Deepseek-API

High-performance, 100% native Golang OpenAI & Anthropic compatible API relay server (http://localhost:8000) powered by personal DeepSeek Web accounts.

Designed specifically to serve as a fast, reliable drop-in backend provider for Agentic Coding CLIs (e.g., Claude Code, OpenCode, OpenAI Codex CLI, Cursor, Aider, Continue, etc.) with native support for Anthropic Messages API, OpenAI Chat Completions, DeepThink R1 Reasoning, Live Web Search & Citations, and Multi-Turn Agent Trajectory Tool Calling.

---

⚡ Features

• Pure Go WASM PoW Solver (wazero) — Zero CGo dependencies. Solves DeepSeek sha3_wasm_bg.wasm challenges natively in ~0.16s.
• Dual API Compatibility (Anthropic + OpenAI) — Dual endpoint support for both Anthropic Messages API (POST /v1/messages) and OpenAI Chat (POST /v1/chat/completions).
• Claude Code Integration — Native support for Claude Code CLI and Anthropic SDKs via /v1/messages with thinking blocks (budget_tokens), tool_use JSON objects, and tool_result history.
• OpenAI Codex CLI Compatibility — Native discovery schema for GET /v1/models including reasoning levels, slug, and context windows.
• Auto Token Exchange & Self-Healing — Accepts raw userToken from LocalStorage, automatically exchanges it for short-lived accessToken via /api/v0/users/current, and auto-refreshes on expiration.
• Clean Reasoning Output Parser — Separates THINK and RESPONSE stream fragments dynamically. reasoning_content & thinking are delivered in standard delta formats without corrupting main answer text.
• Live Web Search & Structured Citations — Supports deepseek-search and deepseek-reasoner-search with automatic conversion of [citation:X] tags and structured markdown footnote citations.
• Multi-Format Tool Calling & Trajectory Replay — Parses <tool>...</tool>, <tool:name>, <tool_call>, and <parameter> shapes, while reconstructing multi-turn assistant tool calls and tool_result outputs with anchoring instructions to prevent agent amnesia and duplicate calls.
• Ephemeral Session Auto-Cleanup — Temporary one-off sessions are automatically deleted from DeepSeek's backend via /api/v0/chat_session/delete, keeping your web dashboard clean.
• CORS & Web Frontend Ready — Built-in CORS middleware and preflight handlers for browser web interfaces (Open WebUI, Chatbox, etc.).

---

🌐 Supported Models & Endpoints

Endpoints

Endpoint	Standard API	Supported Features
POST /v1/messages	Anthropic Messages API	Claude Code CLI, Anthropic SDKs, thinking blocks, tool_use, tool_result
POST /v1/chat/completions	OpenAI Chat API	OpenCode, Cursor, Aider, reasoning_content, tool_calls, Web Search, SSE Streaming
GET /v1/models	OpenAI / Codex Models API	Model Discovery (slug, context_window, supported_reasoning_levels)
GET /healthz	Health Endpoint	Server status check ({"status":"ok"}) with CORS headers

Model Catalog

Model Name / Slug	Mode	Capabilities	Description
deepseek-chat / deepseek-v3	Fast Chat	Text	Standard high-speed conversational model
deepseek-reasoner / deepseek-r1	Reasoning	Text + DeepThink	Reasoning model streaming reasoning_content deltas
deepseek-expert	Expert	Text + DeepThink	High-precision model for complex architecture
deepseek-search	Web Search	Text + Live Web Search	Model with real-time internet search and citation links
deepseek-reasoner-search	Hybrid	DeepThink + Live Search	DeepThink R1 combined with live internet search

---

🛠️ Architecture

• cmd/server — High-performance HTTP server entry point (net/http & log/slog).
• pkg/pow — WebAssembly PoW solver executing sha3_wasm_bg.wasm via wazero.
• pkg/auth — Session storage, token extraction, and access token resolution (session/session.json).
• pkg/client — Direct HTTP client driver, /users/current auto-exchange, /chat_session/delete cleaner, and real-time SSE stream parser.
• pkg/agentic — Multi-format tool parser, trajectory prompt replay builder, thinking extractor, and session cache.
• pkg/server — OpenAI & Anthropic REST API handlers (/v1/messages, /v1/chat/completions, /v1/models, /healthz).

---

🚀 Quick Start

1. Requirements

• Go 1.22+
• A DeepSeek Account (free account from chat.deepseek.com)

2. Setting Up Session Token

Create session/session.json (or set DEEPSEEK_TOKEN in .env):

{
  "token": "YOUR_DEEPSEEK_USER_TOKEN_OR_BEARER",
  "cookies": {
    "ds_session_id": "YOUR_SESSION_ID"
  },
  "user_agent": "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36...",
  "captured_at": 1786623559
}

💡 Easy Token Capture:

1. Open chat.deepseek.com in your browser and log in.
2. Open Developer Tools (F12) → Application → Local Storage → chat.deepseek.com.
3. Copy the value of userToken (raw string or JSON {"value":"..."}).
4. Paste it into "token" in session/session.json. The server will handle authentication and auto-refresh automatically!

3. Building & Running

# Clone the repository
git clone https://github.com/joyccn/Deepseek-API.git
cd Deepseek-API

# Build static binary
CGO_ENABLED=0 go build -o deepseek-api-server ./cmd/server

# Run server (default on http://127.0.0.1:8000)
./deepseek-api-server

---

🔌 CLI & Tool Integrations

Claude Code CLI

export ANTHROPIC_BASE_URL="http://localhost:8000"
export ANTHROPIC_API_KEY="dummy-token"
export ANTHROPIC_MODEL="claude-3-7-sonnet-20250219"

claude

Cursor / Aider / OpenCode

• OpenAI Base URL: http://localhost:8000/v1
• API Key: dummy-token
• Model: deepseek-chat, deepseek-reasoner, or deepseek-search

---

📖 Usage Examples

1. Anthropic Messages API with Thinking (Claude Code)

curl http://localhost:8000/v1/messages \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-3-7-sonnet-20250219",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Explain quantum computing in 2 sentences."}],
    "thinking": {"type": "enabled", "budget_tokens": 1024}
  }'

2. Live Web Search with Citations

curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-search",
    "messages": [{"role": "user", "content": "What is the latest release version of Go language in 2026?"}]
  }'

3. DeepThink R1 Real-Time Streaming

curl -N http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-reasoner",
    "stream": true,
    "messages": [{"role": "user", "content": "Solve: 17 * 29 + 143"}]
  }'

4. Agentic Tool Calling

curl http://localhost:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "deepseek-chat",
    "messages": [{"role": "user", "content": "What is the weather in Tokyo?"}],
    "tools": [{
      "type": "function",
      "function": {
        "name": "get_weather",
        "description": "Get weather for a city",
        "parameters": {
          "type": "object",
          "properties": {"city": {"type": "string"}},
          "required": ["city"]
        }
      }
    }]
  }'

---

🧪 Testing

Run comprehensive unit tests across all packages:

go test ./... -v -count=1

---

📄 License

MIT — Open source under the MIT License.

© 2026-Present Joy.