mchact

mchact is a fork of mchact — a Rust-based multi-platform agentic chat runtime. This fork focuses on building and integrating three companion projects:

• mchact — the core agentic chat engine (this codebase)
• hermes-agent — autonomous agent orchestration layer
• clawteam — multi-agent collaboration framework

What is mchact?

mchact is a channel-agnostic AI chat bot written in Rust. It connects to multiple chat platforms through a single agentic loop, executes tools, manages sessions, and provides persistent memory — all from one binary.

Supported Channels

Channel	Status
Web UI	Built-in (React + Vite)
Telegram	Stable (teloxide)
Discord	Stable (serenity)
Slack	Supported
Feishu / Lark	Supported
Matrix	Supported
WhatsApp	Supported
WeChat	Supported
Signal	Supported
IRC	Supported
Nostr	Supported
DingTalk	Supported
QQ	Supported
iMessage	Supported
Email	Supported

LLM Providers

Provider-agnostic runtime supporting Anthropic, OpenAI, OpenRouter, Ollama, DeepSeek, Google, and any OpenAI-compatible API.

Key Features

• Agentic tool loop — multi-step tool execution with session resume and context compaction
• 22+ built-in tools — bash, file ops, web search/fetch, browser, memory, scheduling, MCP, A2A, and more
• Persistent memory — file-based (AGENTS.md) + structured SQLite/PostgreSQL memory with reflector and deduplication
• Scheduled tasks — cron-based background scheduler with override prompts
• Sub-agents — spawn isolated agent loops with restricted tool access
• Skills & ClawHub — skill discovery, activation, and registry integration
• MCP support — Model Context Protocol server/tool federation
• A2A protocol — agent-to-agent HTTP integration for multi-agent routing
• Hooks — extensible event system (BeforeLLMCall, BeforeToolCall, AfterToolCall)
• Plugins — plugin runtime with manifest-based discovery
• Observability — OpenTelemetry metrics/traces/logs, Langfuse, AgentOps adapters
• Web UI — built-in React interface with auth, sessions, streaming, config panel
• Soul system — customizable personality via SOUL.md with per-chat overrides
• Sandbox — configurable security sandbox for tool execution
• Voice — speech-to-text via OpenAI Whisper or local transcription

Project Focus

This repository (mchact) serves as the integration workspace where mchact, hermes-agent, and clawteam converge:

mchact (core engine)

The upstream agentic chat runtime. Handles LLM communication, tool execution, session management, and channel adapters. This is the foundation everything else builds on.

hermes-agent (orchestration)

Autonomous agent orchestration layer. Hermes manages agent lifecycles, task delegation, and inter-agent communication. It builds on mchact's sub-agent and A2A capabilities to coordinate complex multi-step workflows.

clawteam (collaboration)

Multi-agent collaboration framework. ClawTeam enables teams of specialized agents to work together — planning, executing, reviewing, and iterating — as a coordinated unit. It leverages hermes-agent for orchestration and mchact for execution.

Architecture

crates/
  mchact-core/        Shared types, errors, text utilities
  mchact-storage/     SQLite/PostgreSQL persistence (DataStore trait), memory domain, usage reports
  mchact-tools/       Tool runtime primitives, sandbox, path guards
  mchact-channels/    Channel abstraction and delivery boundary
  mchact-clawhub/     ClawHub registry client, install, lockfile
  mchact-app/         App-level support (logging, skills, transcribe)
  mchact-observability/ OTLP metrics/traces/logs export

src/
  main.rs                CLI entry point (start, setup, doctor, help)
  runtime.rs             App wiring, provider/tool init, channel boot
  agent_engine.rs        Core agentic loop (process_with_agent)
  llm.rs                 Provider implementations + format translation
  web.rs                 Web API routes and streaming
  scheduler.rs           Background task scheduler + memory reflector
  channels/*.rs          Platform adapters (16 channels)
  tools/*.rs             Built-in tool implementations (22+ tools)

web/                     React + Vite web UI (embedded into binary)

Quick Start

# Build
cargo build

# Interactive setup wizard
cargo run -- setup

# Start the bot
cargo run -- start

# Diagnostics
cargo run -- doctor

Configuration

Copy mchact.config.example.yaml to mchact.config.yaml and configure:

• LLM provider and API key
• LLM provider and API key
• Channels to enable (web, telegram, discord, etc.)
• Data directory and working directory
• db_backend: "sqlite" (default) or "postgres"; set db_database_url for Postgres
• Optional: observability, sandbox, voice, plugins

See README-origin.md for the full original documentation.

License

MIT