Moltbot-Lite: Windows-first AI assistant with Web + Mail, workspace files, memory, and scheduled tasks.

• Web UI: Local HTTP server (default port 3000), chat and session list at /
• Mail: IMAP/SMTP (e.g. QQ mailbox), MBCTX protocol for context and agent selection
• LLM: SiliconFlow (OpenAI-compatible), configurable models
• Memory: Per-conversation summary + facts (file-based), optional LLM summarization
• File tools: Workspace-only (WorkspaceFS), read/list without approval; write/delete require approval via /api/approvals
• Scheduler: Cron tasks in data/tasks.json, actions: sendMessage (mail), runTool, runChat

1. Copy .env.example to .env and set at least:
   • SILICONFLOW_API_KEY for LLM
   • Optionally MAIL_USER / MAIL_PASS for Mail channel
2. Install and run:

pnpm install
pnpm dev

1. Open http://localhost:3000 for the Web UI.

• GET /api/sessions – list conversations
• POST /api/chat – send message (body: { sessionId?, message, agentId? })
• GET /api/thread?sessionId=... – get thread
• POST /api/tools/file/list – list workspace (body: { path? })
• POST /api/tools/file/read – read file (body: { path, maxBytes? })
• POST /api/tools/file/write – request write (returns pendingId; then POST /api/approvals/:id/approve)
• POST /api/tools/file/delete – request delete (same approval flow)
• GET /api/approvals – list pending approvals
• POST /api/approvals/:id/approve – approve and run
• POST /api/approvals/:id/reject – reject
• GET /api/tasks – list scheduled tasks
• POST /api/tasks – add task (body: { id?, cron, timezone?, action, enabled? })

• data/conv/<convId>/ – meta.json, thread.jsonl, summary.md, facts.json
• data/workspace/ – file tool root
• data/tasks.json – scheduled tasks
• data/runs.jsonl – scheduler run log
• data/audit.jsonl – tool call audit
• data/inbox_processed.jsonl – mail dedupe

MIT