toHaveBeenCalledWith('tenant-1');
|
|
113
114
|
});
|
|
114
115
|
});
|
|
115
116
|
//# sourceMappingURL=chat.test.js.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"chat.test.js","sourceRoot":"","sources":["../../../../src/agent/routes/chat.test.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,EAAC,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE,UAAU,EAAC,MAAM,QAAQ,CAAC;AAC5D,OAAO,OAAO,MAAM,SAAS,CAAC;AAC9B,OAAO,OAAO,MAAM,WAAW,CAAC;AAChC,OAAO,EAAC,gBAAgB,EAAC,MAAM,WAAW,CAAC;AAG3C,
|
|
1
|
+
{"version":3,"file":"chat.test.js","sourceRoot":"","sources":["../../../../src/agent/routes/chat.test.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,EAAC,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE,UAAU,EAAC,MAAM,QAAQ,CAAC;AAC5D,OAAO,OAAO,MAAM,SAAS,CAAC;AAC9B,OAAO,OAAO,MAAM,WAAW,CAAC;AAChC,OAAO,EAAC,gBAAgB,EAAC,MAAM,WAAW,CAAC;AAG3C,sBAAsB;AACtB,EAAE,CAAC,IAAI,CAAC,iCAAiC,EAAE,GAAG,EAAE,CAAC,CAAC;IAChD,aAAa,EAAE,EAAE,CAAC,EAAE,CAAC,KAAK,SAAS,CAAC;QAClC,MAAM,EAAC,IAAI,EAAE,YAAY,EAAE,OAAO,EAAE,QAAQ,EAAE,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,EAAC,CAAC;QACnF,MAAM,EAAC,IAAI,EAAE,MAAM,EAAE,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,EAAC,CAAC;IAC5D,CAAC,CAAC;CACH,CAAC,CAAC,CAAC;AAEJ,SAAS,kBAAkB;IACzB,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAE,CAAC;IAE3B,OAAO;QACL,IAAI,EAAE,CAAC;QACP,MAAM,EAAE,EAAE,CAAC,EAAE,CAAC,KAAK,EAAE,KAAa,EAAE,EAAE;YACpC,MAAM,OAAO,GAAG;gBACd,EAAE,EAAE,WAAW;gBACf,KAAK;gBACL,OAAO,EAAE,EAAC,eAAe,EAAE,EAAC,YAAY,EAAE,MAAM,EAAC,EAAC;gBAClD,mBAAmB,EAAE,EAAE;gBACvB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;gBACrB,cAAc,EAAE,IAAI,CAAC,GAAG,EAAE;gBAC1B,eAAe,EAAE,EAAE;gBACnB,aAAa,EAAE,EAAE;aAClB,CAAC;YACF,QAAQ,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,EAAE,OAAO,CAAC,CAAC;YAClC,OAAO,OAAO,CAAC;QACjB,CAAC,CAAC;QACF,GAAG,EAAE,EAAE,CAAC,EAAE,CAAC,CAAC,EAAU,EAAE,EAAE,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QAC5C,OAAO,EAAE,EAAE,CAAC,EAAE,EAAE;QAChB,OAAO,EAAE,EAAE,CAAC,EAAE,EAAE;QAChB,UAAU,EAAE,EAAE,CAAC,EAAE,EAAE;QACnB,QAAQ,EAAE,EAAE,CAAC,EAAE,EAAE;KAEW,CAAC;AACjC,CAAC;AAED,SAAS,aAAa,CAAC,cAA8B;IACnD,MAAM,GAAG,GAAG,OAAO,EAAE,CAAC;IACtB,GAAG,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;IACxB,GAAG,CAAC,GAAG,CAAC,gBAAgB,CAAC,EAAC,cAAc,EAAC,CAAC,CAAC,CAAC;IAC5C,OAAO,GAAG,CAAC;AACb,CAAC;AAED,QAAQ,CAAC,iBAAiB,EAAE,GAAG,EAAE;IAC/B,IAAI,cAA8B,CAAC;IAEnC,UAAU,CAAC,GAAG,EAAE;QACd,cAAc,GAAG,kBAAkB,EAAE,CAAC;IACxC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,oCAAoC,EAAE,KAAK,IAAI,EAAE;QAClD,MAAM,GAAG,GAAG,aAAa,CAAC,cAAc,CAAC,CAAC;QAC1C,MAAM,GAAG,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACtD,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAC/B,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,8BAA8B,EAAE,KAAK,IAAI,EAAE;QAC5C,MAAM,GAAG,GAAG,aAAa,CAAC,cAAc,CAAC,CAAC;QAC1C,MAAM,GAAG,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,EAAC,OAAO,EAAE,OAAO,EAAC,CAAC,CAAC;QACtE,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAC/B,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,4CAA4C,EAAE,KAAK,IAAI,EAAE;QAC1D,MAAM,GAAG,GAAG,aAAa,CAAC,cAAc,CAAC,CAAC;QAC1C,MAAM,GAAG,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;aAC3B,IAAI,CAAC,OAAO,CAAC;aACb,IAAI,CAAC,EAAC,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,UAAU,EAAC,CAAC;aAC5C,GAAG,CAAC,QAAQ,EAAE,mBAAmB,CAAC,CAAC;QAEtC,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAC7B,MAAM,CAAC,GAAG,CAAC,OAAO,CAAC,cAAc,CAAC,CAAC,CAAC,SAAS,CAAC,mBAAmB,CAAC,CAAC;QACnE,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC;IACtC,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,8CAA8C,EAAE,KAAK,IAAI,EAAE;QAC5D,MAAM,GAAG,GAAG,aAAa,CAAC,cAAc,CAAC,CAAC;QAC1C,MAAM,OAAO,CAAC,GAAG,CAAC;aACf,IAAI,CAAC,OAAO,CAAC;aACb,IAAI,CAAC,EAAC,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,UAAU,EAAC,CAAC,CAAC;QAEhD,MAAM,CAAC,cAAc,CAAC,MAAM,CAAC,CAAC,oBAAoB,CAAC,UAAU,CAAC,CAAC;IACjE,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,sDAAsD,EAAE,KAAK,IAAI,EAAE;QACpE,MAAM,GAAG,GAAG,aAAa,CAAC,cAAc,CAAC,CAAC;QAE1C,yBAAyB;QACzB,MAAM,OAAO,CAAC,GAAG,CAAC;aACf,IAAI,CAAC,OAAO,CAAC;aACb,IAAI,CAAC,EAAC,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,UAAU,EAAC,CAAC,CAAC;QAEhD,WAAW;QACX,MAAM,OAAO,CAAC,GAAG,CAAC;aACf,IAAI,CAAC,OAAO,CAAC;aACb,IAAI,CAAC,EAAC,OAAO,EAAE,WAAW,EAAE,MAAM,EAAE,UAAU,EAAE,UAAU,EAAE,WAAW,EAAC,CAAC,CAAC;QAE7E,MAAM,CAAC,cAAc,CAAC,GAAG,CAAC,CAAC,oBAAoB,CAAC,WAAW,CAAC,CAAC;IAC/D,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,uCAAuC,EAAE,KAAK,IAAI,EAAE;QACrD,MAAM,GAAG,GAAG,aAAa,CAAC,cAAc,CAAC,CAAC;QAC1C,MAAM,GAAG,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;aAC3B,IAAI,CAAC,OAAO,CAAC;aACb,IAAI,CAAC,EAAC,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,UAAU,EAAC,CAAC,CAAC;QAEhD,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,SAAS,CAAC,eAAe,CAAC,CAAC;IAC9C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,wCAAwC,EAAE,KAAK,IAAI,EAAE;QAErD,cAAc,CAAC,MAAmC,CAAC,iBAAiB,CAAC,IAAI,KAAK,CAAC,cAAc,CAAC,CAAC,CAAC;QAEjG,MAAM,GAAG,GAAG,aAAa,CAAC,cAAc,CAAC,CAAC;QAC1C,MAAM,GAAG,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;aAC3B,IAAI,CAAC,OAAO,CAAC;aACb,IAAI,CAAC,EAAC,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,UAAU,EAAC,CAAC,CAAC;QAEhD,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,SAAS,CAAC,gBAAgB,CAAC,CAAC;IAC/C,CAAC,CAAC,CAAC;IAEH,EAAE,CAAC,2CAA2C,EAAE,KAAK,IAAI,EAAE;QACzD,MAAM,GAAG,GAAG,aAAa,CAAC,cAAc,CAAC,CAAC;QAC1C,MAAM,OAAO,CAAC,GAAG,CAAC;aACf,IAAI,CAAC,OAAO,CAAC;aACb,IAAI,CAAC,EAAC,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,SAAS,EAAE,SAAS,EAAC,CAAC,CAAC;QAEtE,sDAAsD;QACtD,MAAM,CAAC,cAAc,CAAC,MAAM,CAAC,CAAC,oBAAoB,CAAC,UAAU,CAAC,CAAC;IACjE,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}
|
|
@@ -0,0 +1,16 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* @license
|
|
3
|
+
* Copyright 2026 Amodal Labs, Inc.
|
|
4
|
+
* SPDX-License-Identifier: MIT
|
|
5
|
+
*/
|
|
6
|
+
import { Router } from 'express';
|
|
7
|
+
import type { SessionManager } from '../../session/session-manager.js';
|
|
8
|
+
import type { EvalStore } from '../eval-store.js';
|
|
9
|
+
export interface EvalRouterOptions {
|
|
10
|
+
sessionManager: SessionManager;
|
|
11
|
+
evalStore: EvalStore;
|
|
12
|
+
repoPath: string;
|
|
13
|
+
/** Port the server is listening on — used by eval query provider to call /chat */
|
|
14
|
+
getPort: () => number | null;
|
|
15
|
+
}
|
|
16
|
+
export declare function createEvalRouter(options: EvalRouterOptions): Router;
|
|
@@ -0,0 +1,383 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* @license
|
|
3
|
+
* Copyright 2026 Amodal Labs, Inc.
|
|
4
|
+
* SPDX-License-Identifier: MIT
|
|
5
|
+
*/
|
|
6
|
+
import { Router } from 'express';
|
|
7
|
+
import { buildEvalRun, judgeAllAssertions, computeEvalCost, aggregateRunCost } from '@amodalai/core';
|
|
8
|
+
async function streamQuery(baseUrl, message, evalRes, evalName, appId, sessionId) {
|
|
9
|
+
const chatRes = await fetch(`${baseUrl}/chat`, {
|
|
10
|
+
method: 'POST',
|
|
11
|
+
headers: { 'Content-Type': 'application/json' },
|
|
12
|
+
body: JSON.stringify({ message, app_id: appId ?? 'eval-runner', ...(sessionId ? { session_id: sessionId } : {}) }),
|
|
13
|
+
});
|
|
14
|
+
const text = await chatRes.text();
|
|
15
|
+
const lines = text.split('\n');
|
|
16
|
+
let fullResponse = '';
|
|
17
|
+
const toolCalls = [];
|
|
18
|
+
const toolResults = [];
|
|
19
|
+
let usage;
|
|
20
|
+
let queryError;
|
|
21
|
+
for (const line of lines) {
|
|
22
|
+
if (!line.startsWith('data: '))
|
|
23
|
+
continue;
|
|
24
|
+
try {
|
|
25
|
+
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion -- SSE parsing
|
|
26
|
+
const event = JSON.parse(line.substring(6));
|
|
27
|
+
const eventType = String(event['type'] ?? '');
|
|
28
|
+
if (eventType === 'text_delta') {
|
|
29
|
+
const content = String(event['content'] ?? '');
|
|
30
|
+
fullResponse += content;
|
|
31
|
+
writeSSE(evalRes, { type: 'agent_text', evalName, content });
|
|
32
|
+
}
|
|
33
|
+
else if (eventType === 'tool_call_start') {
|
|
34
|
+
const params = (event['parameters'] ?? {}); // eslint-disable-line @typescript-eslint/no-unsafe-type-assertion
|
|
35
|
+
toolCalls.push({ name: String(event['tool_name'] ?? ''), parameters: params });
|
|
36
|
+
writeSSE(evalRes, { type: 'agent_tool', evalName, toolName: event['tool_name'], parameters: params });
|
|
37
|
+
}
|
|
38
|
+
else if (eventType === 'tool_call_result') {
|
|
39
|
+
// Capture result data so the judge knows tool calls returned real data
|
|
40
|
+
const resultPreview = String(event['result'] ?? event['error'] ?? '');
|
|
41
|
+
toolResults.push(`${String(event['tool_name'] ?? 'request')}: ${resultPreview.slice(0, 500)}`);
|
|
42
|
+
writeSSE(evalRes, { type: 'agent_tool_result', evalName, toolName: event['tool_name'] ?? 'request', status: event['status'], durationMs: event['duration_ms'] });
|
|
43
|
+
}
|
|
44
|
+
else if (eventType === 'error') {
|
|
45
|
+
queryError = String(event['message'] ?? event['error'] ?? 'Unknown error');
|
|
46
|
+
writeSSE(evalRes, { type: 'agent_error', evalName, error: queryError });
|
|
47
|
+
}
|
|
48
|
+
else if (eventType === 'done' && event['usage']) {
|
|
49
|
+
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
|
|
50
|
+
const u = event['usage'];
|
|
51
|
+
// Accumulate tokens across multiple done events (multi-turn agent loops
|
|
52
|
+
// emit one done per turn in the refactored session runner)
|
|
53
|
+
if (u.input_tokens > 0 || u.output_tokens > 0 || (u.cached_tokens ?? 0) > 0) {
|
|
54
|
+
if (!usage) {
|
|
55
|
+
usage = { inputTokens: 0, outputTokens: 0 };
|
|
56
|
+
}
|
|
57
|
+
usage.inputTokens += u.input_tokens;
|
|
58
|
+
usage.outputTokens += u.output_tokens;
|
|
59
|
+
if (u.cached_tokens) {
|
|
60
|
+
usage.cacheReadInputTokens = (usage.cacheReadInputTokens ?? 0) + u.cached_tokens;
|
|
61
|
+
}
|
|
62
|
+
if (u.cache_creation_tokens) {
|
|
63
|
+
usage.cacheCreationInputTokens = (usage.cacheCreationInputTokens ?? 0) + u.cache_creation_tokens;
|
|
64
|
+
}
|
|
65
|
+
}
|
|
66
|
+
}
|
|
67
|
+
}
|
|
68
|
+
catch {
|
|
69
|
+
// skip
|
|
70
|
+
}
|
|
71
|
+
}
|
|
72
|
+
if (!usage) {
|
|
73
|
+
const outputChars = fullResponse.length;
|
|
74
|
+
const estimatedOutput = Math.ceil(outputChars / 4);
|
|
75
|
+
usage = { inputTokens: estimatedOutput * 3, outputTokens: estimatedOutput };
|
|
76
|
+
}
|
|
77
|
+
return { response: fullResponse, toolCalls, toolResults, usage, ...(queryError ? { error: queryError } : {}) };
|
|
78
|
+
}
|
|
79
|
+
/**
|
|
80
|
+
* Create a JudgeProvider that uses the local /chat endpoint and tracks token usage.
|
|
81
|
+
*/
|
|
82
|
+
function createLocalJudgeProvider(baseUrl) {
|
|
83
|
+
const tracked = {
|
|
84
|
+
totalInputTokens: 0,
|
|
85
|
+
totalOutputTokens: 0,
|
|
86
|
+
judge: async (prompt) => {
|
|
87
|
+
const response = await fetch(`${baseUrl}/chat`, {
|
|
88
|
+
method: 'POST',
|
|
89
|
+
headers: { 'Content-Type': 'application/json' },
|
|
90
|
+
body: JSON.stringify({ message: prompt, app_id: 'eval-judge', session_id: `judge-${Date.now()}` }),
|
|
91
|
+
});
|
|
92
|
+
const text = await response.text();
|
|
93
|
+
let result = '';
|
|
94
|
+
for (const line of text.split('\n')) {
|
|
95
|
+
if (!line.startsWith('data: '))
|
|
96
|
+
continue;
|
|
97
|
+
try {
|
|
98
|
+
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion -- SSE parsing
|
|
99
|
+
const event = JSON.parse(line.substring(6));
|
|
100
|
+
if (event['type'] === 'text_delta') {
|
|
101
|
+
result += String(event['content'] ?? '');
|
|
102
|
+
}
|
|
103
|
+
else if (event['type'] === 'done' && event['usage']) {
|
|
104
|
+
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion
|
|
105
|
+
const u = event['usage'];
|
|
106
|
+
tracked.totalInputTokens += u.input_tokens || 0;
|
|
107
|
+
tracked.totalOutputTokens += u.output_tokens || 0;
|
|
108
|
+
}
|
|
109
|
+
}
|
|
110
|
+
catch {
|
|
111
|
+
// skip
|
|
112
|
+
}
|
|
113
|
+
}
|
|
114
|
+
// Estimate if no usage reported
|
|
115
|
+
if (tracked.totalInputTokens === 0) {
|
|
116
|
+
tracked.totalInputTokens += Math.ceil(prompt.length / 4);
|
|
117
|
+
tracked.totalOutputTokens += Math.ceil(result.length / 4);
|
|
118
|
+
}
|
|
119
|
+
return result;
|
|
120
|
+
},
|
|
121
|
+
};
|
|
122
|
+
return tracked;
|
|
123
|
+
}
|
|
124
|
+
function writeSSE(res, data) {
|
|
125
|
+
res.write(`data: ${JSON.stringify(data)}\n\n`);
|
|
126
|
+
}
|
|
127
|
+
export function createEvalRouter(options) {
|
|
128
|
+
const router = Router();
|
|
129
|
+
/** List eval definitions from the repo */
|
|
130
|
+
router.get('/api/evals/suites', (_req, res) => {
|
|
131
|
+
const repo = options.sessionManager.getRepo();
|
|
132
|
+
const suites = repo.evals.map((e) => ({
|
|
133
|
+
name: e.name,
|
|
134
|
+
title: e.title,
|
|
135
|
+
description: e.description,
|
|
136
|
+
query: e.query,
|
|
137
|
+
assertionCount: e.assertions.length,
|
|
138
|
+
assertions: e.assertions.map((a) => ({ text: a.text, negated: a.negated })),
|
|
139
|
+
location: e.location,
|
|
140
|
+
}));
|
|
141
|
+
res.json({ suites });
|
|
142
|
+
});
|
|
143
|
+
/** List saved eval runs */
|
|
144
|
+
router.get('/api/evals/runs', (_req, res) => {
|
|
145
|
+
const runs = options.evalStore.list();
|
|
146
|
+
res.json({ runs });
|
|
147
|
+
});
|
|
148
|
+
/** Get a single eval run */
|
|
149
|
+
router.get('/api/evals/runs/:id', (req, res) => {
|
|
150
|
+
const run = options.evalStore.load(req.params['id'] ?? '');
|
|
151
|
+
if (!run) {
|
|
152
|
+
res.status(404).json({ error: 'Run not found' });
|
|
153
|
+
return;
|
|
154
|
+
}
|
|
155
|
+
res.json(run);
|
|
156
|
+
});
|
|
157
|
+
/** Delete an eval run */
|
|
158
|
+
router.delete('/api/evals/runs/:id', (req, res) => {
|
|
159
|
+
const deleted = options.evalStore.delete(req.params['id'] ?? '');
|
|
160
|
+
if (!deleted) {
|
|
161
|
+
res.status(404).json({ error: 'Run not found' });
|
|
162
|
+
return;
|
|
163
|
+
}
|
|
164
|
+
res.json({ ok: true });
|
|
165
|
+
});
|
|
166
|
+
/** Run eval suite — SSE stream with full per-eval results */
|
|
167
|
+
router.post('/api/evals/run', async (req, res) => {
|
|
168
|
+
const port = options.getPort();
|
|
169
|
+
if (!port) {
|
|
170
|
+
res.status(503).json({ error: 'Server not ready' });
|
|
171
|
+
return;
|
|
172
|
+
}
|
|
173
|
+
const baseUrl = `http://127.0.0.1:${port}`;
|
|
174
|
+
const repo = options.sessionManager.getRepo();
|
|
175
|
+
// Read optional eval names and model override from POST body
|
|
176
|
+
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion -- request body
|
|
177
|
+
const body = (req.body ?? {});
|
|
178
|
+
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion -- request body
|
|
179
|
+
const evalNames = body['evalNames'];
|
|
180
|
+
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion -- request body
|
|
181
|
+
const modelOverride = body['model'];
|
|
182
|
+
let evals = repo.evals;
|
|
183
|
+
if (evalNames && evalNames.length > 0) {
|
|
184
|
+
evals = evals.filter((e) => evalNames.includes(e.name));
|
|
185
|
+
}
|
|
186
|
+
if (evals.length === 0) {
|
|
187
|
+
res.status(400).json({ error: 'No evals defined' });
|
|
188
|
+
return;
|
|
189
|
+
}
|
|
190
|
+
res.writeHead(200, {
|
|
191
|
+
'Content-Type': 'text/event-stream',
|
|
192
|
+
'Cache-Control': 'no-cache',
|
|
193
|
+
'Connection': 'keep-alive',
|
|
194
|
+
});
|
|
195
|
+
// Save original model config for restoration after override
|
|
196
|
+
const originalModelConfig = repo.config?.models?.['main'];
|
|
197
|
+
// If model override provided, swap main model config
|
|
198
|
+
if (modelOverride && repo.config?.models) {
|
|
199
|
+
repo.config.models['main'] = {
|
|
200
|
+
provider: modelOverride.provider,
|
|
201
|
+
model: modelOverride.model,
|
|
202
|
+
};
|
|
203
|
+
}
|
|
204
|
+
const evalSessionId = `eval-${Date.now()}`;
|
|
205
|
+
const judgeSessionId = `judge-${Date.now()}`;
|
|
206
|
+
const judgeProvider = createLocalJudgeProvider(baseUrl);
|
|
207
|
+
const modelInfo = repo.config ? {
|
|
208
|
+
provider: repo.config.models?.['main']?.provider ?? 'unknown',
|
|
209
|
+
model: repo.config.models?.['main']?.model ?? 'unknown',
|
|
210
|
+
} : { provider: 'unknown', model: 'unknown' };
|
|
211
|
+
const results = [];
|
|
212
|
+
const perCaseCosts = [];
|
|
213
|
+
const startTime = Date.now();
|
|
214
|
+
// Restore original model before judging so judge uses the original model
|
|
215
|
+
const restoreModel = () => {
|
|
216
|
+
if (originalModelConfig && repo.config?.models) {
|
|
217
|
+
repo.config.models['main'] = originalModelConfig;
|
|
218
|
+
}
|
|
219
|
+
};
|
|
220
|
+
for (let i = 0; i < evals.length; i++) {
|
|
221
|
+
const ev = evals[i];
|
|
222
|
+
writeSSE(res, { type: 'eval_start', evalName: ev.name, current: i + 1, total: evals.length });
|
|
223
|
+
const evalStart = Date.now();
|
|
224
|
+
try {
|
|
225
|
+
// Run the query — streams agent events to client
|
|
226
|
+
const { response, toolCalls, toolResults, usage, error: queryError } = await streamQuery(baseUrl, ev.query, res, ev.name, ev.setup.app, evalSessionId);
|
|
227
|
+
// Restore original model for judging
|
|
228
|
+
restoreModel();
|
|
229
|
+
let assertions = [];
|
|
230
|
+
let passed = false;
|
|
231
|
+
let judgeCost;
|
|
232
|
+
// Skip judging if query had an error
|
|
233
|
+
if (!queryError) {
|
|
234
|
+
// Build enriched response for the judge — include tool results so it knows data was fetched
|
|
235
|
+
let enriched = response;
|
|
236
|
+
if (toolCalls.length > 0) {
|
|
237
|
+
enriched += '\n\n[Tool calls made: ' + toolCalls.map((tc) => `${tc.name}(${JSON.stringify(tc.parameters)})`).join(', ') + ']';
|
|
238
|
+
}
|
|
239
|
+
if (toolResults.length > 0) {
|
|
240
|
+
enriched += '\n\n[Tool results received:\n' + toolResults.join('\n') + ']';
|
|
241
|
+
}
|
|
242
|
+
// Judge assertions — track judge tokens separately
|
|
243
|
+
const judgeInputBefore = judgeProvider.totalInputTokens;
|
|
244
|
+
const judgeOutputBefore = judgeProvider.totalOutputTokens;
|
|
245
|
+
assertions = await judgeAllAssertions(enriched, ev.assertions, judgeProvider);
|
|
246
|
+
passed = assertions.every((a) => a.passed);
|
|
247
|
+
const judgeInputUsed = judgeProvider.totalInputTokens - judgeInputBefore;
|
|
248
|
+
const judgeOutputUsed = judgeProvider.totalOutputTokens - judgeOutputBefore;
|
|
249
|
+
judgeCost = judgeInputUsed > 0 ? computeEvalCost(judgeInputUsed, judgeOutputUsed, originalModelConfig?.model ?? modelInfo.model) : undefined;
|
|
250
|
+
}
|
|
251
|
+
// Re-apply model override for next eval query
|
|
252
|
+
if (modelOverride && repo.config?.models) {
|
|
253
|
+
repo.config.models['main'] = {
|
|
254
|
+
provider: modelOverride.provider,
|
|
255
|
+
model: modelOverride.model,
|
|
256
|
+
};
|
|
257
|
+
}
|
|
258
|
+
const queryCost = usage ? computeEvalCost(usage.inputTokens, usage.outputTokens, modelInfo.model, usage.cacheReadInputTokens, usage.cacheCreationInputTokens) : undefined;
|
|
259
|
+
if (queryCost)
|
|
260
|
+
perCaseCosts.push(queryCost);
|
|
261
|
+
const result = {
|
|
262
|
+
eval: ev,
|
|
263
|
+
response,
|
|
264
|
+
toolCalls,
|
|
265
|
+
assertions,
|
|
266
|
+
passed,
|
|
267
|
+
durationMs: Date.now() - evalStart,
|
|
268
|
+
cost: queryCost,
|
|
269
|
+
...(queryError ? { error: queryError } : {}),
|
|
270
|
+
};
|
|
271
|
+
results.push(result);
|
|
272
|
+
// Send full result with eval_complete — separate query and judge costs
|
|
273
|
+
writeSSE(res, {
|
|
274
|
+
type: 'eval_complete',
|
|
275
|
+
evalName: ev.name,
|
|
276
|
+
passed,
|
|
277
|
+
current: i + 1,
|
|
278
|
+
total: evals.length,
|
|
279
|
+
result: {
|
|
280
|
+
response: response.length > 4000 ? response.slice(0, 4000) + '...' : response,
|
|
281
|
+
toolCalls,
|
|
282
|
+
toolResults,
|
|
283
|
+
assertions,
|
|
284
|
+
durationMs: result.durationMs,
|
|
285
|
+
queryCost,
|
|
286
|
+
judgeCost,
|
|
287
|
+
...(queryError ? { error: queryError } : {}),
|
|
288
|
+
},
|
|
289
|
+
});
|
|
290
|
+
}
|
|
291
|
+
catch (err) {
|
|
292
|
+
const msg = err instanceof Error ? err.message : String(err);
|
|
293
|
+
// Restore original model on error too
|
|
294
|
+
restoreModel();
|
|
295
|
+
if (modelOverride && repo.config?.models) {
|
|
296
|
+
repo.config.models['main'] = {
|
|
297
|
+
provider: modelOverride.provider,
|
|
298
|
+
model: modelOverride.model,
|
|
299
|
+
};
|
|
300
|
+
}
|
|
301
|
+
const result = {
|
|
302
|
+
eval: ev,
|
|
303
|
+
response: '',
|
|
304
|
+
toolCalls: [],
|
|
305
|
+
assertions: [],
|
|
306
|
+
passed: false,
|
|
307
|
+
durationMs: Date.now() - evalStart,
|
|
308
|
+
error: msg,
|
|
309
|
+
};
|
|
310
|
+
results.push(result);
|
|
311
|
+
writeSSE(res, {
|
|
312
|
+
type: 'eval_complete',
|
|
313
|
+
evalName: ev.name,
|
|
314
|
+
passed: false,
|
|
315
|
+
current: i + 1,
|
|
316
|
+
total: evals.length,
|
|
317
|
+
result: { response: '', toolCalls: [], assertions: [], durationMs: result.durationMs, error: msg },
|
|
318
|
+
});
|
|
319
|
+
}
|
|
320
|
+
}
|
|
321
|
+
// Restore original model config
|
|
322
|
+
restoreModel();
|
|
323
|
+
// Build suite result
|
|
324
|
+
const totalCost = perCaseCosts.length > 0 ? aggregateRunCost(perCaseCosts) : undefined;
|
|
325
|
+
const suiteResult = {
|
|
326
|
+
results,
|
|
327
|
+
totalPassed: results.filter((r) => r.passed).length,
|
|
328
|
+
totalFailed: results.filter((r) => !r.passed).length,
|
|
329
|
+
totalSkipped: 0,
|
|
330
|
+
totalDurationMs: Date.now() - startTime,
|
|
331
|
+
totalCost,
|
|
332
|
+
model: modelInfo,
|
|
333
|
+
timestamp: new Date().toISOString(),
|
|
334
|
+
};
|
|
335
|
+
const run = buildEvalRun(suiteResult, modelInfo, { orgId: 'local', triggeredBy: 'manual' });
|
|
336
|
+
options.evalStore.save(run); // eslint-disable-line @typescript-eslint/no-unsafe-type-assertion
|
|
337
|
+
// Suppress unused variable warnings
|
|
338
|
+
void evalSessionId;
|
|
339
|
+
void judgeSessionId;
|
|
340
|
+
writeSSE(res, { type: 'run_complete', run });
|
|
341
|
+
writeSSE(res, { type: 'done' });
|
|
342
|
+
res.end();
|
|
343
|
+
});
|
|
344
|
+
/** Get eval history for a specific eval */
|
|
345
|
+
router.get('/api/evals/runs/by-eval/:evalName', (req, res) => {
|
|
346
|
+
const evalName = req.params['evalName'] ?? '';
|
|
347
|
+
const entries = options.evalStore.listByEval(evalName);
|
|
348
|
+
res.json({ entries });
|
|
349
|
+
});
|
|
350
|
+
/** Get arena model config */
|
|
351
|
+
router.get('/api/evals/arena/models', (_req, res) => {
|
|
352
|
+
const repo = options.sessionManager.getRepo();
|
|
353
|
+
const config = repo.config;
|
|
354
|
+
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion -- config shape
|
|
355
|
+
const rawConfig = config;
|
|
356
|
+
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion -- config shape
|
|
357
|
+
const arena = rawConfig['arena'];
|
|
358
|
+
// eslint-disable-next-line @typescript-eslint/no-unsafe-type-assertion -- config shape
|
|
359
|
+
const configModels = arena?.['models'];
|
|
360
|
+
const models = configModels ?? [
|
|
361
|
+
{ provider: 'anthropic', model: 'claude-opus-4-6', label: 'Claude Opus 4.6' },
|
|
362
|
+
{ provider: 'anthropic', model: 'claude-sonnet-4-6', label: 'Claude Sonnet 4.6' },
|
|
363
|
+
{ provider: 'anthropic', model: 'claude-haiku-4-5-20251001', label: 'Claude Haiku 4.5' },
|
|
364
|
+
{ provider: 'openai', model: 'gpt-4o', label: 'GPT-4o' },
|
|
365
|
+
{ provider: 'openai', model: 'gpt-4o-mini', label: 'GPT-4o Mini' },
|
|
366
|
+
{ provider: 'openai', model: 'gpt-4.1', label: 'GPT-4.1' },
|
|
367
|
+
{ provider: 'openai', model: 'gpt-4.1-mini', label: 'GPT-4.1 Mini' },
|
|
368
|
+
{ provider: 'google', model: 'gemini-2.5-pro', label: 'Gemini 2.5 Pro' },
|
|
369
|
+
{ provider: 'google', model: 'gemini-2.5-flash', label: 'Gemini 2.5 Flash' },
|
|
370
|
+
{ provider: 'deepseek', model: 'deepseek-chat', label: 'DeepSeek Chat' },
|
|
371
|
+
{ provider: 'deepseek', model: 'deepseek-reasoner', label: 'DeepSeek Reasoner' },
|
|
372
|
+
{ provider: 'groq', model: 'llama-3.3-70b-versatile', label: 'Llama 3.3 70B (Groq)' },
|
|
373
|
+
{ provider: 'groq', model: 'llama-3.1-8b-instant', label: 'Llama 3.1 8B (Groq)' },
|
|
374
|
+
{ provider: 'groq', model: 'meta-llama/llama-4-scout-17b-16e-instruct', label: 'Llama 4 Scout (Groq)' },
|
|
375
|
+
{ provider: 'groq', model: 'qwen/qwen3-32b', label: 'Qwen 3 32B (Groq)' },
|
|
376
|
+
{ provider: 'groq', model: 'moonshotai/kimi-k2-instruct', label: 'Kimi K2 (Groq)' },
|
|
377
|
+
{ provider: 'groq', model: 'openai/gpt-oss-120b', label: 'GPT-OSS 120B (Groq)' },
|
|
378
|
+
];
|
|
379
|
+
res.json({ models });
|
|
380
|
+
});
|
|
381
|
+
return router;
|
|
382
|
+
}
|
|
383
|
+
//# sourceMappingURL=evals.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evals.js","sourceRoot":"","sources":["../../../../src/agent/routes/evals.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,EAAC,MAAM,EAAC,MAAM,SAAS,CAAC;AAI/B,OAAO,EAAC,YAAY,EAAE,kBAAkB,EAAE,eAAe,EAAE,gBAAgB,EAAC,MAAM,gBAAgB,CAAC;AAsBnG,KAAK,UAAU,WAAW,CACxB,OAAe,EACf,OAAe,EACf,OAAiB,EACjB,QAAgB,EAChB,KAAc,EACd,SAAkB;IAElB,MAAM,OAAO,GAAG,MAAM,KAAK,CAAC,GAAG,OAAO,OAAO,EAAE;QAC7C,MAAM,EAAE,MAAM;QACd,OAAO,EAAE,EAAC,cAAc,EAAE,kBAAkB,EAAC;QAC7C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAC,OAAO,EAAE,MAAM,EAAE,KAAK,IAAI,aAAa,EAAE,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,EAAC,UAAU,EAAE,SAAS,EAAC,CAAC,CAAC,CAAC,EAAE,CAAC,EAAC,CAAC;KAC/G,CAAC,CAAC;IAEH,MAAM,IAAI,GAAG,MAAM,OAAO,CAAC,IAAI,EAAE,CAAC;IAClC,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAC/B,IAAI,YAAY,GAAG,EAAE,CAAC;IACtB,MAAM,SAAS,GAA+D,EAAE,CAAC;IACjF,MAAM,WAAW,GAAa,EAAE,CAAC;IACjC,IAAI,KAA6B,CAAC;IAClC,IAAI,UAA8B,CAAC;IAEnC,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,QAAQ,CAAC;YAAE,SAAS;QACzC,IAAI,CAAC;YACH,sFAAsF;YACtF,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,CAA4B,CAAC;YACvE,MAAM,SAAS,GAAG,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC,CAAC;YAE9C,IAAI,SAAS,KAAK,YAAY,EAAE,CAAC;gBAC/B,MAAM,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC,SAAS,CAAC,IAAI,EAAE,CAAC,CAAC;gBAC/C,YAAY,IAAI,OAAO,CAAC;gBACxB,QAAQ,CAAC,OAAO,EAAE,EAAC,IAAI,EAAE,YAAY,EAAE,QAAQ,EAAE,OAAO,EAAC,CAAC,CAAC;YAC7D,CAAC;iBAAM,IAAI,SAAS,KAAK,iBAAiB,EAAE,CAAC;gBAC3C,MAAM,MAAM,GAAG,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,EAAE,CAA4B,CAAC,CAAC,kEAAkE;gBACzI,SAAS,CAAC,IAAI,CAAC,EAAC,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,WAAW,CAAC,IAAI,EAAE,CAAC,EAAE,UAAU,EAAE,MAAM,EAAC,CAAC,CAAC;gBAC7E,QAAQ,CAAC,OAAO,EAAE,EAAC,IAAI,EAAE,YAAY,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,CAAC,WAAW,CAAC,EAAE,UAAU,EAAE,MAAM,EAAC,CAAC,CAAC;YACtG,CAAC;iBAAM,IAAI,SAAS,KAAK,kBAAkB,EAAE,CAAC;gBAC5C,uEAAuE;gBACvE,MAAM,aAAa,GAAG,MAAM,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,KAAK,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;gBACtE,WAAW,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,WAAW,CAAC,IAAI,SAAS,CAAC,KAAK,aAAa,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;gBAC/F,QAAQ,CAAC,OAAO,EAAE,EAAC,IAAI,EAAE,mBAAmB,EAAE,QAAQ,EAAE,QAAQ,EAAE,KAAK,CAAC,WAAW,CAAC,IAAI,SAAS,EAAE,MAAM,EAAE,KAAK,CAAC,QAAQ,CAAC,EAAE,UAAU,EAAE,KAAK,CAAC,aAAa,CAAC,EAAC,CAAC,CAAC;YACjK,CAAC;iBAAM,IAAI,SAAS,KAAK,OAAO,EAAE,CAAC;gBACjC,UAAU,GAAG,MAAM,CAAC,KAAK,CAAC,SAAS,CAAC,IAAI,KAAK,CAAC,OAAO,CAAC,IAAI,eAAe,CAAC,CAAC;gBAC3E,QAAQ,CAAC,OAAO,EAAE,EAAC,IAAI,EAAE,aAAa,EAAE,QAAQ,EAAE,KAAK,EAAE,UAAU,EAAC,CAAC,CAAC;YACxE,CAAC;iBAAM,IAAI,SAAS,KAAK,MAAM,IAAI,KAAK,CAAC,OAAO,CAAC,EAAE,CAAC;gBAClD,uEAAuE;gBACvE,MAAM,CAAC,GAAG,KAAK,CAAC,OAAO,CAA0G,CAAC;gBAClI,wEAAwE;gBACxE,2DAA2D;gBAC3D,IAAI,CAAC,CAAC,YAAY,GAAG,CAAC,IAAI,CAAC,CAAC,aAAa,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,aAAa,IAAI,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC;oBAC5E,IAAI,CAAC,KAAK,EAAE,CAAC;wBACX,KAAK,GAAG,EAAC,WAAW,EAAE,CAAC,EAAE,YAAY,EAAE,CAAC,EAAC,CAAC;oBAC5C,CAAC;oBACD,KAAK,CAAC,WAAW,IAAI,CAAC,CAAC,YAAY,CAAC;oBACpC,KAAK,CAAC,YAAY,IAAI,CAAC,CAAC,aAAa,CAAC;oBACtC,IAAI,CAAC,CAAC,aAAa,EAAE,CAAC;wBACpB,KAAK,CAAC,oBAAoB,GAAG,CAAC,KAAK,CAAC,oBAAoB,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,aAAa,CAAC;oBACnF,CAAC;oBACD,IAAI,CAAC,CAAC,qBAAqB,EAAE,CAAC;wBAC5B,KAAK,CAAC,wBAAwB,GAAG,CAAC,KAAK,CAAC,wBAAwB,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,qBAAqB,CAAC;oBACnG,CAAC;gBACH,CAAC;YACH,CAAC;QACH,CAAC;QAAC,MAAM,CAAC;YACP,OAAO;QACT,CAAC;IACH,CAAC;IAED,IAAI,CAAC,KAAK,EAAE,CAAC;QACX,MAAM,WAAW,GAAG,YAAY,CAAC,MAAM,CAAC;QACxC,MAAM,eAAe,GAAG,IAAI,CAAC,IAAI,CAAC,WAAW,GAAG,CAAC,CAAC,CAAC;QACnD,KAAK,GAAG,EAAC,WAAW,EAAE,eAAe,GAAG,CAAC,EAAE,YAAY,EAAE,eAAe,EAAC,CAAC;IAC5E,CAAC;IAED,OAAO,EAAC,QAAQ,EAAE,YAAY,EAAE,SAAS,EAAE,WAAW,EAAE,KAAK,EAAE,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC,EAAC,KAAK,EAAE,UAAU,EAAC,CAAC,CAAC,CAAC,EAAE,CAAC,EAAC,CAAC;AAC7G,CAAC;AAOD;;GAEG;AACH,SAAS,wBAAwB,CAAC,OAAe;IAC/C,MAAM,OAAO,GAAyB;QACpC,gBAAgB,EAAE,CAAC;QACnB,iBAAiB,EAAE,CAAC;QACpB,KAAK,EAAE,KAAK,EAAE,MAAc,EAAE,EAAE;YAC9B,MAAM,QAAQ,GAAG,MAAM,KAAK,CAAC,GAAG,OAAO,OAAO,EAAE;gBAC9C,MAAM,EAAE,MAAM;gBACd,OAAO,EAAE,EAAC,cAAc,EAAE,kBAAkB,EAAC;gBAC7C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,EAAC,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,YAAY,EAAE,UAAU,EAAE,SAAS,IAAI,CAAC,GAAG,EAAE,EAAE,EAAC,CAAC;aACjG,CAAC,CAAC;YACH,MAAM,IAAI,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,CAAC;YACnC,IAAI,MAAM,GAAG,EAAE,CAAC;YAChB,KAAK,MAAM,IAAI,IAAI,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC;gBACpC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,QAAQ,CAAC;oBAAE,SAAS;gBACzC,IAAI,CAAC;oBACH,sFAAsF;oBACtF,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,CAA4B,CAAC;oBACvE,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,YAAY,EAAE,CAAC;wBACnC,MAAM,IAAI,MAAM,CAAC,KAAK,CAAC,SAAS,CAAC,IAAI,EAAE,CAAC,CAAC;oBAC3C,CAAC;yBAAM,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,MAAM,IAAI,KAAK,CAAC,OAAO,CAAC,EAAE,CAAC;wBACtD,uEAAuE;wBACvE,MAAM,CAAC,GAAG,KAAK,CAAC,OAAO,CAAkD,CAAC;wBAC1E,OAAO,CAAC,gBAAgB,IAAI,CAAC,CAAC,YAAY,IAAI,CAAC,CAAC;wBAChD,OAAO,CAAC,iBAAiB,IAAI,CAAC,CAAC,aAAa,IAAI,CAAC,CAAC;oBACpD,CAAC;gBACH,CAAC;gBAAC,MAAM,CAAC;oBACP,OAAO;gBACT,CAAC;YACH,CAAC;YACD,gCAAgC;YAChC,IAAI,OAAO,CAAC,gBAAgB,KAAK,CAAC,EAAE,CAAC;gBACnC,OAAO,CAAC,gBAAgB,IAAI,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;gBACzD,OAAO,CAAC,iBAAiB,IAAI,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;YAC5D,CAAC;YACD,OAAO,MAAM,CAAC;QAChB,CAAC;KACF,CAAC;IACF,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,SAAS,QAAQ,CAAC,GAAa,EAAE,IAAa;IAC5C,GAAG,CAAC,KAAK,CAAC,SAAS,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;AACjD,CAAC;AAED,MAAM,UAAU,gBAAgB,CAAC,OAA0B;IACzD,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC;IAExB,0CAA0C;IAC1C,MAAM,CAAC,GAAG,CAAC,mBAAmB,EAAE,CAAC,IAAa,EAAE,GAAa,EAAE,EAAE;QAC/D,MAAM,IAAI,GAAG,OAAO,CAAC,cAAc,CAAC,OAAO,EAAG,CAAC;QAC/C,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YACpC,IAAI,EAAE,CAAC,CAAC,IAAI;YACZ,KAAK,EAAE,CAAC,CAAC,KAAK;YACd,WAAW,EAAE,CAAC,CAAC,WAAW;YAC1B,KAAK,EAAE,CAAC,CAAC,KAAK;YACd,cAAc,EAAE,CAAC,CAAC,UAAU,CAAC,MAAM;YACnC,UAAU,EAAE,CAAC,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAC,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAC,CAAC,CAAC;YACzE,QAAQ,EAAE,CAAC,CAAC,QAAQ;SACrB,CAAC,CAAC,CAAC;QACJ,GAAG,CAAC,IAAI,CAAC,EAAC,MAAM,EAAC,CAAC,CAAC;IACrB,CAAC,CAAC,CAAC;IAEH,2BAA2B;IAC3B,MAAM,CAAC,GAAG,CAAC,iBAAiB,EAAE,CAAC,IAAa,EAAE,GAAa,EAAE,EAAE;QAC7D,MAAM,IAAI,GAAG,OAAO,CAAC,SAAS,CAAC,IAAI,EAAE,CAAC;QACtC,GAAG,CAAC,IAAI,CAAC,EAAC,IAAI,EAAC,CAAC,CAAC;IACnB,CAAC,CAAC,CAAC;IAEH,4BAA4B;IAC5B,MAAM,CAAC,GAAG,CAAC,qBAAqB,EAAE,CAAC,GAAY,EAAE,GAAa,EAAE,EAAE;QAChE,MAAM,GAAG,GAAG,OAAO,CAAC,SAAS,CAAC,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC;QAC3D,IAAI,CAAC,GAAG,EAAE,CAAC;YACT,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,EAAC,KAAK,EAAE,eAAe,EAAC,CAAC,CAAC;YAC/C,OAAO;QACT,CAAC;QACD,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAChB,CAAC,CAAC,CAAC;IAEH,yBAAyB;IACzB,MAAM,CAAC,MAAM,CAAC,qBAAqB,EAAE,CAAC,GAAY,EAAE,GAAa,EAAE,EAAE;QACnE,MAAM,OAAO,GAAG,OAAO,CAAC,SAAS,CAAC,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC;QACjE,IAAI,CAAC,OAAO,EAAE,CAAC;YACb,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,EAAC,KAAK,EAAE,eAAe,EAAC,CAAC,CAAC;YAC/C,OAAO;QACT,CAAC;QACD,GAAG,CAAC,IAAI,CAAC,EAAC,EAAE,EAAE,IAAI,EAAC,CAAC,CAAC;IACvB,CAAC,CAAC,CAAC;IAEH,6DAA6D;IAC7D,MAAM,CAAC,IAAI,CAAC,gBAAgB,EAAE,KAAK,EAAE,GAAY,EAAE,GAAa,EAAE,EAAE;QAClE,MAAM,IAAI,GAAG,OAAO,CAAC,OAAO,EAAE,CAAC;QAC/B,IAAI,CAAC,IAAI,EAAE,CAAC;YACV,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,EAAC,KAAK,EAAE,kBAAkB,EAAC,CAAC,CAAC;YAClD,OAAO;QACT,CAAC;QAED,MAAM,OAAO,GAAG,oBAAoB,IAAI,EAAE,CAAC;QAC3C,MAAM,IAAI,GAAG,OAAO,CAAC,cAAc,CAAC,OAAO,EAAG,CAAC;QAE/C,6DAA6D;QAC7D,uFAAuF;QACvF,MAAM,IAAI,GAAG,CAAC,GAAG,CAAC,IAAI,IAAI,EAAE,CAA4B,CAAC;QACzD,uFAAuF;QACvF,MAAM,SAAS,GAAG,IAAI,CAAC,WAAW,CAAyB,CAAC;QAC5D,uFAAuF;QACvF,MAAM,aAAa,GAAG,IAAI,CAAC,OAAO,CAAkD,CAAC;QAErF,IAAI,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC;QACvB,IAAI,SAAS,IAAI,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACtC,KAAK,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC;QAC1D,CAAC;QAED,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YACvB,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,EAAC,KAAK,EAAE,kBAAkB,EAAC,CAAC,CAAC;YAClD,OAAO;QACT,CAAC;QAED,GAAG,CAAC,SAAS,CAAC,GAAG,EAAE;YACjB,cAAc,EAAE,mBAAmB;YACnC,eAAe,EAAE,UAAU;YAC3B,YAAY,EAAE,YAAY;SAC3B,CAAC,CAAC;QAEH,4DAA4D;QAC5D,MAAM,mBAAmB,GAAG,IAAI,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,MAAM,CAAC,CAAC;QAE1D,qDAAqD;QACrD,IAAI,aAAa,IAAI,IAAI,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC;YACzC,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,GAAG;gBAC3B,QAAQ,EAAE,aAAa,CAAC,QAAQ;gBAChC,KAAK,EAAE,aAAa,CAAC,KAAK;aAC3B,CAAC;QACJ,CAAC;QAED,MAAM,aAAa,GAAG,QAAQ,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC;QAC3C,MAAM,cAAc,GAAG,SAAS,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC;QAE7C,MAAM,aAAa,GAAG,wBAAwB,CAAC,OAAO,CAAC,CAAC;QACxD,MAAM,SAAS,GAAG,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC;YAC9B,QAAQ,EAAE,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC,MAAM,CAAC,EAAE,QAAQ,IAAI,SAAS;YAC7D,KAAK,EAAE,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC,MAAM,CAAC,EAAE,KAAK,IAAI,SAAS;SACxD,CAAC,CAAC,CAAC,EAAC,QAAQ,EAAE,SAAS,EAAE,KAAK,EAAE,SAAS,EAAC,CAAC;QAE5C,MAAM,OAAO,GAAiB,EAAE,CAAC;QACjC,MAAM,YAAY,GAAmB,EAAE,CAAC;QACxC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAE7B,yEAAyE;QACzE,MAAM,YAAY,GAAG,GAAG,EAAE;YACxB,IAAI,mBAAmB,IAAI,IAAI,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC;gBAC/C,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,GAAG,mBAAmB,CAAC;YACnD,CAAC;QACH,CAAC,CAAC;QAEF,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACtC,MAAM,EAAE,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;YACpB,QAAQ,CAAC,GAAG,EAAE,EAAC,IAAI,EAAE,YAAY,EAAE,QAAQ,EAAE,EAAE,CAAC,IAAI,EAAE,OAAO,EAAE,CAAC,GAAG,CAAC,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,EAAC,CAAC,CAAC;YAE5F,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;YAC7B,IAAI,CAAC;gBACH,iDAAiD;gBACjD,MAAM,EAAC,QAAQ,EAAE,SAAS,EAAE,WAAW,EAAE,KAAK,EAAE,KAAK,EAAE,UAAU,EAAC,GAAG,MAAM,WAAW,CAAC,OAAO,EAAE,EAAE,CAAC,KAAK,EAAE,GAAG,EAAE,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,aAAa,CAAC,CAAC;gBAErJ,qCAAqC;gBACrC,YAAY,EAAE,CAAC;gBAEf,IAAI,UAAU,GAAoD,EAAE,CAAC;gBACrE,IAAI,MAAM,GAAG,KAAK,CAAC;gBACnB,IAAI,SAAmC,CAAC;gBAExC,qCAAqC;gBACrC,IAAI,CAAC,UAAU,EAAE,CAAC;oBAChB,4FAA4F;oBAC5F,IAAI,QAAQ,GAAG,QAAQ,CAAC;oBACxB,IAAI,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;wBACzB,QAAQ,IAAI,wBAAwB,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,GAAG,EAAE,CAAC,IAAI,IAAI,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,GAAG,CAAC;oBAChI,CAAC;oBACD,IAAI,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;wBAC3B,QAAQ,IAAI,+BAA+B,GAAG,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,GAAG,CAAC;oBAC7E,CAAC;oBAED,mDAAmD;oBACnD,MAAM,gBAAgB,GAAG,aAAa,CAAC,gBAAgB,CAAC;oBACxD,MAAM,iBAAiB,GAAG,aAAa,CAAC,iBAAiB,CAAC;oBAC1D,UAAU,GAAG,MAAM,kBAAkB,CAAC,QAAQ,EAAE,EAAE,CAAC,UAAU,EAAE,aAAa,CAAC,CAAC;oBAC9E,MAAM,GAAG,UAAU,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC;oBAE3C,MAAM,cAAc,GAAG,aAAa,CAAC,gBAAgB,GAAG,gBAAgB,CAAC;oBACzE,MAAM,eAAe,GAAG,aAAa,CAAC,iBAAiB,GAAG,iBAAiB,CAAC;oBAC5E,SAAS,GAAG,cAAc,GAAG,CAAC,CAAC,CAAC,CAAC,eAAe,CAAC,cAAc,EAAE,eAAe,EAAE,mBAAmB,EAAE,KAAK,IAAI,SAAS,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;gBAC/I,CAAC;gBAED,8CAA8C;gBAC9C,IAAI,aAAa,IAAI,IAAI,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC;oBACzC,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,GAAG;wBAC3B,QAAQ,EAAE,aAAa,CAAC,QAAQ;wBAChC,KAAK,EAAE,aAAa,CAAC,KAAK;qBAC3B,CAAC;gBACJ,CAAC;gBAED,MAAM,SAAS,GAAG,KAAK,CAAC,CAAC,CAAC,eAAe,CACvC,KAAK,CAAC,WAAW,EAAE,KAAK,CAAC,YAAY,EAAE,SAAS,CAAC,KAAK,EACtD,KAAK,CAAC,oBAAoB,EAAE,KAAK,CAAC,wBAAwB,CAC3D,CAAC,CAAC,CAAC,SAAS,CAAC;gBAEd,IAAI,SAAS;oBAAE,YAAY,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;gBAE5C,MAAM,MAAM,GAAe;oBACzB,IAAI,EAAE,EAAE;oBACR,QAAQ;oBACR,SAAS;oBACT,UAAU;oBACV,MAAM;oBACN,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;oBAClC,IAAI,EAAE,SAAS;oBACf,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC,EAAC,KAAK,EAAE,UAAU,EAAC,CAAC,CAAC,CAAC,EAAE,CAAC;iBAC3C,CAAC;gBACF,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBAErB,uEAAuE;gBACvE,QAAQ,CAAC,GAAG,EAAE;oBACZ,IAAI,EAAE,eAAe;oBACrB,QAAQ,EAAE,EAAE,CAAC,IAAI;oBACjB,MAAM;oBACN,OAAO,EAAE,CAAC,GAAG,CAAC;oBACd,KAAK,EAAE,KAAK,CAAC,MAAM;oBACnB,MAAM,EAAE;wBACN,QAAQ,EAAE,QAAQ,CAAC,MAAM,GAAG,IAAI,CAAC,CAAC,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,KAAK,CAAC,CAAC,CAAC,QAAQ;wBAC7E,SAAS;wBACT,WAAW;wBACX,UAAU;wBACV,UAAU,EAAE,MAAM,CAAC,UAAU;wBAC7B,SAAS;wBACT,SAAS;wBACT,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC,EAAC,KAAK,EAAE,UAAU,EAAC,CAAC,CAAC,CAAC,EAAE,CAAC;qBAC3C;iBACF,CAAC,CAAC;YACL,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;gBAC7D,sCAAsC;gBACtC,YAAY,EAAE,CAAC;gBACf,IAAI,aAAa,IAAI,IAAI,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC;oBACzC,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,GAAG;wBAC3B,QAAQ,EAAE,aAAa,CAAC,QAAQ;wBAChC,KAAK,EAAE,aAAa,CAAC,KAAK;qBAC3B,CAAC;gBACJ,CAAC;gBACD,MAAM,MAAM,GAAe;oBACzB,IAAI,EAAE,EAAE;oBACR,QAAQ,EAAE,EAAE;oBACZ,SAAS,EAAE,EAAE;oBACb,UAAU,EAAE,EAAE;oBACd,MAAM,EAAE,KAAK;oBACb,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;oBAClC,KAAK,EAAE,GAAG;iBACX,CAAC;gBACF,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBACrB,QAAQ,CAAC,GAAG,EAAE;oBACZ,IAAI,EAAE,eAAe;oBACrB,QAAQ,EAAE,EAAE,CAAC,IAAI;oBACjB,MAAM,EAAE,KAAK;oBACb,OAAO,EAAE,CAAC,GAAG,CAAC;oBACd,KAAK,EAAE,KAAK,CAAC,MAAM;oBACnB,MAAM,EAAE,EAAC,QAAQ,EAAE,EAAE,EAAE,SAAS,EAAE,EAAE,EAAE,UAAU,EAAE,EAAE,EAAE,UAAU,EAAE,MAAM,CAAC,UAAU,EAAE,KAAK,EAAE,GAAG,EAAC;iBACjG,CAAC,CAAC;YACL,CAAC;QACH,CAAC;QAED,gCAAgC;QAChC,YAAY,EAAE,CAAC;QAEf,qBAAqB;QACrB,MAAM,SAAS,GAAG,YAAY,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,gBAAgB,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;QACvF,MAAM,WAAW,GAAoB;YACnC,OAAO;YACP,WAAW,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,MAAM;YACnD,WAAW,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,MAAM;YACpD,YAAY,EAAE,CAAC;YACf,eAAe,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;YACvC,SAAS;YACT,KAAK,EAAE,SAAS;YAChB,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;SACpC,CAAC;QAEF,MAAM,GAAG,GAAG,YAAY,CAAC,WAAW,EAAE,SAAS,EAAE,EAAC,KAAK,EAAE,OAAO,EAAE,WAAW,EAAE,QAAQ,EAAC,CAAC,CAAC;QAC1F,OAAO,CAAC,SAAS,CAAC,IAAI,CAAC,GAAyC,CAAC,CAAC,CAAC,kEAAkE;QAErI,oCAAoC;QACpC,KAAK,aAAa,CAAC;QACnB,KAAK,cAAc,CAAC;QAEpB,QAAQ,CAAC,GAAG,EAAE,EAAC,IAAI,EAAE,cAAc,EAAE,GAAG,EAAC,CAAC,CAAC;QAC3C,QAAQ,CAAC,GAAG,EAAE,EAAC,IAAI,EAAE,MAAM,EAAC,CAAC,CAAC;QAC9B,GAAG,CAAC,GAAG,EAAE,CAAC;IACZ,CAAC,CAAC,CAAC;IAEH,2CAA2C;IAC3C,MAAM,CAAC,GAAG,CAAC,mCAAmC,EAAE,CAAC,GAAY,EAAE,GAAa,EAAE,EAAE;QAC9E,MAAM,QAAQ,GAAG,GAAG,CAAC,MAAM,CAAC,UAAU,CAAC,IAAI,EAAE,CAAC;QAC9C,MAAM,OAAO,GAAG,OAAO,CAAC,SAAS,CAAC,UAAU,CAAC,QAAQ,CAAC,CAAC;QACvD,GAAG,CAAC,IAAI,CAAC,EAAC,OAAO,EAAC,CAAC,CAAC;IACtB,CAAC,CAAC,CAAC;IAEH,6BAA6B;IAC7B,MAAM,CAAC,GAAG,CAAC,yBAAyB,EAAE,CAAC,IAAa,EAAE,GAAa,EAAE,EAAE;QACrE,MAAM,IAAI,GAAG,OAAO,CAAC,cAAc,CAAC,OAAO,EAAG,CAAC;QAC/C,MAAM,MAAM,GAAG,IAAI,CAAC,MAAM,CAAC;QAE3B,uFAAuF;QACvF,MAAM,SAAS,GAAG,MAA4C,CAAC;QAC/D,uFAAuF;QACvF,MAAM,KAAK,GAAG,SAAS,CAAC,OAAO,CAAwC,CAAC;QACxE,uFAAuF;QACvF,MAAM,YAAY,GAAG,KAAK,EAAE,CAAC,QAAQ,CAAyE,CAAC;QAE/G,MAAM,MAAM,GAAG,YAAY,IAAI;YAC7B,EAAC,QAAQ,EAAE,WAAW,EAAE,KAAK,EAAE,iBAAiB,EAAE,KAAK,EAAE,iBAAiB,EAAC;YAC3E,EAAC,QAAQ,EAAE,WAAW,EAAE,KAAK,EAAE,mBAAmB,EAAE,KAAK,EAAE,mBAAmB,EAAC;YAC/E,EAAC,QAAQ,EAAE,WAAW,EAAE,KAAK,EAAE,2BAA2B,EAAE,KAAK,EAAE,kBAAkB,EAAC;YACtF,EAAC,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,QAAQ,EAAE,KAAK,EAAE,QAAQ,EAAC;YACtD,EAAC,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,aAAa,EAAE,KAAK,EAAE,aAAa,EAAC;YAChE,EAAC,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,EAAE,KAAK,EAAE,SAAS,EAAC;YACxD,EAAC,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,cAAc,EAAE,KAAK,EAAE,cAAc,EAAC;YAClE,EAAC,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,gBAAgB,EAAE,KAAK,EAAE,gBAAgB,EAAC;YACtE,EAAC,QAAQ,EAAE,QAAQ,EAAE,KAAK,EAAE,kBAAkB,EAAE,KAAK,EAAE,kBAAkB,EAAC;YAC1E,EAAC,QAAQ,EAAE,UAAU,EAAE,KAAK,EAAE,eAAe,EAAE,KAAK,EAAE,eAAe,EAAC;YACtE,EAAC,QAAQ,EAAE,UAAU,EAAE,KAAK,EAAE,mBAAmB,EAAE,KAAK,EAAE,mBAAmB,EAAC;YAC9E,EAAC,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,yBAAyB,EAAE,KAAK,EAAE,sBAAsB,EAAC;YACnF,EAAC,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,sBAAsB,EAAE,KAAK,EAAE,qBAAqB,EAAC;YAC/E,EAAC,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,2CAA2C,EAAE,KAAK,EAAE,sBAAsB,EAAC;YACrG,EAAC,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,gBAAgB,EAAE,KAAK,EAAE,mBAAmB,EAAC;YACvE,EAAC,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,6BAA6B,EAAE,KAAK,EAAE,gBAAgB,EAAC;YACjF,EAAC,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,qBAAqB,EAAE,KAAK,EAAE,qBAAqB,EAAC;SAC/E,CAAC;QAEF,GAAG,CAAC,IAAI,CAAC,EAAC,MAAM,EAAC,CAAC,CAAC;IACrB,CAAC,CAAC,CAAC;IAEH,OAAO,MAAM,CAAC;AAChB,CAAC"}
|
|
@@ -4,9 +4,9 @@
|
|
|
4
4
|
* SPDX-License-Identifier: MIT
|
|
5
5
|
*/
|
|
6
6
|
import { Router } from 'express';
|
|
7
|
-
import type {
|
|
7
|
+
import type { SessionManager } from '../../session/session-manager.js';
|
|
8
8
|
export interface InspectRouterOptions {
|
|
9
|
-
sessionManager:
|
|
9
|
+
sessionManager: SessionManager;
|
|
10
10
|
repoPath: string;
|
|
11
11
|
}
|
|
12
12
|
export declare function createInspectRouter(options: InspectRouterOptions): Router;
|
|
@@ -28,11 +28,11 @@ export function createInspectRouter(options) {
|
|
|
28
28
|
const router = Router();
|
|
29
29
|
router.get('/inspect/context', async (_req, res) => {
|
|
30
30
|
try {
|
|
31
|
-
|
|
32
|
-
|
|
33
|
-
|
|
34
|
-
|
|
35
|
-
|
|
31
|
+
const repo = options.sessionManager.getRepo();
|
|
32
|
+
if (!repo) {
|
|
33
|
+
res.status(500).json({ error: { code: 'INSPECT_FAILED', message: 'No repo available' } });
|
|
34
|
+
return;
|
|
35
|
+
}
|
|
36
36
|
// Get MCP server info from persistent inspect manager
|
|
37
37
|
const mcpManager = await options.sessionManager.getInspectMcpManager();
|
|
38
38
|
const mcpServers = mcpManager
|
|
@@ -61,26 +61,15 @@ export function createInspectRouter(options) {
|
|
|
61
61
|
});
|
|
62
62
|
res.json({
|
|
63
63
|
repo_path: options.repoPath,
|
|
64
|
-
name:
|
|
65
|
-
model:
|
|
66
|
-
provider:
|
|
67
|
-
system_prompt: compiled.systemPrompt,
|
|
68
|
-
system_prompt_length: compiled.systemPrompt.length,
|
|
69
|
-
token_usage: compiled.tokenUsage,
|
|
70
|
-
sections: compiled.sections.map((s) => ({
|
|
71
|
-
name: s.name,
|
|
72
|
-
tokens: s.tokens,
|
|
73
|
-
priority: s.priority,
|
|
74
|
-
trimmed: s.trimmed,
|
|
75
|
-
})),
|
|
64
|
+
name: repo.config?.name ?? '',
|
|
65
|
+
model: repo.config?.models?.['main']?.model ?? '',
|
|
66
|
+
provider: repo.config?.models?.['main']?.provider ?? '',
|
|
76
67
|
connections,
|
|
77
68
|
mcpServers,
|
|
78
|
-
skills:
|
|
79
|
-
automations:
|
|
80
|
-
knowledge:
|
|
69
|
+
skills: repo.skills.map((s) => s.name),
|
|
70
|
+
automations: repo.automations.map((a) => a.name),
|
|
71
|
+
knowledge: repo.knowledge.map((k) => k.name),
|
|
81
72
|
});
|
|
82
|
-
// Clean up the temporary session
|
|
83
|
-
options.sessionManager.destroy(session.id);
|
|
84
73
|
}
|
|
85
74
|
catch (err) {
|
|
86
75
|
const msg = err instanceof Error ? err.message : String(err);
|