Trigger.dev

The v4.5.0 release candidate series has reached rc.6, bringing durable chat agents and versioned AI prompts closer to stable.

Chat agent primitives keep expanding. chat.agent persists conversations in a Session that survives restarts and redeploys. The rc series added a tools option to thread tool history correctly across turns, useTriggerChatTransport as a drop-in Vercel AI SDK ChatTransport, and chat.headStart to run the first LLM turn server-side while the agent boots — measured at cutting cold start from 2801ms to 1218ms. AI SDK v7 (ESM-only) is now supported alongside v5 and v6; install @ai-sdk/otel on v7 to keep telemetry spans flowing.

AI Prompts let teams version templates without redeploying. Define with prompts.define({ id, model, config, variables, content }); every deploy snapshots the version. Dashboard overrides swap text or model per environment. Spread resolved.toAISDKTelemetry() into any generateText / streamText call to link each span to its prompt version.

trigger init now wires up your AI coding assistant. The setup flow lets you pick the MCP server, agent skills, or both, and scaffolds the project. A new getting-started skill teaches the assistant how to bootstrap a Trigger.dev project end to end.

Dashboard and reliability improvements. The runs page now updates statuses in place and banners new runs live. Uncaught exceptions fail attempts fast instead of drifting to MAX_DURATION_EXCEEDED. The Errors page fingerprints failed runs into groups and supports bulk replay after a fix ships.

Infrastructure additions. Input streams (streams.input<T>) added bidirectional typed communication with running tasks, with four receiving patterns from non-blocking to suspending. AWS PrivateLink support lets tasks reach private VPC resources (RDS, ElastiCache, internal APIs) without public endpoints. Trigger.dev Cloud now meets HIPAA Business Associate requirements — sign a BAA to run workloads that process PHI.