Tomis

An AI agent orchestrator that coordinates tasks by delegating to specialized models, with support for skills and MCP (Model Context Protocol) integration.

Features

• Orchestrator Agent — Coordinates complex tasks by delegating to specialized models (coding, fast search, vision, review)
• Skills System — Load expert workflows from ~/.agents/skills/ for specialized tasks
• MCP Client — Connect to external MCP servers for additional tools and resources
• Interactive TUI — Real-time terminal interface with streaming output
• Memory System — Persistent knowledge base across sessions
• Approval Workflow — Guard-based approval for sensitive operations

Quick Start

# Build
cargo build --bin tomis

# Run with default config
cargo run --bin tomis -- --config config.toml

# Or with inline request
cargo run --bin tomis -- --config config.toml -r "What files are in this directory?"

Configuration

Create a config.toml:

endpoint = "http://127.0.0.1:11434"
model = "llama3.1:8b"
log_level = "info"
think_visible = true

# Model registry for delegation
[[models]]
capability = "coding"
endpoint = "http://127.0.0.1:11434"
model = "qwen3.5:cloud"
local = false

[[models]]
capability = "fast"
endpoint = "http://127.0.0.1:11434"
model = "qwen3.5:cloud"
local = false

# MCP servers (optional)
[[mcp]]
name = "filesystem"
command = "npx"
args = ["-y", "@modelcontextprotocol/server-filesystem", "."]
auto_connect = true

See docs/configuration.md for full reference.

Project Structure

tomis/
├── src/
│   ├── bin/tomis.rs      # CLI entry point
│   ├── dev_session.rs    # Interactive development session
│   ├── orchestrator.rs   # Model client wrapper
│   ├── sandbox.rs        # Sandboxed tool execution
│   ├── skills.rs         # Skills loading and management
│   ├── mcp_client.rs     # MCP client implementation
│   ├── tools.rs          # Built-in tool definitions
│   └── tui.rs            # Terminal UI
├── docs/                 # Documentation
└── config.toml     # Configuration file

Documentation

• Getting Started — Installation and first steps
• Configuration — Config file reference
• Skills — Skills system and how to create them
• MCP Integration — Connecting to MCP servers
• Architecture — System design overview

Development

# Format and lint
cargo fmt && cargo clippy

# Run with logging
RUST_LOG=info cargo run --bin tomis -- --config config.toml

# Run tests
cargo test

License

MIT