Overview
Recent usage
Per model, per day. Costs are exact integer micro-dollars from the ledger.
Chats never leave this browser. Each reply is billed from your credits.
API keys
Keys authenticate inference traffic. We store only a hash — each key is shown once, at creation.
Create a key
This is the only time the full key appears. Store it now.
Your keys
Usage
Every request that cost anything is here. Nothing else about it was kept — no prompts, no IPs, no fingerprints.
By day and model
Billing
Usage-based first: credits in, per-token charges out, one ledger. A subscription only changes how credits arrive — at a discount, monthly — never what anything costs.
Subscription
Optional. Each renewal grants credits worth more than the price — plus a higher rate limit.
Models
The live catalog, straight from GET /v1/models — a public endpoint, no key required, so anyone can audit prices before signing up.
Catalog
Text is priced per million tokens; images per image; video per clip — charged from your credit balance per request. Deprecations get a published removal date — models never disappear silently.
Privacy tiers
The suffix on a model name is a routing decision, not marketing.
| anonymous | Private by policy. No prompt logging, no IPs, no fingerprints — the same zero-retention rules as every tier, enforced by what the schema cannot store. |
| tee | Hardware pilot — no live models yet. Served only by enrolled nodes that presented attestation evidence for a published measurement. Every response carries the serving node's identity and measurement. |
| e2ee | Hardware pilot — no live models yet. You seal the prompt to one attested node's public key before it leaves your machine. The gateway relays bytes it cannot read and bills from the node's signed receipt. |
Docs
The short version — enough to make a first request. The full reference, including image, video, errors and the privacy architecture, lives at /docs (public, no login).
Base URL & auth
Authenticate with an API key (create one under API keys) in a standard bearer header. Session logins never touch the inference API.
Quickstart — curl
Python — official OpenAI SDK
Node — official OpenAI SDK
Images & video
Image generation is OpenAI-shaped; video takes a prompt and returns a URL. Both are billed flat per generated unit at the catalog price — no token meters.
Attested tiers — hardware pilot
The tee and e2ee tiers — attested confidential-computing nodes, sealed requests, signed-receipt billing — are built and tested, but no catalog model carries those labels until the first attested hardware nodes onboard. A label you can't verify is worthless. Discovery is already public API:
Errors
Errors carry a code and a request id, never an echo of your request — error bodies are a classic place prompts leak, so ours can't.
| HTTP | Code | Meaning |
|---|---|---|
| 400 | bad_request | Malformed body or parameters. |
| 401 | unauthorized | Missing, wrong, or revoked API key. |
| 402 | insufficient_credits | Balance too low. Nothing was charged — top up under Billing. |
| 404 | model_not_found | No such model in the catalog. |
| 410 | model_retired | Model passed its published removal date. |
| 429 | rate_limited | Over your plan's requests/minute. Keyed to your account, never your IP. |
| 503 | no_route_available | Model exists but no healthy node can serve it right now. Retryable. |
| 502 / 504 | upstream_error / upstream_timeout | A serving node failed or timed out. |