Sentry for AI agents
Catch runaway agent loops and surprise API bills before your users do.
Two-line SDK. Workflow-level runs, not per-call traces. Real-time alerts by email on Starter; Slack and PagerDuty on Pro and Scale — triage from alert to root cause without leaving your stack.
- 25K runs/mo free
- No credit card
- Anthropic & OpenAI
- Framework-agnostic
Trakr Alert: Cost spike — invoice-extractor
Run cost $0.84 exceeded your $0.50 threshold — open run to investigate.
Loop detected on code-review-agent
View run in Trakr →
Production blind spots
Your agents ship to production. Your visibility doesn't.
Teams shipping AI workflows hit these failure modes before they find Trakr — or after they've already paid the bill.
The API bill lands before anyone notices the spike
Your agent 10×'d token usage overnight. Stripe emailed you — your trace dashboard still looks normal.
The same tool call fails hundreds of times — and nothing pages you
Same tool call, same error, on repeat. Per-call tracing logs each attempt; it doesn't flag a runaway loop.
Fifty thousand spans — still no answer on which workflow is bleeding cash
LangSmith and Langfuse excel at debugging calls. You still can't tell which workflow is driving the bill.
How billing works
We bill workflow runs, not every LLM call
One startRun() or @track_agent invocation is one billable run — the full workflow from start to finish.
Every LLM and tool call inside that run is tracked for cost attribution, loop detection, and alerts. You are not charged per internal observation.
1 Trakr run — invoice-extractor
├── LLM · tool:search
├── LLM · tool:extract
├── LLM · retry
└── … more events inside one runTypical agent: ~10 LLM calls per run. 25K Trakr runs can mean 250K+ API events.
Product tour
From alert to root cause in one workflow
Triage-first: resolve incidents, then drill into Overview and workflow-level cost.
From fire to fix without context switching
Built for production agent ops
Loop detection & alerts
Loop detection on every run. Starter: cost and error alerts by email. Pro and Scale: all rule types plus Slack and PagerDuty.
Learn moreWorkflow cost attribution
Per-run, per-workflow, and per-tool cost with 24h deltas on Overview.
Learn moreRunning in production in under 5 minutes
Install the SDK, send your first run, and set up alerts — no framework lock-in.
Install SDK
npm install @trakr/monitorSend first run
import Trakr from '@trakr/monitor';
Trakr.init();
const run = Trakr.startRun('invoice-extractor');
// ... your agent code ...
await run.end({ status: 'success' });Set up alerts
// In the Trakr dashboard:
// Onboarding → cost spike preset (email on Starter)
// Or Alerts → Create rule → cost, error (Starter); loop, latency (Pro+)
// Slack & PagerDuty notifications on Pro and ScaleWhen an alert fires: view run → compare vs previous (Pro/Scale) → resolve or edit rule.
Why teams choose Trakr
Workflow-level ops and alerts without running your own observability stack.
| Aspect | Trace-first tools | Trakr |
|---|---|---|
| Unit | Per span / observation | Per workflow run |
| Pricing | Often per-seat | Flat team price |
| Loops | Manual analysis | Built-in loop detection |
| Triage | Search traces | Alert → run → compare |
Self-hosting Langfuse is free; Trakr is for teams who want managed workflow ops and alerts without running infra.
Meet your team where they already work. Slack and PagerDuty on Pro and Scale; email, REST, Grafana, and CI export on all plans.
Slack · Email · PagerDuty · REST API · Grafana · GitHub Actions
Start free in production
25K workflow runs per month free — no credit card. Pro is $79/mo per org (up to 3 users), not per seat.
Starter
Solo dev validating agents in production
For individual developers validating agents in production.
- 25K runs/mo
- 14 days retention
- 1 user
Pro
Small team shipping agents
For teams shipping agents to production with real traffic.
- 100K runs/mo
- 90 days retention
- 3 users
Upgrade anytime in the dashboard
Scale
Multiple pipelines at volume
For companies running multiple agent pipelines at volume.
- 1M runs/mo
- 1 year retention
- Unlimited
Upgrade anytime in the dashboard
Priced per workflow run, not per LLM call. See full pricing
Common questions
- What counts as a run?
- A run is one complete agent workflow execution — from startRun() to end(). Each LLM call inside a run is tracked separately for cost, but billing counts one run per workflow execution.
- How is Trakr different from Langfuse?
- Trakr is built for managed workflow-level ops: one billable run per agent execution, built-in loop detection, and alert-first triage from alert to run to compare. Langfuse is stronger on evals, prompts, and self-hosted MIT deployments when you want to run your own stack.
- What happens when I hit my limit?
- Starter pauses alert notifications at 25,000 runs per month — runs keep ingesting with no data loss. Pro and Scale allow overage at $0.0008 per additional run.
Start monitoring your agents today
Free tier includes 25K runs per month. No credit card required.