-
Notifications
You must be signed in to change notification settings - Fork 1
Dashboard Tabs
Four views on one local URL, bound to 127.0.0.1, fully offline, no CDN. The React app is the single dashboard; an unbuilt checkout serves a small "run the build" stub instead of a second UI.
The home view: a live pixel scene of who is doing what. Every open Claude Code tab on the project is a character at a desk in a shared room — carpet, wood desks, glowing monitors, plants — animated by what it is doing right now:
- typing when editing, reading when reading, running when a command is going, thinking / idle when waiting
- a speech bubble shows the live file it is touching
- the monitor glows while the tab is active and the character dims when it goes quiet
- tokens saved is the hero figure at the top of the floor
Click any character to read its session story (the digest plus a tight timeline). The floor only shows tabs active in the last few minutes, fed by the cross-tab bus, so what you see is genuinely live. See Live-Agent-Dashboard for the data path and Subagents for how the bus coordinates multiple tabs.
Digest-first. Each session is synthesised server-side into one readable paragraph with headline counts (prompts, tools, files, exchanges), grouped by day. Open a session to expand a tight timeline — what was asked and what was done — rather than a dump of every prompt and tool call. A per-session Markdown report is one click away.
Powered by /api/session-digest and /api/story.
The knowledge slipstream has accumulated across sessions, made legible:
- Habits forming — instincts tagged recurring topic or hot file, each with how many times and across how many sessions it has recurred, and a confidence bar with a strength label (forming → strong → established).
- Recurring work — topic clusters with their dominant action, session count and the files involved.
- Durable facts — what has been explicitly remembered, surviving every future session.
The overview summary and its bullets (active areas, most-touched files, last activity) sit at the top.
An interactive dependency graph: files as nodes sized by how many files depend on them, imports as edges, the god nodes everything flows through ringed in white, colour grouped by area. Search to highlight, scroll to zoom, drag to pan, click a node for its imports and importers.
The views are wired through the /api/* surface, including:
| Endpoint | What it returns |
|---|---|
/api/overview |
Identity, map areas, recent work, counts, tokens and dollars saved |
/api/agents |
Every session on the bus as a character: thread, files, last tool, derived mood/verb, active flag |
/api/session-digest?session=S |
One-paragraph digest of a session plus headline counts |
/api/story?session=S |
The said-to-did timeline for a session |
/api/memory/overview |
Summary, durable facts, lessons / recurring work |
/api/instincts |
Recurring topics and hot files with confidence |
/api/codegraph |
The dependency graph nodes and edges |
/api/resume?session=S |
Open thread, suggested next, files in flight |
All bound to 127.0.0.1. No telemetry. No phone home.
- Live-Agent-Dashboard — the data path behind the office and the live strip
-
Health-Endpoint —
/api/healthand the version-aware restart - Observation-Memory — the store the views read from
- Cross-IDE-Support — how the same dashboard runs in Cursor, Windsurf, Antigravity, VS Code and JetBrains
Start here
Install paths
v0.6.0 features
- Map watcher
- Token forecast
- Replay export
- Configurable redaction
- Drift detection
- Per-skill opt %
- CI mode
- Lessons
Headline features
- MCP tools
- Observation memory & search
- Cross-IDE support
- Lossless compaction
- Memory recall
- Live agent dashboard
- Statusline
- Output style
- Subagents
Token efficiency
Skills
Internals
- Architecture
- Memory system
- Hooks
- Mind map and status
- Configuration and tuning
- Data formats
- Performance and benchmarks
- Design decisions
- Security model
- Testing strategy
Reference
SarmaLinux . sarmalinux.com