Skip to content

Capture work in memory

In this tutorial you'll record two entries to your project's memory tree — a decision and a checkpoint — and then watch them show up two ways: through the agent's recall and search tools, and in the dashboard. By the end you'll have durable, git-committed knowledge that any future session can pick up.

Everything is driven from a normal Cursor chat; the agent calls the memory tools for you.

Before you begin

Finish Getting started first. You need an onboarded project (a memory tree already exists in .memory/) with the angelo-memory MCP server green in Settings → MCP.

1. Record a decision

A decision captures a choice you made and why. In your chat, paste:

Record a decision to memory: "Use Markdown for all docs pages" — because the
docs site is built with MkDocs and Markdown keeps pages portable and reviewable.

The agent calls the memory record tool with type='decision' and places it in the tree (under the active phase, or the project root if you're just starting).

Expected result: the agent confirms it recorded the entry and a new markdown file appears under .memory/entries/ (a deci-…md file with YAML frontmatter).

2. Record a checkpoint

A checkpoint is a snapshot of where things stand — what's done and what's next. Paste:

Record a checkpoint to memory: "Docs tutorials started" — the first tutorial
pages are drafted; next up is wiring them into the nav.

Expected result: the agent confirms a second entry, this time a chec-…md (or similarly named) file under .memory/entries/.

Why record at all?

The memory tree is the system's long-term store — it persists across sessions and machines because it's committed to git. Recording now means a future you (or a future agent) can recall exactly what was decided.

3. See them with recall

Ask the agent to show you the tree:

Recall the memory tree and show me where my new decision and checkpoint landed.

The agent calls recall, which returns the nested tree plus a summary.

Expected result: both new entries appear in the tree under your project, with their titles and types — your decision and your checkpoint, right where the agent placed them.

recall shows structure; search finds entries by keyword. Paste:

Search memory for "docs".

Expected result: the search results include both entries you just recorded (they both mention docs), each with its id and title.

5. View them in the dashboard

Finally, see the tree visually. Ask:

Launch the memory dashboard.

The agent calls launch_dashboard and gives you a local URL. Open it.

  • The tree view renders your project, phases, and the two entries you added.
  • The Sessions view shows your live chat as a session card (it reads the live session sidecar under .memory/active-sessions/).

Expected result: you can spot your decision and checkpoint nodes in the tree, and your current session appears as a live card. If the card doesn't appear, reload Cursor (or restart the angelo-memory MCP server) so the latest sidecar is written.

What you learned

  • How to record the two most common entry types — decision and checkpoint — by just asking the agent.
  • Three ways to see your knowledge back: recall (structure), search (keyword), and the dashboard (visual tree + sessions).
  • Memory lives as plain markdown in .memory/, committed to git, so it's durable across sessions and machines.

Next steps