Trigger.dev is an open-source TypeScript orchestration platform for AI workflows and agents, overcoming Serverless timeout limits via Checkpoint-Resume, with native LLM SDK integration, human-in-the-loop, auto-retry, and full observability.
Overview#
Trigger.dev is an open-source platform maintained by the triggerdotdev organization, focused on building, deploying, and managing long-running AI agents and workflows in TypeScript. Licensed under Apache-2.0, currently at v4.4.5 (617 releases), supporting both cloud-hosted and self-hosted modes.
Core Capabilities#
No Timeout Execution#
Unlike AWS Lambda and Vercel, Trigger.dev has no execution time limits, suitable for video processing, multi-day email campaigns, and long-chain AI inference.
Checkpoint-Resume Mechanism#
Based on CRIU (Checkpoint/Restore In Userspace) technology, snapshots memory, CPU registers, and file descriptors when tasks wait for subtasks or pause, releasing resources and restoring precisely from breakpoints when events trigger.
High Reliability#
- Auto-retry: Configurable exponential backoff, max attempts, randomization
- Idempotency: Idempotent task definitions to prevent duplicate execution
- Atomic deployments: Each deployment creates an immutable atomic version; running tasks unaffected by new code
AI Agent & Workflow Orchestration#
- Multi-paradigm: Autonomous agents, prompt chaining, routing, parallelization, multi-agent orchestration, evaluator-optimizer loops
- Native Tool Calling: Deep integration with OpenAI, Anthropic, Vercel AI SDK, OpenAI Agents SDK (TS & Python), Claude Agent SDK, Mastra
- MCP Protocol: Built-in MCP (Model Context Protocol) server support
- Skills & Rules: Definable agent skills and rules
Human-in-the-Loop#
- Real-time streaming: Stream LLM responses to frontend in real-time
- Waitpoints: Pause tasks at critical decision points for human approval, rejection, or feedback
- Input Streams: Interact with running tasks via streaming interfaces
Concurrency & Scheduling#
- Task-level concurrency rule configuration
- Priority queues and batch triggering (
batchTrigger()) - Persistent Cron scheduling, up to one-year cycles
Runtime Extensions#
- Execute Python scripts within tasks (auto-installing requirements.txt)
- Built-in FFmpeg binary for video processing
- Puppeteer, Browserbase, Lightpanda browser automation
- esbuild plugins, apt-get system packages, Prisma, LibreOffice build extensions
Observability#
- Full tracing and logging per run
- Tags, metadata, advanced filtering
- Real-time alerts (email, Slack, Webhook)
- Dashboard and query tools
Typical Use Cases#
- AI Agent development: Multi-model orchestration, tool calling, multi-step reasoning
- Data pipelines & ETL: Large-scale timeout-free data processing
- Media processing: Batch audio/video/image processing
- AI media generation: DALL·E, Fal.ai, Replicate, etc.
- Marketing automation: Multi-day email campaign orchestration
- Human-in-the-loop workflows: Critical decision flows requiring human approval
- Webhook handling: Stripe, Supabase, custom webhooks
- Platform migration: Guides from Mergent, n8n, Temporal
Architecture#
Serverless architecture with isolated execution environments, elastic scaling via task queues, schedulers, and worker pools, supporting multi-region worker deployment. Engineering uses pnpm workspace + Turborepo for monorepo management with apps/, packages/, internal-packages/, docker/, hosting/ directories, and Changeset for versioning.
Quick Start#
Cloud-hosted:
npx trigger.dev@latest login
npx trigger.dev@latest init
npx trigger.dev@latest dev
npx trigger.dev@latest deploy
Self-hosted: Docker Compose and Kubernetes (Helm Chart) options, CLI supports multi-profile for different environments.
Compatible frameworks: Next.js, Node.js, Bun, Remix, SvelteKit
Unconfirmed Information#
- Commercial pricing details not confirmed in collected pages
- CRIU implementation code paths not deeply verified
- Known users' (Supabase, Resend, Unkey) deployment scale unconfirmed
- MCP Server specific tools/protocols details not deeply collected