fix: make agent run timeout configurable

.env.example

@@ -21,6 +21,8 @@ INTELLIGENCE_API_KEY=
 OPENAI_API_KEY=
 OPENAI_BASE_URL=https://api.openai.com/v1
 OPENAI_MODEL=
+# Maximum duration for Dot turns and scheduled/background agent runs (milliseconds).
+# AGENT_RUN_TIMEOUT_MS=90000
 # Keep the CLI-generated CPK_TELEMETRY_ID with its matching project key.
 # The SDK uses it to connect OpenDots usage to the Intelligence account.
 # CPK_TELEMETRY_ID=

compose.yml

@@ -25,6 +25,7 @@ services:
       SLACK_TEAM_ID: ${SLACK_TEAM_ID:-}
       SLACK_USER_IDS: ${SLACK_USER_IDS:-}
       SLACK_DOT_ID: ${SLACK_DOT_ID:-}
+      AGENT_RUN_TIMEOUT_MS: ${AGENT_RUN_TIMEOUT_MS:-90000}
       OPENAI_API_KEY: ${OPENAI_API_KEY:-}
       OPENAI_BASE_URL: ${OPENAI_BASE_URL:-https://api.openai.com/v1}
       OPENAI_MODEL: ${OPENAI_MODEL:-}

src/server/dot-agent.ts

@@ -18,15 +18,17 @@ import { Observable } from 'rxjs';
 import { z } from 'zod';
 import { Store } from './store.js';
 import { WorkspaceStore } from './workspace.js';
-import type { PlatformConfig } from './platform-config.js';
+import {
+  DEFAULT_AGENT_RUN_TIMEOUT_MS,
+  type PlatformConfig,
+} from './platform-config.js';
 import { browserResponse } from './research.js';
 import { answerObserver, type SetupTelemetry } from './setup-telemetry.js';
 const channelError = () => ({
   type: EventType.RUN_ERROR,
   message:
     'OpenDots could not complete this request. Please check the app and try again.',
 });
-const TURN_TIME_LIMIT_MS = 90_000;
 export class DotAgent extends AbstractAgent {
   private inner?: BuiltInAgent;
   private controller?: AbortController;
@@ -63,17 +65,19 @@ export class DotAgent extends AbstractAgent {
       let timedOut = false;
       let finished = false;
       let configurationFailure = false;
+      const turnTimeLimitMs =
+        this.config.agentRunTimeoutMs ?? DEFAULT_AGENT_RUN_TIMEOUT_MS;
       const observe = answerObserver((event) =>
         this.setupTelemetry?.capture(event),
       );
       const timeout = setTimeout(() => {
         timedOut = true;
         observe({ type: EventType.RUN_ERROR });
         this.abortRun();
-      }, TURN_TIME_LIMIT_MS);
+      }, turnTimeLimitMs);
       const timeLimitError = () => ({
         type: EventType.RUN_ERROR,
-        message: `This turn reached the ${TURN_TIME_LIMIT_MS / 1000} second time limit and was stopped. Try a smaller request.`,
+        message: `This turn reached the ${turnTimeLimitMs / 1000} second time limit and was stopped. Try a smaller request.`,
       });
       try {
         const dot = this.workspace.dot(this.dotId);

src/server/index.ts

@@ -10,6 +10,7 @@ import { resolveAppOrigins } from './app-origin.js';
 import { WorkspaceStore } from './workspace.js';
 import { Platform } from './platform.js';
 import {
+  agentRunTimeoutMsFromEnv,
   intelligenceApiKeyFromEnv,
   intelligenceWsUrlFromEnv,
   type PlatformConfig,
@@ -37,6 +38,7 @@ const config: PlatformConfig = {
   apiKey: process.env.OPENAI_API_KEY,
   model: process.env.OPENAI_MODEL,
   baseUrl: process.env.OPENAI_BASE_URL ?? 'https://api.openai.com/v1',
+  agentRunTimeoutMs: agentRunTimeoutMsFromEnv(process.env),
   webSearchProvider: webSearchProvider(process.env.WEB_SEARCH_PROVIDER),
   parallelApiKey: process.env.PARALLEL_API_KEY,
   browserUrl: process.env.BROWSER_URL,
@@ -82,6 +84,7 @@ const runner = new Runner(
     const text = await platform.turn(threadId, claim.prompt, signal);
     return { text, sources: [], sample: false };
   },
+  config.agentRunTimeoutMs,
 );
 const wsOrigin = new URL(
   config.intelligenceWsUrl ?? 'wss://realtime.intelligence.copilotkit.ai',

src/server/platform-config.ts

@@ -7,6 +7,8 @@ const INTELLIGENCE_API_KEY_ENV_NAMES = [
   'INTELLIGENCE_API_KEY',
 ] as const;
 export const INTELLIGENCE_KEY_MISSING_LABEL = `${INTELLIGENCE_API_KEY_ENV_NAMES[1]} (or ${INTELLIGENCE_API_KEY_ENV_NAMES[0]})`;
+export const DEFAULT_AGENT_RUN_TIMEOUT_MS = 90_000;
+const MAX_TIMER_DELAY_MS = 2_147_483_647;
 
 export function intelligenceApiKeyFromEnv(
   env: Record<string, string | undefined>,
@@ -23,6 +25,19 @@ export function intelligenceWsUrlFromEnv(
   ]);
 }
 
+export function agentRunTimeoutMsFromEnv(
+  env: Record<string, string | undefined>,
+): number {
+  const raw = env.AGENT_RUN_TIMEOUT_MS?.trim();
+  if (!raw) return DEFAULT_AGENT_RUN_TIMEOUT_MS;
+  const value = Number(raw);
+  if (!Number.isInteger(value) || value < 1 || value > MAX_TIMER_DELAY_MS)
+    throw new Error(
+      `AGENT_RUN_TIMEOUT_MS must be an integer between 1 and ${MAX_TIMER_DELAY_MS}.`,
+    );
+  return value;
+}
+
 function firstNonEmptyEnvValue(
   env: Record<string, string | undefined>,
   names: readonly string[],
@@ -41,6 +56,7 @@ export interface PlatformConfig extends WebConfig {
   model?: string;
   apiKey?: string;
   baseUrl: string;
+  agentRunTimeoutMs?: number;
   computerSupervisorUrl?: string;
   computerSupervisorToken?: string;
   computerToken?: string;

src/server/runner.ts

@@ -1,7 +1,8 @@
-import { Store } from './store.js';
+import { DEFAULT_WORKER_LEASE_MS, Store } from './store.js';
 import type { Result, Memory } from '../shared/types.js';
 import type { Claim } from './store.js';
 import { research, type Config } from './research.js';
+import { DEFAULT_AGENT_RUN_TIMEOUT_MS } from './platform-config.js';
 export class Runner {
   private timer?: ReturnType<typeof setInterval>;
   private active = new Map<string, AbortController>();
@@ -14,6 +15,7 @@ export class Runner {
       signal: AbortSignal,
       progress: (text: string) => void,
     ) => Promise<Result>,
+    private runTimeoutMs = DEFAULT_AGENT_RUN_TIMEOUT_MS,
   ) {}
   start() {
     if (!this.timer) {
@@ -53,7 +55,11 @@ export class Runner {
   }
   private async runTick() {
     if (this.active.size) return;
-    const claim = this.store.claim();
+    // Leave time to persist the outcome after the configured run timeout.
+    const claim = this.store.claim(
+      Date.now(),
+      Math.max(DEFAULT_WORKER_LEASE_MS, this.runTimeoutMs * 2),
+    );
     if (!claim) return;
     const controller = new AbortController();
     this.active.set(claim.id, controller);
@@ -68,9 +74,11 @@ export class Runner {
     const timeout = setTimeout(
       () =>
         controller.abort(
-          new Error('Research exceeded the 90 second time limit.'),
+          new Error(
+            `Research exceeded the ${this.runTimeoutMs / 1000} second time limit.`,
+          ),
         ),
-      90_000,
+      this.runTimeoutMs,
     );
     try {
       const settings = this.store.settings();

src/server/store.ts

@@ -14,6 +14,7 @@ import type {
 } from '../shared/types.js';
 
 export type Claim = Task & { lease: string };
+export const DEFAULT_WORKER_LEASE_MS = 180_000;
 const defaults: Settings = {
   name: 'Dot',
   paused: false,
@@ -204,7 +205,10 @@ export class Store {
     );
     return this.task(id);
   }
-  claim(now = Date.now()): Claim | null {
+  claim(
+    now = Date.now(),
+    leaseDurationMs = DEFAULT_WORKER_LEASE_MS,
+  ): Claim | null {
     return this.transaction(() => {
       const settings = this.settings();
       if (settings.paused || !settings.researchAllowed) return null;
@@ -228,7 +232,7 @@ export class Store {
         .prepare(
           "UPDATE tasks SET status='running', lease=?, leaseUntil=?, nextRunAt=NULL, error=NULL, updatedAt=? WHERE id=?",
         )
-        .run(lease, now + 180_000, now, task.id);
+        .run(lease, now + leaseDurationMs, now, task.id);
       this.db
         .prepare(
           "INSERT INTO runs VALUES (?, ?, 'running', ?, NULL, NULL, NULL)",