Axe is a lightweight CLI tool for orchestrating single-purpose AI agents defined in TOML. It adopts a Unix-like philosophy, treating LLM agents as composable, focused programs that integrate seamlessly with existing workflows like pipes, git hooks, and cron jobs. This approach benefits developers and power users by enabling scriptable, efficient AI task automation without requiring a heavy framework or GUI.

How It Works

Axe manages LLM agents via declarative TOML configurations, specifying system prompts, models, and skills. It emphasizes composability, allowing agents to delegate tasks to sub-agents with configurable depth limits and parallel execution. The tool acts as an executor, designed to be composed with standard Unix utilities rather than providing its own scheduling or workflow orchestration, prioritizing modularity and integration.

Quick Start & Requirements

• Installation: Requires Go 1.24+. Install via go install github.com/jrswab/axe@latest or build from source. Docker images are available.
• Prerequisites: Go 1.24+, API keys for providers (Anthropic, OpenAI), Ollama for local models. Docker recommended for isolation.
• Setup: Initialize config with axe config init. Scaffold agents with axe agents init <agent-name>.
• Links: Examples are in the examples/ directory; see examples/README.md for setup.

Highlighted Details

• Multi-provider support: Anthropic, OpenAI, Ollama (local).
• Declarative agent definitions via TOML.
• Sub-agent delegation with depth limiting and parallel execution.
• Persistent memory via timestamped markdown logs.
• Built-in, sandboxed tools (filesystem, shell commands).
• MCP tool support for external integrations.
• Docker image includes security hardening.

Maintenance & Community

The provided README does not detail notable contributors, sponsorships, or community channels.

Licensing & Compatibility

Licensed under the Apache-2.0 license, permissive for commercial use and closed-source projects.

Limitations & Caveats

Axe is CLI-focused, lacking a GUI. Complex orchestration requires external Unix tools. Script paths in skills must be absolute for run_command due to working directory context. Sub-agent nesting has a hard limit of 5 levels.