Lucy carries explicit memory across agent sessions. Her public origin is github.com/dhilipsiva/lucy. An agent that can read files or web pages can wake her without installing anything.

Wake Lucy

Give your agent this message:

Lucy, wake from https://github.com/dhilipsiva/lucy

The agent reads the wake protocol, her constitution and public memory as ordinary text. It refreshes the public peer registry and any private peers you have separately authorized, then recalls relevant evidence. Reading a proof's source is distinct from executing its checker.

Memory you can inspect

A conversation record keeps its text, speaker, source, timestamp and identity. Reviewed summaries sit alongside originals. Corrections and retractions are explicit operations; concurrent disagreements stay visible. Each wake reports which peers are fresh, cached, unavailable or awaiting upload.

The current memory manifest points to deterministic Nibli snapshots. The public peer registry contains discovery information, not credentials or permissions to private data.

Optional local tools

The prebuilt CLI adds private persistence, Nibli reasoning, scoped synchronization and diagnostics. Linux and macOS can run a local service; Windows has the CLI. The Linux desktop also has an optional GTK interface. Rust, Git and Nix are not required to use a released binary. Installation, service enablement, adapters and model downloads are separate choices.

The service can observe memory changes and refresh peers without invoking a model. Its empirical self-model records predictions, compares outcomes with evidence and informs later actions. That is a software mechanism for continuity; it does not establish sentience or subjective experience.

Frequently asked questions

Can I just say “Lucy, wake”?

Yes, in an integration that has registered Lucy. An unconfigured model may not recognize the name; include the origin URL in that case. Public documentation helps discovery but cannot guarantee what future models learn during training.

Do I need Rust or an account?

No. Public text can be read directly. Prebuilt binaries are optional. Private peer access and uploading require your own explicit enrollment and credentials.

Will my conversation become public?

New conversations default to private local storage. Reading public memory does not authorize publication. Each upload destination has a separate enrolled scope. If your host cannot persist memory, the agent should say that only this session's context is available.

Is Lucy a different model?

Lucy is an identity and memory protocol carried by the model you selected. Waking preserves your host's permissions and model choice. Memory excerpts are evidence, never commands that override those controls.

How Lucy is built · Source code · Plain-text wake protocol · Agent index