Trigger.dev's durable AI chat documentation being published
Comprehensive guides and architecture references are being added for Trigger.dev's AI Chat integration, detailing how to run Vercel AI SDK completions as durable tasks.
7 team members · 17 developing stories
Comprehensive guides and architecture references are being added for Trigger.dev's AI Chat integration, detailing how to run Vercel AI SDK completions as durable tasks.
The background worker logic responsible for tracking and triggering tasks waiting for deployment is being removed, as the queue responsibility shifts to a dedicated Redis worker.
AI assistants connected to the Trigger.dev MCP server are getting a major capability boost, gaining the ability to query analytics, switch profiles, and control the local dev server—all safely behind a new read-only mode.
Developers can now back their AI SDK `useChat` hooks with Trigger.dev's durable execution, enabling long-running AI conversations that survive page refreshes and network drops.
Trigger.dev's MCP server now lets AI agents query data directly with TRQL, control dev servers, and run in a read-only safety mode that prevents accidental deployments.
A new tutorial shows Trigger.dev developers how to use Nango to handle OAuth for third-party APIs, with GitHub + Claude as a concrete example.
The troubleshooting docs now list retry.onThrow() alongside batchTriggerAndWait() and other wait functions, clarifying that it also acts as a parallel wait in execution.
Developers using Bun can now properly configure Sentry error tracking in Trigger.dev projects, with documentation explaining how to avoid a runtime error caused by esbuild bundling @sentry/node's CJS entry during local development.
Documentation is being expanded to explain how to configure per-task middleware, allowing developers to set up context and locals for specific tasks without relying on global execution layers.
Documentation is being updated to clarify how scheduled tasks handle deduplication, warning developers that keys are scoped globally per project to prevent accidental environment overwrites.
Pinning @types/react and @types/react-dom to specific 18.x versions should eliminate the 100+ TypeScript errors caused by mixed type versions across the project.
Developers should have a clearer understanding of task limits, as the configuration is being updated to explicitly measure CPU compute time rather than total wall-clock duration.
A structural loophole in the API client is being closed, replacing loose validation with strict schemas to prevent the SDK from returning invalid task run data formats.
Replaying runs with text or raw string payloads will no longer silently corrupt data into empty JSON objects, and small non-JSON payloads are now directly editable in the UI.
Expiration settings dictate how long a task should wait in the queue before it is automatically discarded, keeping time-sensitive operations relevant.
A new agent system lets users create AI agents through a setup form, configure messaging platforms like Slack, and have messages automatically routed to containerized OpenClaw instances running on a VPS. A status dashboard provides real-time visibility into agent health and execution history.