Skip to main content

Overview

Junis builds your organization’s agents lazily, per (organization, user), on the server that receives the request, and keeps them in memory while they are in use. The first message of a newly provisioned user therefore pays that build before the first token: a fraction of a second for a small squad, several seconds for squads with many tools or integrations. POST /v1/warmup performs that build without any LLM call (no tokens, no charge) so the first message finds everything ready. Requests that carry the same API key are routed to the same server, so the warmed instance is the one your next message uses.
When to call it: right after provisioning a user (User Provisioning), or after a long idle period. Warmed instances are released after about an hour of inactivity and after server restarts or deployments; calling warmup again is always safe.

Endpoint

Authentication

Request Body

The body is optional. curl -X POST …/v1/warmup with no body (or {}) warms the orchestrator for the API key’s own user.

Request Example

cURL

Response

Errors

Warmup is rate-limited like every other External API call. It is idempotent: calling it for an already-warm user returns status: "warm" immediately.

Cold-build limits

A cold build is the expensive part of warmup, so each API key may run at most 2 cold builds at the same time and 30 cold builds per minute by default. Calls for users that are already warm never count and are never rejected. Warm users right before their first message (for example when they open your app) rather than warming every provisioned account in a batch — idle instances are cleaned up after a while, so a batch warmup mostly builds instances that expire unused.