The Lambda Symbolics team released Autolith, a self‑modifying general‑purpose Lisp AI agent that runs as a terminal agent inside a live Common Lisp image. The installer can be invoked with a single curl‑into‑sh command:
curl -fsSL https://sh.lambda-symbolics.com/autolith | sh
Architecture and platform reach
Autolith embeds an agent inside a running Common Lisp image, allowing it to observe, introspect, and rewrite its own code while you work. It supports Linux (x86_64, aarch64), macOS (x86_64, aarch64), and the BSDs via pre‑built binaries, while Windows runs from a source checkout using the provided PowerShell bootstrap script. Platform support can be extended by contributors.
Installing and authenticating providers
Besides the curl installer, the runtime is also available through Nix (nix run github:lambda-symbolics/autolith). After installation you can authenticate LLM providers with commands like autolith auth chatgpt or autolith auth anthropic. Once authenticated, launch the agent with autolith or start a fullscreen session via autolith --fullscreen.
Core capabilities
- Inline REPL – The message window is a Lisp REPL; you can mix prose prompts with Lisp forms that execute directly in the active image.
- Self‑modification tools – Commands such as
lisp.evalandself.statuslet the agent edit its own code, create isolated SBCL workers, and roll back changes via checkpoints and generations. - Recursive inference – The
infer,rlm-map, andrlm-completeprimitives let Autolith process large contexts (e.g., log files) without exceeding model token limits, persisting traces underdata/inferences/. - Vault for crash recovery – Conversations are stored in append‑only files; commands like
(vault),(vault-restore), and(vault-discard)let you inspect or replay prior messages after a crash.
Cautionary notes
The announcement does not include pricing, usage limits, or a Windows binary; Windows users must build from source, which adds friction for teams that rely on simple installers. Also, the curl‑into‑sh installer requires trusting the script before execution.
When to try it – If your startup already uses Common Lisp or needs a REPL‑centric AI assistant that can modify its own runtime, give Autolith a spin on Linux or macOS. Keep an eye on upcoming Windows binaries and community‑contributed packages before adopting it in a production environment.
