API Compatibility Endpoints

When launched with jeikcode --host or jeikcode serve, JeikCode exposes standard LLM-compatible HTTP endpoints on the same port alongside the WebUI. Third-party clients, IDE plugins, or external automated systems can connect directly using standard AI API formats.

---

1. Supported Endpoints

Method	Path	Protocol & Description
POST	/v1/chat/completions	OpenAI Chat Completions compatible protocol (supports SSE streaming, thinking blocks, and tool calls)
POST	/v1/responses	OpenAI Responses native streaming protocol
POST	/v1/messages	Anthropic Claude Messages protocol
GET	/v1/models	List all currently mounted models (IDs formatted as account/model)

---

2. Authentication & Execution Mechanics

• Authorization Header: Pass the bearer token specified during service startup:

  Authorization: Bearer <your-token>

  (Omit if launched with --no-token).
• Direct Agent Execution: These endpoints are directly backed by the local JeikCode Agent runtime rather than acting as a simple proxy. Requests automatically load repository rules (AGENTS.md), streaming back thinking blocks and tool call progress in real time.

---

3. Example Request (cURL)

curl -X POST http://127.0.0.1:13457/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer your-token" \
  -d '{
    "model": "deepseek/deepseek-v4.1-flash",
    "messages": [
      {"role": "user", "content": "Analyze the architecture of this codebase"}
    ],
    "stream": true
  }'