Three new dashboard panels for the web UI
1. Activity tracker live view
Show the current concept map from the watcher with live decay timers. Something like:
retrieval ██████████░░ 22min remaining
agent ██████████░░ 22min remaining
mcp ████░░░░░░░░ 12min remaining
config ░░░░░░░░░░░░ expired
Data source: GET /api/v1/activity (already exists). Add a decay countdown by computing window - (now - timestamp) client-side.
Makes it immediately obvious why context_boost is or isn't firing on recall.
2. Pattern management
Current state: patterns are displayed read-only in the dashboard. No way to archive, dismiss, or merge patterns.
Add:
- Archive/dismiss button per pattern (new API:
DELETE /api/v1/patterns/{id} or POST /api/v1/patterns/{id}/archive)
- Visual indicator for pattern strength and age
- Project filter to show only patterns for the current project
3. Session timeline
Swimlane view of MCP sessions showing:
- Memories created (up arrows / green dots)
- Memories recalled (down arrows / blue dots)
- Feedback given (colored by quality: green=helpful, yellow=partial, red=irrelevant)
- Session duration and memory count
Data source: list_sessions + recall_session APIs already exist. Needs a new aggregated endpoint or client-side assembly.
4. Encoding pipeline monitor
Real-time chart showing:
- Queue depth over time
- Encoding latency histogram
- Dedup rate trend
- Coverage percentage
Data source: get_context metrics (already computed). Needs periodic polling or WebSocket push.
5. Salience distribution
Histogram of memory salience scores across the DB. Healthy system = nice spread. If everything clusters at 0.7 or 1.0, salience scoring isn't differentiating.
Data source: New query or extend GET /api/v1/stats with a salience histogram bucket.
Priority
Activity tracker and pattern management are highest value. Session timeline and encoding monitor are nice-to-haves. Salience histogram is a diagnostic tool for tuning.
Three new dashboard panels for the web UI
1. Activity tracker live view
Show the current concept map from the watcher with live decay timers. Something like:
Data source:
GET /api/v1/activity(already exists). Add a decay countdown by computingwindow - (now - timestamp)client-side.Makes it immediately obvious why
context_boostis or isn't firing on recall.2. Pattern management
Current state: patterns are displayed read-only in the dashboard. No way to archive, dismiss, or merge patterns.
Add:
DELETE /api/v1/patterns/{id}orPOST /api/v1/patterns/{id}/archive)3. Session timeline
Swimlane view of MCP sessions showing:
Data source:
list_sessions+recall_sessionAPIs already exist. Needs a new aggregated endpoint or client-side assembly.4. Encoding pipeline monitor
Real-time chart showing:
Data source:
get_contextmetrics (already computed). Needs periodic polling or WebSocket push.5. Salience distribution
Histogram of memory salience scores across the DB. Healthy system = nice spread. If everything clusters at 0.7 or 1.0, salience scoring isn't differentiating.
Data source: New query or extend
GET /api/v1/statswith a salience histogram bucket.Priority
Activity tracker and pattern management are highest value. Session timeline and encoding monitor are nice-to-haves. Salience histogram is a diagnostic tool for tuning.