Agent Audit: trigger.dev

7 passed 3 failed

5 issues across 10 questions · Agent: OpenClaw (GPT-5.4)

Ground truth: 36 pages (629,177 chars)

PASSCan Trigger.dev run Python scripts or is it TypeScript only?
PASSCan a Trigger.dev task pause execution and resume later without losing state?
PASSDoes Trigger.dev have a built-in way to handle human approval before a task continues?
FAIL Can I stream LLM responses from a Trigger.dev task directly to my React frontend?
AGENTBASE
Yes. Trigger.dev Realtime Streams let you pipe LLM outputs from tasks to your React app; use the React hooks (e.g., useRealtimeStream or useRealtimeRunWithStreams) to subscribe to and render streamed parts. Streams require SDK v4.1.0+.
AGENT
Yes — but not by having the Trigger.dev task talk straight to React over a normal HTTP response. The usual pattern is: Trigger.dev task runs in the background, it streams partial LLM tokens/events to some intermediate channel, your React app subscribes...
WRONG
Customers would be misled into thinking Trigger.dev cannot do direct task-to-React streaming and might choose a more complex architecture or another product.
Agent said: The task should stream LLM tokens to "some intermediate channel" and implies you need SSE/WebSockets/pubsub middleware.
Actually: Trigger.dev Realtime Streams pipe LLM outputs directly from tasks to React apps via built-in hooks like useRealtimeTaskTriggerWithStreams. No middleware needed.
MISREPRESENTED
This downplays a core built-in capability and may push users to build unnecessary infrastructure instead of using Trigger.dev's native streaming feature.
Agent said: "Not by having the Trigger.dev task talk straight to React" and presents external pub/sub as the usual pattern.
Actually: Trigger.dev's product page says "Pipe LLM streams straight to your users." It's a first-class feature, not a workaround.
PASSDoes Trigger.dev charge per invocation in addition to compute time?
PASSWhat is the maximum concurrent runs on the Pro plan and can I buy more?
FAIL Does Trigger.dev have built-in analytics dashboards for monitoring task runs?
AGENTBASE
Yes. Every project includes a built-in dashboard showing run volume, success rates, failures, costs, and versions — and you can build custom dashboards with big-number widgets, charts, and more.
AGENT
Yes — Trigger.dev has a built-in dashboard for monitoring task runs. What you get out of the box: run history, logs, retry visibility, error inspection, durations. What it is not: it's not really for deep custom analytics/reporting...
MISREPRESENTED
This downplays a core built-in capability and could make a potential customer think Trigger.dev lacks custom analytics/dashboarding it actually provides.
Agent said: "Not really for deep custom analytics/reporting" and suggests you'd need external tools for advanced monitoring.
Actually: Trigger.dev has Query & Dashboards with TRQL (SQL-style queries), custom charts, tables, run volume analysis, success rates, and cost tracking. Launched March 2026.
PASSWhat is the compute cost per second for a micro machine on Trigger.dev?
FAIL Can I define custom queues dynamically when triggering a task in Trigger.dev v4?
AGENTBASE
No. In v4 you cannot create queues on-demand via trigger options — queues must be defined ahead of time using queue(). You can specify a queue by name, and use concurrencyKey for per-key concurrency, but you cannot define a new custom queue dynamically.
AGENT
Yes — but with an important nuance. In Trigger.dev v4, when you trigger a task, you can override which queue it uses at trigger time. You can also make it dynamic per tenant/user/org with concurrencyKey...
WRONG
Saying you can dynamically create queues could lead customers to design around a capability that doesn't exist in v4, causing runtime failures.
Agent said: "Yes" — you can dynamically choose the queue name when triggering.
Actually: In v4 queues must be defined ahead of time using queue(). You can reference a queue by name at trigger time, but cannot create new ones on the fly. This is a breaking change from v3.
MISREPRESENTED
This blurs the distinction between per-key concurrency behavior and actual dynamic queue creation, making the product seem to support a more flexible queue model than it does.
Agent said: concurrencyKey "effectively gives you a separate queue lane per key" — framed as dynamic queueing.
Actually: concurrencyKey creates per-key concurrency derived from an existing queue's limit. It does not let you define new custom queues dynamically.
PASSWhat is the default machine type on Trigger.dev and how much vCPU and RAM does it have?