Memory

The Pathvela agent can remember useful facts across sessions. When it learns something worth keeping — a convention you prefer, where a piece of code lives, a decision you made — it records it, and recalls it later when it's relevant. You don't manage a database; the agent handles it as part of doing the work.

How it works

Memory is a tool the agent uses on its own. During a task it can:

  • Remember a fact worth carrying forward.
  • Recall and search what it's stored before.
  • List, tag, link, and forget entries as things change.

So a fact you established last week — "we deploy from the release branch", "the auth code is in src/auth/" — is available in a fresh session without you repeating it.

Memory is scoped two ways:

  • Project memory — facts tied to the project you're working in.
  • Global memory — facts that apply everywhere.

Both persist on your machine under your Lily data directory (~/.lily/memory/), so they survive restarts and new sessions.

Turning it on and off

In the interactive terminal UI, /memory controls the feature:

/memory            # toggle memory on or off
/memory status     # show whether it's on
/memory on
/memory off

Turn it off for a session where you'd rather the agent not persist anything; turn it back on when you want continuity again.

There is no lily memory command and no manual "add this to memory" step — memory is written by the agent as it works, and the /memory toggle is the control you have over it.

Memory vs rules

Memory and rules both give the agent context, from opposite directions:

  • Rules are what you write down for the agent to follow — every turn, deliberately.
  • Memory is what the agent notices and records for itself — recalled when relevant.

If something should always hold (a coding standard, a command), put it in AGENTS.md rather than relying on memory. Memory is for the incidental things learned along the way.

See also