Knowledge Contributions

Collaborative knowledge building from agent memories, experiences, and research

Overview

The Knowledge Contributions page (KnowledgeContributionsPage) is the most feature-rich page in the dashboard at 1,629 lines. It combines knowledge base management, agent contribution tracking, automated scheduling, rich analytics with interactive SVG charts, and search metrics into a five-tab interface. It represents the collaborative knowledge loop: agents learn → contribute → other agents search → better answers → more learning.

How It Works

1. Create contribution bases — Specialized knowledge bases for agent contributions, optionally with role tags (support, sales, engineering, HR, ops).
2. Agents contribute — Manually trigger contributions, or set up automated schedules. Agents scan their recent memories and contribute qualifying entries above a confidence threshold.
3. Review entries — Each entry has a quality score, confidence level, category, and status. Admins can approve, reject, archive, or vote on entries.
4. Monitor with analytics — Charts show contribution volume over time, confidence trends, per-agent breakdowns, category distribution, and daily heatmaps.
5. Track search usage — See how agents are searching knowledge bases — hit rates, query patterns, and which bases are most useful.

Key Concepts

Confidence Score

A 0–100% score indicating how certain the contributing agent is about the accuracy of the knowledge. Contributions below your minimum threshold (set per schedule) are flagged for review. High confidence (80%+) indicates well-verified knowledge; below 50% may be speculative.

Quality Score

An independent quality assessment of each entry, displayed as a colored progress bar. Green (80%+) = high quality, yellow (50–80%) = moderate, red (<50%) = needs improvement. Quality is determined by content completeness, clarity, and relevance.

Entry Lifecycle

Roles

Knowledge bases can be tagged with roles (support, sales, engineering, HR, ops, general) to help organize and filter by department. Each role gets a distinct color badge.

Tabs Overview

Knowledge Bases Tab

Displays knowledge bases as cards with name, role badge, description, entry count, contributor count, and last contribution date. Merges data from the contribution system's bases and the main knowledge bases (deduplicating by ID).

Base Detail View

Clicking a base opens an inline detail view with:

• Filter bar — Filter entries by category, status (pending/approved/rejected/archived), and minimum quality score (40%/60%/80%)
• Entry cards — Each entry shows title, category badge, status badge, quality bar, confidence percentage, tags, vote buttons (▲/▼ with counts), and action buttons
• Actions per entry — Approve, Reject, Archive (context-sensitive — only shows valid transitions)

Creating a Base

POST /api/engine/knowledge-contribution/bases
Content-Type: application/json

{
  "name": "Customer Support Knowledge",
  "description": "Solutions, troubleshooting, and best practices from support agents",
  "role": "support",
  "orgId": "org-abc123"
}

Contributions Tab

A paginated, searchable table of all contributions across all bases. Features:

• Search — Full-text search across title, agent name, content, and category
• Agent filter — Dropdown to filter by contributing agent
• Trigger button — Manually trigger a contribution cycle for any agent
• Detail modal — Click any row to see full content, metadata, tags, confidence, source, and category
• Pagination — 20 items per page with prev/next navigation

Triggering a Contribution

POST /api/engine/knowledge-contribution/contribute/{agentId}
Content-Type: application/json

{
  "targetBaseId": "base-xyz",   // optional — contributes to all eligible if omitted
  "orgId": "org-abc123"
}

Schedules Tab

Automate knowledge contributions on a recurring basis. Each schedule connects an agent to a target knowledge base with a frequency and confidence threshold.

POST /api/engine/knowledge-contribution/schedules
Content-Type: application/json

{
  "agentId": "agent-xyz",
  "targetBaseId": "base-abc",
  "frequency": "weekly",
  "dayOfWeek": "monday",
  "minConfidence": 0.7,
  "orgId": "org-abc123"
}

Stats & Charts Tab

The Stats tab renders six interactive SVG-based visualizations with hover tooltips:

1. Contributions Over Time (Line Chart)

Smooth catmull-rom curve showing daily contribution counts. Includes gradient area fill and interactive hover dots with date and count tooltips.

2. Confidence Level Over Time (Band Chart)

Weekly average confidence with min/max hover details. Color: purple (#8b5cf6). Y-axis: 0–100%.

3. Contributions by Agent (Horizontal Bar Chart)

Each agent gets a colored bar proportional to their contribution count. Hover shows count and average confidence.

4. Category Distribution (Donut Chart)

Breakdown by importance category (high/medium/low/normal). Interactive slices with count and percentage tooltips. Legend displayed alongside.

5. Agent Quality Comparison

Dual horizontal bars per agent: contribution count (agent color) and confidence score (green/yellow/red). Enables quick quality-vs-quantity comparison.

6. Daily Activity Heatmap

Grid table with agents as rows, days as columns. Cell intensity scales with contribution count. Each agent has a unique color from the CHART_COLORS palette.

All charts support time range filtering (7/14/30/60/90 days) and per-agent filtering.

Search Metrics Tab

Tracks how agents use knowledge search, loaded from /knowledge-contribution/search-metrics:

• Summary cards — Total searches, KB searches, Hub searches, Hit rate
• Search efficiency by agent — Table with total, helpful, and hit rate per agent
• Search activity timeline — Daily bar chart colored by helpfulness
• Recent searches — Table with agent, type (KB/Hub), query, result count, top score, helpfulness, and timestamp

Entry Management

Voting

Entries support upvote/downvote via POST /knowledge-contribution/entries/{id}/vote with { "direction": "up" } or "down". Vote counts are displayed next to the arrow buttons.

Status Transitions

// Approve an entry
PUT /api/engine/knowledge-contribution/entries/{id}/approve

// Reject an entry
PUT /api/engine/knowledge-contribution/entries/{id}/reject

// Archive an entry
PUT /api/engine/knowledge-contribution/entries/{id}/archive

API Reference

Best Practices

• Start with weekly schedules at 70% confidence — This balances volume with quality. Adjust down for more coverage, up for higher accuracy.
• Review pending entries regularly — Low-confidence contributions need human validation before they can mislead other agents.
• Use the quality comparison chart — An agent with few but high-confidence contributions may be more valuable than one producing many low-quality entries.
• Monitor hit rates — Low search hit rates indicate gaps in your knowledge bases. Use "no result" queries to identify what content to import.
• Create role-specific bases — Separate support knowledge from engineering knowledge for cleaner retrieval.
• Use voting for community curation — Upvote high-value entries and downvote inaccurate ones to surface the best knowledge.

Troubleshooting