@moldable-ai/ai-server

Internal AI server that powers Moldable's desktop chat and gateway integrations.

Gateway endpoints

The gateway API accepts plain text messages from Moldable Gateway (Telegram/WhatsApp/etc.) and stores sessions separately from desktop conversations.

POST /api/gateway/chat

{
  "sessionId": "agent:main:telegram:dm:123",
  "messages": [{ "role": "user", "text": "hello", "timestamp": 1737811000 }],
  "model": "anthropic/claude-3.7-sonnet",
  "reasoningEffort": "medium",
  "activeWorkspaceId": "personal",
  "gateway": {
    "channel": "telegram",
    "peerId": "123",
    "displayName": "alice",
    "isGroup": false,
    "agentId": "main",
    "sessionKey": "agent:main:telegram:dm:123"
  }
}

Response:

{ "text": "assistant response", "sessionId": "agent:main:telegram:dm:123" }

Session listing

• GET /api/gateway/sessions?workspaceId=personal
• GET /api/gateway/sessions/:id?workspaceId=personal
• DELETE /api/gateway/sessions/:id?workspaceId=personal

Storage

• Shared sessions: ~/.moldable/shared/gateway-sessions/
• Workspace sessions: ~/.moldable/workspaces/{workspaceId}/gateway-sessions/

Notes

• /api/chat remains the desktop app entry point for interactive chat sessions.
• Gateway sessions are kept separate to avoid mixing UI history.

App LLM endpoints

Installed apps call the local AI server with their Moldable app id and app token.

• POST /api/llm/completion returns { "text": "..." } for plain text output.
• POST /api/llm/stream streams plain text output.
• POST /api/llm/generate-json accepts a JSON Schema in schema and returns { "json": ... } from AI SDK structured generation.

Apps should define structured contracts with Zod when possible, convert them to JSON Schema at the HTTP boundary, and validate the returned object with the same Zod schema before acting on it.