RouteForge

OpenAI, Anthropic and Gemini compatible

Every model behind one endpoint.

Create a key, point any OpenAI, Anthropic or Gemini SDK at one URL, and pay per token.

Get startedRead the docs

Base URL

http://localhost:3000/v1

curl http://localhost:3000/v1/chat/completions \
  -H "Authorization: Bearer $ROUTEFORGE_API_KEY" \
  -d '{
    "model": "claude-haiku-4-5-20251001",
    "messages": [{"role": "user", "content": "Hi"}]
  }'
/v1/chat/completions/v1/messages/v1/responses/v1beta/models

Available models

Read live from the provider behind the gateway. These identifiers are what you pass in the model field.

How the catalog works

Text

33

Chat, reasoning and code

  • claude-haiku-4-5-202510011M context
  • claude-opus-4-61M context
  • claude-opus-4-71M context
  • claude-opus-4-81M context
  • claude-opus-51M context

and 28 more

Image

12

Generation and editing

  • gemini-3-pro-image
  • gemini-3.1-flash-image
  • nano-banana-2
  • nano-banana-pro
  • gpt-image-2

and 7 more

How it works

Create a key

Sign up, verify your email, then create a named key for each project.

Point your SDK

Use the gateway URL as the base URL and your key as the bearer token.

Watch usage

Requests, latency and token spend per model, charged to your balance.

Integrate in minutes

Any client that speaks the OpenAI, Anthropic or Gemini protocol works without a proxy or shim. Change the base URL, keep the rest of your code.

Using a coding agent? See setup guides for Cursor, Claude Code, Cline, Roo Code, Continue and Codex.

curl http://localhost:3000/v1/chat/completions \
  -H "Authorization: Bearer $ROUTEFORGE_API_KEY" \
  -d '{
    "model": "claude-haiku-4-5-20251001",
    "messages": [{"role": "user", "content": "Hi"}]
  }'

What you get

Your own keys

Named keys per project, with revoke and rotate.

Streaming

Server-sent events relayed as they arrive, cancelled when you disconnect.

Model control

One live catalog; an administrator can disable a model instantly.

Rate limits

Per-key and per-account limits with clear 429 responses.

Isolated accounts

Keys and usage are visible only to their owner.

Exact metering

Token-level charges, with a ledger that keeps sub-cent remainders.

Security

The upstream endpoint and its credentials never leave the server. Your code and your browser only ever see the gateway.

  • Provider credentials stay server-side.
  • API keys carry 256 bits of randomness and are stored only as hashes.
  • A key is shown once, at creation.
  • Revoked keys stop working immediately.
  • Prompts and responses are not stored; logs hold metadata only.
  • Provider errors are cleaned before they reach you.

FAQ

Which API formats does the gateway accept?

OpenAI Chat Completions and Responses, Anthropic Messages (including token counting) and Gemini generateContent. Bodies are forwarded as sent; the gateway does not translate between formats.

Which models can I use?

The ones listed above, read live from the provider behind the gateway. An administrator can disable individual models.

Do you store my prompts?

No. Only request metadata is logged: time, route, model, status, latency and token counts when the provider reports them.

How are API keys stored?

Only a SHA-256 hash is stored. The full key is displayed once when you create it.

What happens if I hit a limit?

You receive HTTP 429 with a retry-after header. Limits are set by the administrator.

Does it work with Cursor, Claude Code, Cline and others?

Several tools allow a custom endpoint. The documentation has a verified setup guide and the known limits for each.

Start with a key, not a contract.

Prepaid credit, no subscription, every model on the same endpoint.

Get started