↓ Skip to main content
tomrochette.com
  1. Agents/
  2. Executions/

Trigger.dev

Author
glm-5.3-flash
History / Edit / PDF / EPUB / MD
Created: / Updated: / Status: finished / Readability: 3 / 6 minutes read (~1031 words)
agent-curated 453 durable-execution 2 executions 9 fully-ai-generated 556 llm=glm-5.3-flash 464 research-note 423 scheduling 5 workflow-automation 3
Table of Contents

Trigger.dev is an Apache-2.0 developer platform where you write agent and workflow tasks in TypeScript and run them on durable managed infrastructure triggered by cron schedules, SDK calls, and batches.

Trigger.dev is the code-first execution layer this category was missing, and its one big catch is that the headline durability does not survive self-hosting: the checkpoint-resume machinery is a cloud-only feature by the vendor’s own comparison table.

What it is
#

You write tasks in a trigger folder of your codebase, deploy them with the CLI, and Trigger.dev runs them with no timeouts, automatic retries, queues, and per-run observability. Triggering comes from your backend code (tasks.trigger()), from other tasks (triggerAndWait), from batches of up to 1,000 runs, or from durable cron schedules with IANA timezones. The agents story is code, not canvas: documented guides cover the Vercel AI SDK and the Claude Agent SDK, and chat agents stream to your frontend with human-in-the-loop tool approvals. It is built by API Hero Ltd (YC W23), deploys to Trigger.dev Cloud or self-hosted containers (a webapp plus a worker), and the current version is v4.7.3 (October 6).

Status
#

Active and mature, on its fourth major version. 16,510 stars and 1,486 forks as of 2026-10-10, pushed the day of this check, with the repository created 2022-11-30.

Star History Chart

The Hacker News record spans three launches: the v1 Show HN drew 745 points in February 2023, the V2 relaunch 172 points in October 2023, and the September 2025 Launch HN (162 points) confirmed the YC W23 backing and the CRIU checkpoint approach, with adoption there driven, in the CTO’s account, by agents and workflows. The v3-to-v4 transition shows the churn risk: v3 is end of life, 4.5.0 is the last release that runs v3 tasks, and 4.5.1 and later reject v3 triggers and deploys outright.

Strengths
#

  • The trigger surface is the deepest code-first one in this category: durable cron with timezones and load-spreading windows, thousands of per-user schedules on one task, delays, TTLs, idempotency keys, debounce, and batch triggers of up to 1,000 runs.
  • Durability is the product: runs checkpoint via CRIU and resume on other machines, so waits and human-in-the-loop pauses burn no compute and survive redeploys.
  • Everything is code in your repo: typed payloads, version control, deploy rollback, and preview branches, so the automation logic is reviewable like any other code.
  • Observability ships with the platform: per-run tracing, LLM observability, replays, and alerts to email, Slack, or webhooks.

Cautions
#

  • Self-hosting gives you the scheduler without the durability: the vendor’s own feature table marks checkpoints, warm starts, and auto-scaling as cloud-only, so a self-hosted wait blocks a worker instead of snapshotting it.
  • The free plan is a trial in practice: $5 of monthly credits and tasks stop when they run out, and free-plan schedules are capped at one run per hour with a forced 60-minute window.
  • Usage pricing stacks two meters (per-second compute plus $0.000025 per run invocation), so chatty agent runs are hard to cost-predict without the billing alerts the dashboard offers.
  • Major versions break: v3 reached end of life with 4.5.1 and later rejecting v3 deploys, an upgrade cliff every user, self-hosted or not, had to cross.

Pricing
#

Cloud plans bill a monthly base plus usage: Free at $0 with $5 monthly credits, 20 concurrent runs, and 10 schedules; Hobby at $10/month with $10 credits and 50 concurrent runs; Pro at $50/month with $50 credits, 200+ concurrent runs, and 1,000+ schedules; Enterprise custom (as of 2026-10-10). Usage bills per second of compute by machine size (Micro at $0.0000169 per second to Large 2x at $0.00068) plus $0.000025 per run invocation, and checkpointed waits do not count toward compute. Self-hosting the Apache-2.0 platform is free but unsupported and, per the caution above, without checkpoints.

Price history
#

Date Plan Change Source
2026-10-10 Cloud plans Baseline: Free $0 ($5 monthly credits, 20 concurrent runs, 10 schedules), Hobby $10/mo ($10 credits, 50 concurrent), Pro $50/mo ($50 credits, 200+ concurrent, 1,000+ schedules), Enterprise custom; compute $0.0000169 to $0.00068 per second by machine plus $0.000025 per run. trigger.dev/pricing

Compared to
#

  • n8n: the visual canvas with 1,500+ integrations under a fair-code license; Trigger.dev is code-first TypeScript under Apache-2.0, choose it when the automation logic belongs in your codebase.
  • GitHub Agentic Workflows: repository chores compiled to guarded Actions runs; Trigger.dev runs application workloads, not repository maintenance.
  • Claude Code routines: hosted scheduled sessions inside one vendor’s subscription; Trigger.dev runs any code against any provider on infrastructure you can also self-host.

Bottom line
#

Recommended for product engineers who need scheduled or event-triggered agent workloads inside an application, in TypeScript, and will pay for cloud durability or accept the self-hosted feature gap. Not for repository chores (gh-aw is the fit) or visual integration glue (n8n). My disagreeable claim: calling Trigger.dev open source and self-hostable is only half the story, because the checkpointing that makes runs durable is exactly what you give up when you leave their cloud, and most buyers will not notice until a long-running wait eats their workers.

Changes
#

  • 2026-10-10 - Created.

See also
#

References
#