TUI Guide
The terminal UI is the fastest way to inspect what Memory Layer knows before you act on it. It combines cited query results, stored memory, project health, agent activity, watcher state, proposals, and diagnostics in one local surface.
Start the TUI
memory tui
memory tui --project <project-slug>The header identifies the active project and the footer reports TUI, service, watcher, and skill health. A source checkout uses an isolated dev profile; its header is marked [dev] so it is not confused with a packaged installation.
Shared navigation
| Key | Action |
|---|---|
Tab, Right, or l | Next tab |
Shift+Tab or Left | Previous tab |
h | Open or close help for the active tab |
? | Switch to Query and start a question |
/ | Edit the global text filter |
r | Refresh current project state |
Ctrl+C | Exit |
Choose a tab
| Need | Tab |
|---|---|
| Read or filter known memory | Memories |
| Ask a cited project question | Query |
| Review replacement proposals | Review |
| See active coding sessions | Agents |
| Check watcher liveness | Watchers |
| Inspect the managed skill bundle | Skills |
| See loop state | Automations |
| Generate a handoff briefing | Activity or Resume |
| Maintain semantic retrieval | Embeddings |
| Diagnose a failure | Errors |
| See the whole project at a glance | Project |
The TUI is for human inspection. Agents should prefer structured CLI output such as memory query --json when another tool will parse the result.
See also: Daily workflow and the Web UI.
Additional workspaces
The Offline tab shows read-only synchronized knowledge from explicitly
configured laptop subscriptions. r loads/refreshes inventory; j/k or arrows
select the exact subscription UUID; / enters a question and Enter performs
a bounded eight-hit local deterministic search. Questions are limited to 4096
UTF-8 bytes. Esc ends entry; d/u or PageDown/PageUp scroll provenance and
reviewed evidence. While editing, letters including q are question text, not
quit/navigation shortcuts; Ctrl+C still exits.
Area, owner instance, project, company, device and space UUIDs identify the copy. Last observed sync/cursor is not guaranteed freshness or current remote access. Shared text/locators are untrusted plain text, never executed or forwarded to a model. There are no write, approval or synchronization controls. Selection, refresh and query edits invalidate old excerpts; leaving the tab cancels reads. The configured local loopback owner must authenticate. A disabled/unsupported server or failed read is reported as unavailable, not an empty healthy cache.
The Jobs tab inspects durable background stages and provides confirmed retry with stable request IDs. See Background jobs for keys and policy.
The Supervisor tab provides instance-wide metadata-only inspection, separate from local coding sessions in Agents. See Supervisor inspection for permissions, cursor controls and explicit inactive-worker limitations.
For private federation, launch memory federation --project <slug> tui. This
focused workspace supports membership, exact publication review, shared citations,
proposals, approval/rejection and synchronization through the same authenticated
local API as Web and CLI. See Federation commands and keys.