Capabilities

Memory is built around a small loop: build the product graph once, query it on demand, save product-meaningful changes, and reconcile at session end. Every command serves one of those jobs or helps a human inspect and repair storage.

Build the graph

memory init
memory init --brief

init creates .memory/ storage, installs the guidance block and generated product map into AGENTS.md/CLAUDE.md, starts the viewer, and prints the indexing brief. The brief is the agent’s instructions for the initial graph: explore the repo, interview the user for intent and stage, save once. --brief reprints it any time.

Query on demand

memory query "why do webhook retries run in the worker?"
memory query "state of batch exports" --budget 1200
memory inspect feature.batch-exports

• query returns a token-budgeted Markdown subgraph: full-text matches, their one-hop relations, and connected open questions.
• inspect opens one node and its direct relations.

There is no per-task loading step. The product map in the agent instruction files is the always-on overview; everything deeper is a query away.

Save product changes

memory save --stdin
memory save --stdin --dry-run

save takes intent JSON — {task, nodes, stale, supersede, delete} — and validates it into graph writes. Update existing nodes by id; retire wrong memory with stale, supersede, or delete instead of duplicating. Every save refreshes the product map.

Tip

When a correction reveals old memory was wrong, update, stale, supersede, or delete the existing node instead of creating a near-duplicate.

Keep it honest

memory sync
memory sync --dry-run
memory status
memory status --all
memory check

• sync is the session-end ritual: diff-driven anchor verification, a report of changed/orphaned/unanchored nodes and coverage gaps, and a pre-filled save skeleton when reconciliation is needed.
• status summarizes features by stage, open questions, stale anchors, and last activity. status --all is the cross-project dashboard.
• check validates storage and warns about dead anchors and out-of-date product map sections.

Human inspection

memory view --open
memory diff
memory projects list
memory docs

view starts the local browser viewer: projects dashboard, memory list, node detail, and the relation graph. diff shows tracked and untracked .memory/ changes. projects manages the user-level registry behind status --all and the viewer. docs prints bundled docs topics or opens the hosted site.

Repair and recovery

memory rebuild
memory reset
memory reset --all

• rebuild recreates the generated index from canonical files.
• reset backs up storage to .memory/.backup/ and clears it; reset --all does it for every registered project. After a breaking package upgrade, the recovery path is memory reset then memory init — there is no storage migration pre-1.0.

MCP

MCP covers query, save, status, and inspect when your client launches memory-mcp. Init, sync, viewer, registry, and maintenance stay in the CLI. See the MCP guide and Reference.