code_memory

code_memory gives your AI agent a persistent brain. Decisions, gotchas, architectural notes, and fix patterns survive across sessions — the agent picks up exactly where it left off.

Actions

Action	Description
save	Store a memory
recall	Retrieve relevant memories
chat	Conversational query with auto-memory
browse	Paginated listing of all memories
update	Correct an existing memory
delete	Remove a memory by ID
status	Check connection and project KB

Saving a memory

code_memory(action="save", text="processPayment must never be called without a valid idempotency key — caused a double-charge incident in prod")

Recalling memories

code_memory(action="recall", text="payment gotchas")
→ processPayment must never be called without a valid idempotency key...
→ Stripe webhook retries can fire up to 3 times — always check event.id

Project isolation

Each project gets its own Knowledge Base, automatically created on first save. Memories from one project never appear in another project’s recall.

The KB ID is cached in .gl/kb_id. No manual management needed.

What to save

• User preferences: “always use X over Y”
• Gotchas discovered: “this function has a race condition under concurrent writes”
• Decisions made: “we chose approach A because of constraint B”
• Fix patterns: “when error X happens, the fix is Y”
• Left-off notes: “TODO: finish token refresh logic in auth middleware”