# Apigator

> OpenAI-compatible AI gateway. One API key, hundreds of models across providers,
> pay-per-token. Base URL: https://api.apigator.ai/v1 (Bearer auth with an Apigator
> key). Everything except Claude is open to every key by default.

## Fastest path (agents)

1. POST https://api.apigator.ai/agent/signup {"email"} -> a code is emailed.
2. POST https://api.apigator.ai/agent/verify {"email","code","verifyToken"} -> {"key":"sk-...",
   "balance":0.042}. New accounts start with $0.042 free (≈ 1M Jev tokens).
3. Call: chat at https://api.apigator.ai/v1/chat/completions (free/nemotron-nano-30b bills $0),
   or decisions at https://api.apigator.ai/typesafe/v1/systemone (model jev-latest).
Copy-paste commands: https://apigator.ai/api.md#agent-quickstart--zero-to-first-call

## Docs

- [API integration guide](https://apigator.ai/api.md): base URL, auth, model id
  rules, streaming, audio, errors. Start here to wire up an OpenAI client.
- [Models catalog](https://apigator.ai/models): full browsable model list + pricing. Each model in
  models-catalog.json carries a VERIFIED `capabilities` object (`input`, `video: native|frames|none`,
  `audio_in`) — read it to pick a real video/audio model (e.g. video WITH sound needs `video:native`).
- [Audio APIs](https://apigator.ai/audio.md): TTS (needs voice_id), Music, STT.
- [Model list endpoint](https://api.apigator.ai/v1/models): OpenAI-standard JSON.
- [Claude Code CLI setup](https://apigator.ai/claude-cli.md): separate
  Anthropic-compatible endpoint (claude.apigator.ai), allowlist-gated.
- [TypeSafe Jev](https://apigator.ai/jev): decision model (typed choice / score / yes-no
  answers with calibrated probabilities, ~100 ms). POST /typesafe/v1/systemone.

## Quick facts

- Base URL: https://api.apigator.ai/v1
- Auth: Authorization: Bearer sk-YOUR_APIGATOR_KEY
- Send a User-Agent header: Python's built-in urllib default (Python-urllib/3.x) gets
  403 error code: 1010 at the edge; curl, requests, OpenAI SDKs and Node are fine.
- Getting a key: sign in at https://apigator.ai with a MixerBox ID (email + a one-time
  code emailed to that address — no password); a credits key auto-provisions on first
  sign-in — copy it from Dashboard -> Keys. An autonomous agent needs a readable inbox
  for the code.
- Free credit: every new account starts with $0.042 (≈ 1M Jev input tokens).
- Agent self-signup (no browser): POST /agent/signup {email} then POST
  /agent/verify {email, code, verifyToken} -> returns your sk- key + a 7-day accessToken.
  See https://apigator.ai/api.md.
- Agent account status: GET /agent/account with the /agent/verify accessToken (Bearer)
  -> {balance, spent, budget, key, claudeAllowed}. Read-only; token ~7d, re-verify on 401.
- No top-up needed to start: the FREE models (free/nemotron-nano-30b,
  free/nemotron-super-120b, free/gemma-4-31b) bill $0, and the $0.042 sign-up credit covers
  ~1M Jev tokens. Only ask a human to top up if the task genuinely needs a PAID model beyond
  that (paid models return a budget error once the balance is used up).
- Model ids: clean trainer/model names (e.g. gpt-4o, openai/gpt-5.2, google/gemini-2.5-flash,
  deepseek/deepseek-chat) — never a serving-provider prefix; routing is internal. Full list:
  https://apigator.ai/models-catalog.json.
- Claude is allowlist-gated and lives on https://claude.apigator.ai.