See your real LLM cost and cheaper on-par alternatives for free while your tracked model spend is under $300/mo. Above that, plans scale with your spend — and stay a small single-digit-percent slice of the bill they help you cut.
One tier per account. Each band is a superset of the one before it.
Everything you need to see what you run. No credit card.
First real production usage.
Scaling product teams.
High-spend platform teams.
Compliance & very high spend.
Optional, and independent of your tracked spend. Watch every model you run and get deprecation & retirement alerts pushed to webhook, Slack, or PagerDuty — so you never learn about a retirement from a 4xx in production.
The rolling-30-day cost of the model calls your telemetry agent reports, priced at ingest from published provider rates. You stay free while that's under $300/mo — the threshold measures itself from your own usage, so nothing to configure.
Ingestion returns a clear 402 and the dashboard prompts you to pick the band that matches your spend. Your existing data is never re-priced or deleted — historic cost is frozen at ingest.
A raw percentage grows unbounded and feels like a tax on success. Banded fixed prices are budgetable and the effective rate falls as you grow (Starter ≈1% of its ceiling, Scale ≈0.5%) — the right incentive to expand.
No. The $5/mo Pro reliability add-on unlocks unlimited watches and webhook/Slack/PagerDuty push independent of your tracked spend — so cheap, hands-off reliability is never hostage to a spend conversation.
Never. The agent sends metadata only — model id, token counts, timestamp, and an optional environment tag. Prompts and completions never leave your process.
Create an account, drop instrument() around your client, and watch your cost dashboard fill in. Upgrade only when your spend says it's worth it.