Set up in four steps
- Mint a token for a seat — a principal (who), a tenant (their org), and a spend cap.
- Hand over the secret — it is shown once, on creation. The user pastes it into the desktop app (Settings → Proxy key) or sends it as
Authorization: Bearer. - Set an org pool — a shared ceiling across every seat in the org, on top of each seat's own cap.
- Watch usage — durable per-seat and per-org spend for the current window, across all replicas.
Data-plane endpoints
Every call authenticates with a minted token as Authorization: Bearer zmc-….
/v1/messagesAnthropic-Messages-compatible chat, streaming included./v1/chat/completionsOpenAI-compatible chat — same auth, budgets, and metering./v1/embeddingsOpenAI-compatible embeddings (when configured)./v1/modelsModels this deployment advertises./v1/budgetThe caller's own seat + org budget status — apps draw budget bars from it.Tokens
0| Seat | Org | Usage | Expires | Key | Status |
|---|
No tokens match that filter.
No tokens yet
Mint the first seat token to let a user through the proxy.
Org pool budgets
A shared ceiling across every seat in an org, on top of each token's own cap. A request is blocked if either the seat cap or the org pool is exhausted.
| Org | Pool | Used | Remaining | Window |
|---|
No org pools yet
Seats spend against their own caps until an org pool adds a shared ceiling.
Durable fixed-window spend, per org and per seat — consistent across replicas and reset on the displayed schedule.
No usage yet
Mint a token and make a call — spend shows up here per org and per seat.
Models
Limits & defaults
Platform
read-only
Values are set with PROXY_* environment variables on the
deployment and change via a rollout, not from the console.