Memory
Studio's window into repo Memory — Remembered lines as they happen in chat, and a dedicated Memory view over what's serving the agent, fading, or forgotten.
Studio surfaces the platform's Memory feature directly in the desktop app — the same repo-scoped knowledge base, curated without leaving the chat pane.
The flaky step was a race on the toast dismiss — fixed and re-ran.
✻ Remembered · decision — “Wait for the toast's exit animation before asserting checkout state”
⎿ distilling durable insights · /memory to browse
- Checkout toast dismiss is a known race — assert after animation end
- /cart API returns 502 under 3+ concurrent adds (flagged, not yet fixed)
This page is about the Studio surfaces
For what Memory is, its four categories, and how it's read back by Ask AI, the CLI, and MCP agents, see Memory (AI) → Memory. This page covers Memory 2.0's Studio-specific UI: the transcript lines, the Memory view, and local capture.
Remembered lines
While you chat, a captured memory shows up inline — ✻ Remembered · decision — "…" — instead of
disappearing silently into a database. A pending capture shows a dim
⎿ distilling durable insights · /memory to browse line until it resolves. At the end of a
session, a "What the agent learned this session" card summarizes everything captured, with a
Review memory link straight into the Memory view.
The Memory view
Memory is organized into three lanes — who can see it, and where it lives:
| Lane | Who sees it | Where it lives |
|---|---|---|
| Project memory | Your whole team | TestRelic Cloud |
| This device | Just you | Local disk, until you sync it |
| About you | Just you | The assistant's personal recall across your chats (read-only) |
Project memory and This device each offer the same four views:
- Lifecycle (the default landing) — what's actively serving the agent right now, what's fading, and what's been forgotten (soft-deleted or auto-archived entries, each restorable).
- List — a sortable table with pin, archive, restore, and delete curation.
- Graph — how a memory links to the tests and pages it's tied to.
- Coverage — which memories are mapped to a real test versus orphaned, with a Run mapping in chat action that seeds the composer.
Lifecycle and Coverage go deeper than a quick glance shows — Lifecycle tracks a memory's whole life (what's serving the agent right now against its context budget, what's fading, and a restorable Forgotten lane), while Coverage turns the same navigation data into a per-page pass-rate breakdown:
Graph plots the same data as a memory → test → page chain, so a memory with no linked test (or a test on no mapped page) is visibly dangling instead of silently missing:
Teal arrows = a full memory → test → page chain. Amber flags a memory with no linked test. Click a memory (in Studio) to open it in the reader.
Local capture
In Local mode, the agent can save a memory to a per-repo local store on disk without anything leaving the machine — nothing is uploaded unless you say so. The first local capture shows a one-time consent picker: Keep local only, Sync to cloud, or Ask me later. The choice is a toggle you can revisit any time from Command Center → "Back up memories to TestRelic Cloud."
Every capture after that first choice lands in a per-device table — each memory shows who wrote it, when it was saved, and whether it's synced yet — with sync (individual or bulk) always going through the same explicit confirmation, never silently:
Written automatically from your chats — Studio remembers durable facts at the end of each turn, and you can ask it directly ("remember that …"). Auto-sync is on: new memories upload to your project's shared cloud memory on a timer (Memory sync, top right).
| Memory | Category | Written by | Saved | Synced |
|---|---|---|---|---|
| Wait for toast's exit animation before asserting checkout state | Maintenance | ✻ Agent | 2h ago | |
| Feature-flag rollout uses a 10% canary | Decision | Auto | 1d ago | ✓ 1d ago |
| Prefer data-testid over text selectors here | Context | You | 3d ago | ✓ 3d ago |