Agent Runtime

A portable, production-ready agent runtime built in Rust with support for multiple LLM providers, tool calling, session management, and inter-agent communication.

Features

• 🤖 Multi-Provider LLM Support: OpenAI, Anthropic, and more
• 🔧 Tool Calling: Extensible function calling system
• 💾 Flexible Session Storage: In-memory, cache, and database options
• 🛡️ Guardrails: Safety mechanisms and control systems
• 👤 Human-in-the-Loop: Approval workflows for sensitive operations
• 🧠 Dynamic Context Management: Intelligent context window handling
• 🔌 MCP Support: Model Context Protocol integration
• 🌐 Agent Communication: In-memory and network-based collaboration

Project Structure

This is a Cargo workspace with multiple crates:

Implemented (Phases 0-5):

• agent-core: Core foundation (error handling, logging, config)
• agent-llm: LLM providers (OpenAI, Anthropic) + conversations
• agent-session: Session management (memory, cache, layered storage)
• agent-tools: Tool system (registry, built-in tools)
• agent-runtime: Agent struct - ties everything together
• agent-cli: Command-line interface

Future (Phases 6+):

• agent-guardrails: Safety systems
• agent-hitl: Human-in-the-loop
• agent-context: Context management
• agent-mcp: MCP implementation
• agent-comms: Communication layer

Getting Started

Prerequisites

• Rust 1.75+ (install from rustup.rs)
• API keys for LLM providers (OpenAI and/or Anthropic)
• cargo-watch (optional): cargo install cargo-watch

Setup Environment Variables

# Copy the example environment file
cp .env.example .env

# Edit .env with your API keys
nano .env

Your .env file should contain:

OPENAI_API_KEY=sk-your-actual-key
ANTHROPIC_API_KEY=sk-ant-your-actual-key

Note: The .env file is git-ignored and safe for local development. See ENV_VARS.md for production deployment options.

Building

# Build all crates
cargo build

# Build in release mode
cargo build --release

Running Tests

# Run all tests
cargo test

# Run tests with output
cargo test -- --nocapture

# Run tests for a specific crate
cargo test -p agent-core

Development

# Watch for changes and run tests automatically
cargo watch -x test

# Check code without building
cargo check

# Run linter
cargo clippy

# Format code
cargo fmt

Development Phases

This project is being developed in phases:

• ✅ Phase 0: Project foundation
• ✅ Phase 1: LLM provider abstraction
• ✅ Phase 2: Message & conversation management
• ✅ Phase 3: Session management
• ✅ Phase 4: Tool calling system
• ✅ Phase 5: Agent runtime (current)
• 🔄 Phase 6: Database storage (SQLite)

See implementation_plan.md for full roadmap.

Milestone: Phases 0-5 complete! You have a working agent runtime!

License

Licensed under either of:

• Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
• MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)

at your option.