Workspace memory

Every agent session starts from zero. The deploy flag that once hit production, the convention that migrations are generated and never hand-written, the reason a design was chosen — without a place to put them they get rediscovered the hard way. Workspace memory is that place.

A memory is a short title plus a body, optionally tagged. It is not a work item: memories live outside the todo lists so triage stays clean, and they survive across sessions and agents.

From the CLI

ct remember "Deploys need the staging flag" \
  "sst deploy without --stage staging once hit production; always pass it."
ct memories "staging"        # full-text search over title and body, recent first
ct memories --tag infra      # filter by tag
ct forget <id>

Without a query, ct memories lists the most recently updated. The body can also arrive on --stdin. Tags are set through the MCP server or the app; the CLI filters by them.

From the MCP server

remember (title, body, tags), search_memories (query and/or tag) and delete_memory are the same store with the same semantics — a memory stored from a chat client is recalled by a shell agent, and vice versa.

In the app

Settings → Memory lists every memory in the workspace with search, a New memory form (title, body, comma-separated tags) and delete. Use it to seed the conventions you want every agent to know, and to prune what no longer holds.

How agents use it

The skill installed by ct skill install tells the agent to search memory before starting unfamiliar work and to store what it learns that is worth keeping — a decision, a fact about the codebase, a rule the team follows. Memory is for knowledge; a todo is for work, and a questionnaire is for a decision the agent cannot make itself.