LLM API Translator

Multi-format LLM API proxy: accept Anthropic Messages API and OpenAI Chat API, forward to any OpenAI-compatible backend (DeepSeek, Groq, Together, etc.), return responses in the original format.

• Version: 1.0.0, published 25 May 2026.
• License: MIT
• Runtime: Node.js 18+

Why

LLM tools speak different API formats. Claude Code and Cursor use Anthropic Messages API. Most cost-effective providers (DeepSeek, Groq, Together) use OpenAI Chat API. Instead of modifying SDKs or maintaining vendor-specific adapters, run this proxy — it translates between formats transparently.

Use cases:

• Run Claude Code against DeepSeek instead of Anthropic's API ($0 vs $100/mo)
• Point any OpenAI SDK at a cheaper backend without code changes
• Bridge Anthropic-format agents to OpenAI-compatible infrastructure
• Experiment with multiple backends by swapping a single environment variable

Quick Start

git clone https://github.com/ahmdngi/llm-api-translator.git
cd llm-api-translator

# Set your API key
export DEEPSEEK_API_KEY="sk-..."

# Start the proxy
node translator-proxy.mjs

The proxy listens on http://localhost:3800.

Usage

Anthropic format (Claude Code, Cursor, etc.)

# One-shot
ANTHROPIC_BASE_URL=http://localhost:3800 claude -p "Hello from DeepSeek!"

# Or set globally
export ANTHROPIC_BASE_URL=http://localhost:3800

OpenAI format (any OpenAI SDK)

curl http://localhost:3800/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "x-api-key: sk-your-key" \
  -d '{
    "model": "deepseek-chat",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

Force a different response format

By default, responses match the input format. Override with DEEPSEEK_RESPONSE_FORMAT:

# Send Anthropic format, get OpenAI format back
DEEPSEEK_RESPONSE_FORMAT=openai node translator-proxy.mjs

Architecture

  ┌─ Input ─────────────┐    ┌─ Normalizer ─┐
  │ POST /v1/messages   │───▶│  Anthropic   │──┐
  │ (Anthropic format)  │    └──────────────┘  │
  └─────────────────────┘                      │
                                               │
  ┌─ Input ─────────────┐    ┌─ Normalizer ─┐  │
  │ POST /v1/chat/      │───▶│  OpenAI      │──┤
  │   completions       │    └──────────────┘  │
  │ (OpenAI format)     │                      │
  └─────────────────────┘                      │
                                               ▼
                                    ┌──────────────────┐
                                    │  Canonical IR    │
                                    │  (format-agnostic│
                                    │   message shape) │
                                    └────────┬─────────┘
                                             │
                                             ▼
                                    ┌──────────────────┐
                                    │  DeepSeek / Groq │
                                    │  Together / etc. │
                                    └────────┬─────────┘
                                             │
                                    ┌────────┴─────────┐
                                    │  Denormalizer    │
                                    │  (back to input  │
                                    │   format)        │
                                    └──────────────────┘

Endpoints

| Endpoint | Method | Description | |----------|--------|-------------| | /v1/messages | POST | Accept Anthropic Messages API, return Anthropic format | | /v1/chat/completions | POST | Accept OpenAI Chat API, return OpenAI format | | /v1/models | GET | List models (format-appropriate names) | | /health | GET | Upstream status and config |

Configuration

| Variable | Default | Description | |----------|---------|-------------| | DEEPSEEK_API_KEY or ANTHROPIC_API_KEY | — | API key for the upstream provider | | DEEPSEEK_BASE_URL | https://api.deepseek.com | OpenAI-compatible API base URL | | OPENAI_MODEL | deepseek-chat | Model name sent to upstream | | PROXY_PORT | 3800 | Listen port | | DEEPSEEK_RESPONSE_FORMAT | (auto) | Override: openai or anthropic | | PROXY_DEBUG | — | Enable debug logging to stderr |

Provider examples

Route to Groq:

DEEPSEEK_BASE_URL=https://api.groq.com/openai \
DEEPSEEK_API_KEY="gsk-..." \
OPENAI_MODEL="llama-3.3-70b-versatile" \
node translator-proxy.mjs

Route to Together:

DEEPSEEK_BASE_URL=https://api.together.xyz/v1 \
DEEPSEEK_API_KEY="..." \
OPENAI_MODEL="mistralai/Mixtral-8x22B-Instruct-v0.1" \
node translator-proxy.mjs

Translation Reference

| Anthropic (inbound) | OpenAI (outbound) | |---------------------|-------------------| | system string or block | messages[0].role="system" | | messages[].content blocks | Flat string or tool role | | tools[].input_schema | function.parameters | | tool_choice auto/any/tool | tool_choice auto/required/function | | tool_use response block | tool_calls in choice | | tool_result in user message | role: "tool" message | | SSE stream | Anthropic events ↔ OpenAI SSE |

Roadmap

• [x] Anthropic Messages API ↔ OpenAI Chat API
• [x] Bidirectional (any input → any output format)
• [ ] Google Gemini API format
• [ ] Configurable model list via env
• [ ] Docker image
• [ ] Rate limiting / auth on inbound

License

MIT