Update
Bring your own model — route agent LLM calls to your own endpoint
TL;DR
For teams that already pay for a model deployment, or need model traffic on infrastructure they control. Paste a base URL, model id and key on the Agents page; the platform test-calls it live before saving, and a dead endpoint never persists. Tokens on your endpoint cost zero platform credits.
What shipped
- Any OpenAI-compatible endpoint — Azure AI, vLLM, OpenRouter, or a model you host yourself.
- A live test call before the configuration saves, so a wrong URL, key or model id fails fast rather than silently.
- LLM tokens on your endpoint zero-rated in platform credits — you pay your provider directly.
- Tool, browser, sandbox and media calls keep metering normally, since those costs are real regardless of which model answers.
- Available on paid plans.
The honest cost picture
BYOM is not free. A platform-infrastructure fee still applies, covering the serverless compute, database and streaming work of running a task — set as a fraction of the default model's blended rate. It's a meaningful discount versus metered credits, not zero.
Frequently asked questions
- Is BYOM completely free?
- No. LLM tokens on your own endpoint are zero-rated in platform credits, but a platform-infrastructure fee still applies for the compute, database and streaming cost of running each task.
- What happens if my endpoint is unreachable when I save it?
- The configuration doesn't persist. Saving runs a live test call first, so a dead URL, key or model id is caught immediately rather than failing later mid-task.