Unofficial API Gateway

⚠️ Unofficial & Experimental

This project uses undocumented, unofficial APIs from DeepSeek and Google (Gemini). These are not officially supported by any provider.

• APIs can break or change without notice
• Rate limits and throttling apply
• Credentials (tokens/cookies) expire and must be refreshed
• Not affiliated with DeepSeek or Google
• For prototypes, research, and personal projects only

OpenAI-compatible REST API for DeepSeek and Gemini with multi-profile dynamic load balancing.

📖 Architecture Overview — project structure, lifecycle, client types, streaming, auth, Docker

🔄 Provider Conversion Details — OpenAI field mapping per provider

🔑 Authentication & Session Management — API key management, Swagger UI setup, conversation context isolation

Getting Started

1. Pick a provider → create profiles or follow its credential guide
2. Copy and fill .env:

   cp .env.example .env

3. Run the server:

   ./run.sh                  # local
   # or
   docker compose up -d      # Docker

4. Open Swagger UI: http://localhost:8088/docs
5. Generate an API key (required for chat requests if DISABLE_AUTH=false):

   curl -X POST http://localhost:8088/v1/keys/generate \
     -H "Content-Type: application/json" \
     -d '{"name": "my-key"}'

   Save the returned api_key — pass it as Authorization: Bearer <key> on every request.

Provider	GitHub Repository	Credentials Guide	Specific Endpoints
DeepSeek	2noScript/deepseek-api	docs/deepseek.md	reasoning_content
Gemini	HanaokaYuzu/Gemini-API	docs/gemini.md	Chats, Gems, Deep Research

Profiles & Load Balancing

Credentials are stored and managed via Profiles saved in data/profiles.json (or configured via API).

• Multi-Profile Load Balancing: Chat completion requests automatically load-balance across active profiles (is_active: true) using a Round-Robin algorithm.
• Session-Profile Sticky Affinity: When passing X-Session-Id, the conversation binds to a specific profile to ensure multi-turn context continuity. If the profile becomes inactive, auto-failover seamlessly routes to another active profile.

Profiles API (/v1/profiles)

Endpoint	Method	Description
POST /v1/profiles	POST	Create a new profile (deepseek requires token, gemini requires cookie)
GET /v1/profiles	GET	List all profiles (optional query parameter ?type=deepseek or ?type=gemini)
GET /v1/profiles/{id}	GET	Get profile details by ID
PUT /v1/profiles/{id}	PUT	Update profile name, credentials, or is_active status
DELETE /v1/profiles/{id}	DELETE	Delete a profile by ID

Configuration

cp .env.example .env

Env var	Required	Description
UNOFFICIAL_API_DATA_DIR	❌	Directory for profiles.json, api_keys.json, sessions.json, machine_id. Default: data
DISABLE_AUTH	❌	Disable API key authentication in development mode (true/false). Default: true
SESSION_TTL_DAYS	❌	Session lifetime in days after last use. 0 = never expire. Default: 7
SESSION_MAX_SESSIONS	❌	Max sessions kept in memory. Default: 5000
API_KEY_SECRET	❌	HMAC secret for API key signing. Change in production.

Run

Docker

# Run
docker compose up -d
docker compose logs -f
docker compose down

Local

./run.sh

Common (OpenAI-compatible) Endpoints

These share the same format across all providers.

Endpoint	Method	DeepSeek	Gemini
GET /v1/{provider}/models	GET	✅	✅
POST /v1/{provider}/chat/completions	POST	✅	✅

Request body

Field	Type	Required	Default	Description
model	string	✅	—	Model ID (e.g. deepseek-v3, deepseek-r1, gemini-3-flash, gemini-3-pro)
messages	array	✅	—	[{"role": "user", "content": "..."}]
stream	bool	❌	false	Enable SSE streaming

Response

{
  "id": "chatcmpl-1719000000",
  "object": "chat.completion",
  "created": 1719000000,
  "model": "deepseek-v3",
  "choices": [
    {
      "index": 0,
      "message": { "role": "assistant", "content": "..." },
      "finish_reason": "stop"
    }
  ],
  "usage": { "prompt_tokens": 0, "completion_tokens": 42, "total_tokens": 42 }
}

Streaming SSE format:

data: {"choices": [{"delta": {"content": "..."}}]}
data: [DONE]

System

Endpoint	Method	Description
GET /health	GET	Provider connection status
GET /	GET	Redirects to Swagger UI

API Key Management

Endpoint	Method	Description
POST /v1/keys/generate	POST	Generate a new API key
GET /v1/keys	GET	List all API keys (masked)
POST /v1/keys/revoke	POST	Deactivate an API key

See docs/auth.md for full details.

Examples

# DeepSeek
curl -s http://localhost:8088/v1/deepseek/chat/completions \
  -H "Authorization: Bearer <key>" \
  -H "Content-Type: application/json" \
  -d '{"model": "deepseek-v3", "messages": [{"role": "user", "content": "Hi"}]}'

# Gemini
curl -s http://localhost:8088/v1/gemini/chat/completions \
  -H "Authorization: Bearer <key>" \
  -H "Content-Type: application/json" \
  -d '{"model": "gemini-3-flash", "messages": [{"role": "user", "content": "Hi"}]}'

# Session persistence with Sticky Profile Affinity
curl -s http://localhost:8088/v1/gemini/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "X-Session-Id: my-chat-session-001" \
  -H "Content-Type: application/json" \
  -d '{"model": "gemini-3-flash", "messages": [{"role": "user", "content": "What did I just ask?"}]}'