Axe

Axe is an open source command line tool that runs LLM powered agents the way Unix runs programs. Each agent does one focused job, is defined in a TOML file with its own system prompt, model, skill files, context files, working directory and persistent memory, and is run from the terminal. Data is piped in and results come out, so a developer can run something like a git diff straight into a review agent, chain agents together, or trigger them from cron, git hooks or CI. There is no daemon, no GUI and no framework to adopt; it ships as a single binary plus configuration. It suits developers who want small, composable automation rather than a long running chat session. Axe supports Anthropic, OpenAI, Ollama for local models, OpenCode and AWS Bedrock. Agents can delegate to sub agents through tool use, with depth limits and parallel execution. Memory is kept as timestamped markdown logs, with LLM assisted garbage collection to trim it. Built in tools cover file reading, writing and editing sandboxed to the working directory, shell commands, URL fetching and web search, with a hostname allowlist and blocking of private IP ranges. Token budgets cap usage per run, MCP servers can supply extra tools, retries use configurable backoff, a dry run mode shows the resolved context without calling a model, and JSON output supports scripting. Axe is free and open source under the Apache 2.0 licence. There is no paid plan; the user pays only for the model provider they connect. Axe describes itself as the executor, not the scheduler, and deliberately leaves scheduling and workflow orchestration to existing Unix tools.