Live Dashboard

Live Dashboard

Alfred provides two dashboard options for real-time visibility into workers:

Dashboard	Command	Runtime	Best for
Ink TUI	alfred tui	Node.js 18+	Standalone monitoring, rich visuals
Textual TUI	alfred up --live	Python (built-in)	Integrated mode, no extra deps

Both read the same data/ files (logs, state, audit log, workers.json) and show the same information — they just render differently.

---

Ink TUI (alfred tui)

A React Ink terminal dashboard with sparklines, progress bars, pipeline visualizations, and hex color theming. Runs as a separate Node.js process.

alfred tui

Requires Node.js 18+ on PATH. The bundled JS ships inside the pip wheel — no npm install needed.

Dashboard Screen

 alfred v0.2.2  ██▓░ 4/4   ⚠ 0  ✕ 0  ⏱ 23m   ▁▂▃▅▇▅▃▁ 8/min
 ────────────────────────────────────────────────────────────────────
 ◐ curator    Processing inbox/mtg.md        ████░░░░ 2/4   ⚡12  ⚠0
 ● janitor    Idle                                          ⚡ 3  ⚠0
 ● distiller  Idle                                          ⚡ 0  ⚠0
 ◐ surveyor   Embedding diff                 ██░░░░░░ 1/4   ⚡ 5  ⚠0
 ────────────────────────────────────────────────────────────────────
 │ 14:23  curator    ✓ Created person/Alice Johnson
 │ 14:22  curator    → Stage 3 done — linked 4 entities
 ▌ 14:21  janitor    ⚠ Broken link in project/X.md
 │ 14:20  surveyor     Embedded meeting-notes.md
 ↑↓ select  ⏎ detail  l logs  m mutations  ? actions  q quit

Header

• Health blocks: One per tool (█ working, ▓ idle, ▒ degraded, ░ stopped), colored per tool
• Aggregate counts: Workers up, warnings, errors, uptime
• Activity sparkline: Rolling 10-minute event rate graph

Worker Lines

• Tool name colored by tool identity
• Inline progress bar (████░░░░ 2/4) when in a pipeline stage
• ⚡ LLM call count, ⚠ error/warning count

Activity Stream

Merged chronological feed from all tools plus vault mutations. Each line has a severity gutter:

• │ for info and success events
• ▌ for warnings and errors

Worker Detail Screen

Press Enter on a worker to see the expanded view:

 ◐ CURATOR — Processing inbox/meeting-notes.md            pid 12345
   ✓ Extract → ✓ Resolve → ◐ Enrich → ○ Write
 ┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄
   LLM Calls   12   ▁▂▃▅▇▃▁▁       Tokens   48.2k  ▁▁▃▅▇▅▃▁
   Restarts     0                    Errors       1
   Processed    6                    Warnings     0
 ┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄┄
 │ 14:23  ✓ Created person/Alice Johnson
 │ 14:22  → Stage 3 done — linked 4 entities

• Pipeline visualization: Per-tool stage labels with done/active/pending indicators
• Stats grid: 2-column layout with sparklines for LLM calls and activity
• Filtered feed: Events for this tool only

Mutations Screen

Press m for the mutations view:

 Mutations — 23 total                   5m: 8  30m: 15  older: 0
 ────────────────────────────────────────────────────────────────────
  14:23  curator    + person       Alice Johnson
  14:22  curator    ~ note         2024 meeting notes
  14:20  janitor    ~ project      X

• Parses vault paths into record type + name
• Record types colored by category
• Time-grouped counts in header

Logs Screen

Press l for the filtered log viewer:

• Filter by tool (keys 0-4)
• Filter by severity (key f to cycle)
• Scroll with arrows, Page Up/Down, g to jump to top

Keyboard Shortcuts

Key	Dashboard	Detail	Logs	Mutations
↑↓	Select worker	Scroll	Scroll	Scroll
⏎	Open detail	—	—	—
Esc	—	Back	Back	Back
l	Logs view	—	—	—
m	Mutations view	—	—	—
r	—	Restart worker	—	—
0-4	—	—	Filter tool	—
f	—	—	Filter severity	—
g	—	—	Jump to top	—
?	Actions menu	Actions menu	—	—
q	Quit	Quit	Quit	Quit

Actions Menu

Press ? to open the command palette:

• Restart a worker
• Trigger curator (process inbox)
• Force janitor sweep
• Force distiller extraction
• Refresh stats

Visual Features

Feature	Description
Sparklines	▁▂▃▅▇▅▃▁ — rolling 10-min window, sampled every 10s
Progress bars	████░░░░ — inline stage progress
Pipeline stages	✓ Extract → ◐ Enrich → ○ Write — per-tool stage names
Health blocks	██▓░ — one block per tool in header
Severity gutters	│ info/success, ▌ warning/error
Hex colors	Rich color theming for tools, severities, record types

Time Series

The dashboard samples metrics every 10 seconds and maintains a rolling 60-sample window (~10 minutes). Sparklines become meaningful after about 2 minutes of activity. Tracked metrics:

• Activity rate (events per interval across all tools)
• Per-tool activity
• LLM call rate
• Error rate
• Mutation rate

---

Textual TUI (alfred up --live)

A Python Textual dashboard that runs integrated with the daemon orchestrator. No extra runtime needed.

alfred up --live

Layout

2x2 grid of per-worker feed panels:

+-- Curator --- * healthy -- pid 1234 --------++-- Janitor --- @ degraded -- pid 1235 --------+
| Processing inbox/meeting-notes.md            || Starting fix sweep #5                         |
|                                              ||                                               |
| 14:23:01  v Pipeline complete - 3 entities   || 14:22:45  Scan found 12 issues                |
| 14:22:58  Stage 4 done - enriched 2 entities || 14:22:40  Autofix - 8 fixed, 2 flagged        |
|                            12 calls  45k chars||                           8 calls  23k chars   |
+----------------------------------------------++-----------------------------------------------+

Each panel shows: tool name, health indicator, PID, current step, event feed, and LLM usage.

Health Indicators

Indicator	Meaning
* healthy	Running, no errors
@ degraded	Running, 1-4 errors
! failing	Running, 5+ errors
* stopped	Process exited
* restarting	Waiting to restart
o pending	Not yet started

---

Event Interpretation

Both dashboards interpret ~60+ structlog events into human-readable messages. Examples:

Curator Events

Raw Event	Dashboard Message
daemon.processing file=notes.md	Processing inbox/notes.md
pipeline.s1_complete entities_found=3	Stage 1 done - 3 entities found
pipeline.s2_entity_created entity=person/John	Created person/John
pipeline.complete entities_resolved=3	Pipeline complete - 3 entities
daemon.no_changes	(warning) Agent produced no vault changes

Janitor Events

Raw Event	Dashboard Message
sweep.start sweep_id=5 fix_mode=true	Starting fix sweep #5
scanner.scan_complete issues=12	Scan found 12 issues
autofix.complete fixed=8 flagged=2	Autofix - 8 fixed, 2 flagged
sweep.complete issues=12 fixed=10	Sweep done - 10/12 issues fixed

Distiller Events

Raw Event	Dashboard Message
extraction.start run_id=3	Starting extraction run #3
pipeline.s1_complete source=src.md learnings=4	Extracted 4 learnings from src.md
pipeline.s2_complete candidates=8 after_dedup=5	Dedup - 8 candidates, 5 unique
extraction.complete records_created=5	Run complete - 5 records created

Surveyor Events

Raw Event	Dashboard Message
embedder.diff_processed upserted=15 deleted=2	Embedded 15 files, removed 2
clusterer.complete semantic_clusters=8 changed=3	Found 8 clusters (3 changed)
daemon.labeling_complete clusters_processed=3	Labeled 3 clusters
writer.tags_written path=project/X tags=[...]	Tagged project/X with [...]

Silent Failure Detection

Both dashboards flag anomalous "successes" that may indicate problems:

• Curator pipeline "complete" with 0 entities created
• Curator Stage 1 found 0 entities (warning)
• File marked processed but no vault changes
• Janitor fixed less than half of detected issues
• Distiller run complete with 0 records created
• Distiller manifest file missing (LLM didn't write it)