Skip to content

Capture

What to Capture in AI Work Memory

Use a simple test for deciding what belongs in Origin and what should stay out of memory.

Qi-Xuan LuUpdated 5 min read

Article packet

01

Concepts

02

Origin users building better memory habits

03

5 min read

01

Capture one durable idea at a time.

02

Include why the fact matters for future sessions.

03

Skip raw logs, filler, and temporary status.

01

Quick answer

Capture something when a future AI session would waste time or make a worse decision without it. Good captures are durable, atomic, specific, and include the reason.

Origin's canonical memory types are identity, preference, decision, lesson, gotcha, and fact. Constraints and corrections are still valid capture content, but they usually become one of those types: a project constraint may be a fact or decision, and a correction should name what it supersedes.

02

When this problem appears

Most memory systems get noisy when users store everything. AI work memory should not become a junk drawer of transcripts, logs, and temporary todos.

03

Use the capture test

Before capturing, ask whether this fact will change a future AI session.

  • Capture decisions and why the chosen path won.
  • Capture gotchas that would cause repeated debugging.
  • Capture corrections when previous memory is stale or wrong.
  • Capture project constraints that are not obvious from source files.
  • Skip raw output; capture the durable conclusion, root cause, error signature, or command that proved the fix when future agents need it.

Capture examples

/capture We chose source-backed pages because summaries need provenance.
/capture Gotcha: origin-mcp must talk to the local daemon on 127.0.0.1:7878.
/capture Supersedes mem_abc123: Windows setup now uses a Task Scheduler ONLOGON task.

Bad: /capture worked on repo
Better: /capture Origin docs now separate Learn SEO articles from product setup docs because users need different navigation paths.

Bad: /capture tests failed
Better: /capture Gotcha: pnpm build fails if Learn relatedSlugs point to missing article slugs; run the slug audit before build.

04

What to check next

If a source file is already the authority, do not duplicate the file into memory. Capture the reasoning or consequence instead.

Try the local memory loop

Install Origin, connect your AI client, and verify that capture, recall, and handoff work on your machine.

FAQ

Should I capture every meeting note or chat summary?+
No. Capture the durable decisions, constraints, and follow-ups that should affect future AI work.
What if I captured the wrong thing?+
Recall or find the old memory ID, capture a corrected self-contained statement, and name what it supersedes. Use forget only when the old record should not remain.