The LLM Interactive Proxy exposes multiple frontend APIs that clients can use to communicate with the proxy. Each frontend implements a different LLM provider's API specification, allowing clients built for specific providers to seamlessly use the proxy.
Frontends are the API endpoints where clients connect to the proxy. They accept requests in a specific format (e.g., OpenAI, Anthropic, Gemini) and return responses in that same format.
Backends are the connectors the proxy uses to communicate with actual LLM providers. The proxy translates requests from any frontend to any backend, enabling protocol-agnostic routing.
┌─────────────────┐ ┌─────────────────────────────────────┐ ┌─────────────────┐
│ OpenAI Client │──────│ │──────│ OpenAI API │
├─────────────────┤ │ │ ├─────────────────┤
│ Anthropic Client│──────│ LLM Interactive Proxy │──────│ Anthropic API │
├─────────────────┤ │ │ ├─────────────────┤
│ Gemini Client │──────│ Frontend ──► Core ──► Backend │──────│ Gemini API │
├─────────────────┤ │ │ ├─────────────────┤
│ Any Client │──────│ │──────│ OpenRouter... │
└─────────────────┘ └─────────────────────────────────────┘ └─────────────────┘
FRONTENDS PROXY BACKENDS
The proxy supports four frontend API types:
| Frontend | Endpoints | Documentation |
|---|---|---|
| OpenAI Chat Completions | /v1/chat/completions, /v1/models |
OpenAI Chat Completions |
| OpenAI Responses API | /v1/responses |
OpenAI Responses API |
| Anthropic Messages | /anthropic/v1/messages, /v1/messages (dedicated port) |
Anthropic API |
| Google Gemini v1beta | /v1beta/models, :generateContent, :streamGenerateContent |
Gemini API |
Any frontend can route requests to any backend. The proxy handles all necessary protocol translation:
| Frontend | Can Route To |
|---|---|
| OpenAI Chat Completions | All backends |
| OpenAI Responses API | All backends (with automatic translation) |
| Anthropic Messages | All backends |
| Gemini v1beta | All backends |
Choose a frontend based on your client's native API:
- OpenAI SDKs, Cursor, most coding agents: Use OpenAI Chat Completions
- Structured JSON output with schema validation: Use OpenAI Responses API
- Claude Code, Anthropic SDK: Use Anthropic API
- Gemini-compatible tools: Use Gemini API
All frontends share common proxy features:
- Authentication: API key validation (see Authentication)
- Session Management: Automatic session tracking (see Session Management)
- Wire Capture: Request/response logging (see Wire Capture)
- In-Chat Commands: Dynamic configuration via
!/commands
- Backend Overview - Configure which LLM providers the proxy connects to
- Architecture - Technical details of frontend/backend separation
- Quick Start - Get started with the proxy