7/24 Office -- Self-Evolving AI Agent System

A production-running AI agent built in ~3,500 lines of pure Python with zero framework dependency. No LangChain, no LlamaIndex, no CrewAI -- just the standard library + 3 small packages (croniter, lancedb, websocket-client).

26 tools. 8 files. Runs 24/7.

Built solo with AI co-development tools in under 3 months. Production 24/7.

Features

Tool Use Loop -- OpenAI-compatible function calling with automatic retry, up to 20 iterations per conversation
Three-Layer Memory -- Session history + LLM-compressed long-term memory + LanceDB vector retrieval
MCP/Plugin System -- Connect external MCP servers via JSON-RPC (stdio or HTTP), hot-reload without restart
Runtime Tool Creation -- The agent can write, save, and load new Python tools at runtime (create_tool)
Self-Repair -- Daily self-check, session health diagnostics, error log analysis, auto-notification on failure
Cron Scheduling -- One-shot and recurring tasks, persistent across restarts, timezone-aware
Multi-Tenant Router -- Docker-based auto-provisioning, one container per user, health-checked
Multimodal -- Image/video/file/voice/link handling, ASR (speech-to-text), vision via base64
Web Search -- Multi-engine (Tavily, web search, GitHub, HuggingFace) with auto-routing
Video Processing -- Trim, add BGM, AI video generation -- all via ffmpeg + API, exposed as tools
Messaging Integration -- WeChat Work (Enterprise WeChat) with debounce, message splitting, media upload/download

Architecture

                    +-----------------+
                    |  Messaging      |
                    |  Platform       |
                    +--------+--------+
                             |
                    +--------v--------+
                    |   router.py     |  Multi-tenant routing
                    |  (per-user      |  Auto-provision containers
                    |   containers)   |
                    +--------+--------+
                             |
                    +--------v--------+
                    | xiaowang.py     |  Entry point
                    |  HTTP server    |  Callback handling
                    |  Debounce       |  Media download
                    |  ASR pipeline   |  File persistence
                    +--------+--------+
                             |
                    +--------v--------+
                    |    llm.py       |  Tool Use Loop (core)
                    |  LLM API call   |  Session management
                    |  System prompt  |  Cross-session context
                    |  Multimodal     |  Memory injection
                    +--------+--------+
                             |
              +--------------+--------------+
              |              |              |
     +--------v---+  +------v------+  +----v--------+
     |  tools.py   |  | memory.py  |  |scheduler.py |
     | 26 built-in |  | 3-stage    |  | cron + once |
     | tools +     |  | pipeline:  |  | jobs.json   |
     | plugin sys  |  | compress   |  | persistent  |
     | + MCP bridge|  | deduplicate|  | tz-aware    |
     +------+------+  | retrieve   |  +-------------+
            |          +------------+
     +------v------+
     |mcp_client.py|  JSON-RPC over stdio/HTTP
     | MCP protocol|  Namespace: server__tool
     | Auto-reconnect  Hot-reload support
     +-------------+

Memory System

Layer 1: Session (short-term)
  - Last 40 messages per session, JSON files
  - Overflow triggers compression

Layer 2: Compressed (long-term)
  - LLM extracts structured facts from evicted messages
  - Deduplication via cosine similarity (threshold: 0.92)
  - Stored as vectors in LanceDB

Layer 3: Retrieval (active recall)
  - User message -> embedding -> vector search
  - Top-K relevant memories injected into system prompt
  - Zero-latency cache for hardware/voice channels

Tool List (26 built-in)

Category	Tools
Core	`exec`, `message`
Files	`read_file`, `write_file`, `edit_file`, `list_files`
Scheduling	`schedule`, `list_schedules`, `remove_schedule`
Media Send	`send_image`, `send_file`, `send_video`, `send_link`
Video	`trim_video`, `add_bgm`, `generate_video`
Search	`web_search` (multi-engine: Tavily, web, GitHub, HuggingFace)
Memory	`search_memory`, `recall` (vector semantic search)
Diagnostics	`self_check`, `diagnose`
Plugins	`create_tool`, `list_custom_tools`, `remove_tool`
MCP	`reload_mcp`

Quick Start

Clone and configure:

git clone https://github.com/your-username/724-office.git
cd 724-office
cp config.example.json config.json
# Edit config.json with your API keys

Install dependencies:

pip install croniter lancedb websocket-client
# Optional: pilk (for WeChat silk audio decoding)

Set up workspace:

mkdir -p workspace/memory workspace/files

Create personality files (optional but recommended):

# workspace/SOUL.md  -- Agent personality and behavior rules
# workspace/AGENT.md -- Operational procedures and troubleshooting guide
# workspace/USER.md  -- User preferences and context

Run:

python3 xiaowang.py

The agent starts an HTTP server on port 8080 (configurable). Point your messaging platform webhook to http://YOUR_SERVER_IP:8080/.

Configuration

See config.example.json for the full configuration structure. Key sections:

models -- LLM providers (any OpenAI-compatible API)
messaging -- Messaging platform credentials
memory -- Three-layer memory system settings
asr -- Speech-to-text API credentials
mcp_servers -- MCP server connections

Design Principles

Zero framework dependency -- Every line is visible and debuggable. No magic. No hidden abstractions.
Single-file tools -- Adding a capability = adding one function with @tool decorator in tools.py.
Edge-deployable -- Designed to run on Jetson Orin Nano (8GB RAM, ARM64 + GPU). RAM budget under 2GB.
Self-evolving -- The agent can create new tools at runtime, diagnose its own issues, and notify the owner.
Offline-capable -- Core functionality works without cloud APIs (except the LLM itself). Local embeddings supported.

License

MIT

724-office

Specs

README