OpenAI-compatible LLM gateway

One API key for Codex and Claude.

LLMApi puts your local AI engines behind a single OpenAI-compatible endpoint — issue scoped keys, enforce quotas and concurrency, stream responses, generate images, and meter every token. Drop it into any OpenAI SDK.

Open the console

One gateway. Every engine.

Your apps speak the OpenAI API. LLMApi sits in the middle — it authenticates the key, enforces scopes and quotas, meters every call, then routes it to the right local engine through its adapter.

Your apps

OpenAI-compatible clients

PythonNode.jscurlGo

Change the base URL + key. Nothing else.

LLMApi gateway

the door between your apps and engines

API keys Scopes Daily quotas Concurrency Usage metering Streaming

Local engines · via adapters

Codex

GPT-5 family

live

Claude

Opus · Sonnet · Haiku

live

Everything a gateway needs

OpenAI-compatible

Point any OpenAI SDK at /v1 — chat completions, token-level streaming and image generation work unchanged.

Codex + Claude

Two AI engines behind one unified interface, with live per-account model discovery — no unsupported-model surprises.

Sensible API keys

SHA-256-hashed keys with scopes, daily quotas, concurrency limits and instant revocation.

Usage metering

Every request is metered — requests, tokens and cost, broken down by model and by individual key.

Locked-down by default

Each request runs read-only, with no network, in a throwaway workspace. One fresh run per call.

Streaming & images

Server-sent token streaming plus aspect-ratio image generation, all behind the same key.

Drop-in OpenAI compatibility

Point any OpenAI SDK at your gateway — change the base URL and the key, nothing else.

bash

curl https://your-host/v1/chat/completions \
  -H "Authorization: Bearer sk-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-sonnet-4-6",
    "messages": [{ "role": "user", "content": "Hello!" }],
    "stream": true
  }'

/v1

OpenAI-compatible surface

AI engines: Codex + Claude

SHA-256

Hashed API keys

Plaintext secrets stored

Frequently asked

Is it really OpenAI-compatible?

Yes. /v1/chat/completions, /v1/models and /v1/images/generations follow the OpenAI wire format, so existing SDKs work by changing only the base URL and key.

Which models can I call?

Whatever your Codex and Claude accounts expose — the model list is discovered live from each CLI, so you never hit an unsupported-model error.

How are API keys secured?

Only a SHA-256 hash is stored; the secret is shown once at creation. Keys carry scopes, daily quotas and concurrency limits, and can be revoked instantly.

Can I see how much each key uses?

The console shows requests, tokens and cost over time, broken down by model and by individual API key.

Issue a key and make your first call.

Open the console