Autolith docs

Autolith is a terminal programming agent that lives in a running Common Lisp image. It works in your repository, keeps project context, and can inspect and extend its own runtime. These pages take you from install to daily work in short, complete steps.

Read the three Start pages in order. Concepts explains one mechanism per page. Cookbook solves concrete tasks. Advanced is operator machinery. Every example was checked against Autolith 0.47.1.

Start

• Install — put Autolith on your machine and keep it current.
• Your first session — launch, authenticate, and hold a conversation.
• The workspace — what Autolith reads, where its state lives, what a workspace is.

Concepts

• Tools and permissions — the tool registry, approvals, and guard rails.
• Lisp workers — disposable REPLs beside the prompt-window REPL.
• The live image — journaled, exercisable self-modification.
• Commit and recover — generations, crash capsules, recovery.
• State — conversations, memories, agenda, papercuts.
• Recursive inference — corpora larger than the model window.
• Agents and jobs — child agents, roles, and headless jobs.
• Providers — sign-in methods and models.
• MCP — external tools through MCP servers.
• Skills — loadable instruction bundles.

Reference

• Slash commands — every interactive command in one table.

Cookbook

• Find and resume past work
• Keep a helper across restarts
• Search a corpus larger than the window
• Run Autolith from CI
• Add an MCP server
• Set up a project for Autolith

Advanced

• The management endpoint — an authenticated REPL for trusted programs.
• Context tuning — init.lisp and ephemeral notes.