Skip to main content
The dev dashboard is a self-contained tool for visually inspecting kernle memory stacks. It serves an interactive dark-themed web UI on localhost, giving you a complete view of raw entries, promoted memories, provenance chains, anxiety scores, processing status, and audit trails.

​
Quick Start

The dashboard opens in your browser automatically. No additional dependencies are required beyond kernle itself — the dashboard uses only Python standard library (http.server, json).

​
Options

​
Dashboard Tabs

​
Overview

Stats cards showing counts for each memory type (raw entries, episodes, beliefs, values, goals, notes, relationships, drives, suggestions). Below: anxiety dimension breakdown with bar charts and processing pipeline configuration.

​
Raw Entries

Filterable table of all raw entries with:
  • Status indicator — green dot (processed) or yellow circle (unprocessed)
  • Blob preview — first 120 characters of the raw content
  • Source — where the entry came from (cli, mcp, sdk, import)
  • Processed into — what memories were created from this entry
  • Strength bar — visual indicator of memory strength
Click any row to expand full details including the complete blob text and provenance chain (what memories were derived from this raw entry). Filters: All / Unprocessed / Processed. Adjustable limit (50-500).

​
Memories

Sub-tabs for each memory type: Episodes, Beliefs, Values, Goals, Notes, Relationships, Drives. Each shows a table with type-specific fields: Click any row to see full detail including all fields and the provenance chain — both what this memory was derived from and what memories were derived from it.

​
Suggestions

Pending memory suggestions with status (pending/promoted/rejected), type, confidence score, content preview, source raw IDs, and promotion target.

​
Audit Log

Chronological list of all memory operations with timestamp, operation type, memory type, memory ID, and actor. Adjustable limit.

​
Settings

Stack settings (enforce_provenance, stack_state, etc.) and processing pipeline configuration showing each transition layer with its enabled status, model, batch size, and thresholds.

​
Header Bar

The sticky header shows:
  • Stack ID — which stack you’re inspecting
  • Memory counts — quick summary of counts per type
  • Anxiety badge — colored badge showing overall anxiety score and level (Calm/Aware/Elevated/High/Critical)
  • Auto-refresh toggle — when enabled, refreshes data every 5 seconds

​
Typical Workflow

After seeding a corpus and processing it:
In the dashboard:
  1. Overview — verify memory counts match expectations, check anxiety score
  2. Raw Entries — filter to “Unprocessed” to see if anything was missed
  3. Memories → Beliefs — check that promoted beliefs are sensible
  4. Memories → Episodes — verify episode objective/outcome quality
  5. Audit Log — trace the processing pipeline’s decisions

​
API Endpoints

The dashboard exposes a read-only JSON API that can also be used programmatically: All memory endpoints include forgotten and weak memories (include_forgotten=True, include_weak=True) to give a complete view of the stack.

​
Security

The dashboard is designed for local development use:
  • Binds to 127.0.0.1 only (not accessible from the network)
  • Read-only — no mutation endpoints (GET only)
  • No CORS headers (same-origin only)
  • X-Content-Type-Options: nosniff on all responses
  • Generic error messages (no internal detail leakage)