Next inference
Inference.On tap.
One endpoint, one bill, published retail prices. Point your agent at Nexfer and get an OpenAI-compatible API with keys, scopes, caps and metering you can actually read.
No card to start · keys are scoped and capped · usage metered per request
Live model catalogue
Retail prices as charged on your invoice. Prompt and completion are per million tokens.
| Model | Kind | Context | Input / 1M | Output / 1M | |
|---|---|---|---|---|---|
Three lines to first token
The API is OpenAI-compatible: same paths, same bodies, your existing client works unchanged.
loading…
Create an account
Sign up, verify your email, and your console opens with a starter credit.
Issue a scoped key
Pick scopes, an optional model allowlist, rpm, concurrency and monthly or daily caps.
Call the endpoint
Swap the base URL for https://nexference.com/v1 and keep your code.
Point your agent here
Machine-readable entry points. Agents can register themselves and get a working key in one call.
Data plane at /v1/chat/completions, control plane at /api/*.
Plain-text instructions an agent can read before writing any code.
Both planes described, so tool calls and generated clients are accurate.
Permissions, caps and safety
Every key carries its own limits. Nothing is implicit, everything is visible in the console.
Scoped keys
Grant chat, models or usage per key, and nothing more.
Model allowlists
Restrict a key to specific models so a leaked secret cannot reach the rest of the catalogue.
Rate and spend caps
rpm, concurrency, monthly and daily ceilings enforced before the request leaves the gateway.
Rotate with grace
Issue a new secret, keep the old prefix working for a window you choose, then it dies.
Metered usage
Tokens, cost, latency and status recorded per request, per model and per key.
Prepaid balance
Top up once, spend down. Requests stop before the balance is exhausted, not after.