Trigger.dev's chat.agent turns AI chats into durable tasks that survive crashes and redeploys
CodeByPoonam · x · 2026-09-15
The author tested Trigger.dev's newly launched chat.agent with Claude Code: each conversation becomes a long-lived, stateful task that survives page refreshes, redeploys, and crashes without manual state management.
Key points:
- Unlike stateless HTTP chat backends, chat.agent gives every session a durable machine that sleeps when idle and wakes on new messages, keeping in-memory state and on-disk workspace across interruptions.
- Tools can declare needsApproval: true, parking the run on a durable waitpoint until a human approves — then it resumes exactly where it stopped.
- The entire backend is a single chat.agent definition (streamText with claude-sonnet-4-5 in the example), with AI SDK's useChat pointed directly at it, no API route needed.
GA since July 2 and in production since June, it has served millions of sessions and over 84 years of compute; Arena built its Agent Mode on top of it. Trigger.dev is an open-source TypeScript platform (16.3k stars) for durable agents with retries, queues, HITL, and observability.
Related event: Trigger.dev Launches chat.agent for Crash-Resilient AI Chats(2 posts)→
More from coding & agent
- GPT-6 Astra says you can delete scaffolding — just not the control kind — Slight_Republic_4242 · 2026-09-15
- You Don't Need Ultra: Power User's Rule of Thumb Says Low/Medium Tiers Suffice for Daily Tasks — soumitrashukla9 · 2026-09-15
- Stop Asking AI Agents 'Can You Build This?' — Use This Inspection-First Prompt Instead — taherdhanera · 2026-09-15
- Claude Cowork 101: A Beginner Tutorial Written for Academic Researchers — JeremyNguyenPhD · 2026-09-15
- ASC: a fast Android decompiler front-end built for AI agents, 1K stars in days — MG1937 · 2026-09-15
- AI agent toolkit pi hits 105K GitHub stars with unified LLM API and coding agent CLI — earendil-works · 2026-09-15