conduit-ui/knowledge

AI-powered knowledge base with semantic search and ChromaDB integration

dev-master 2026-01-04 20:20 UTC

Requires

• php: ^8.2
• illuminate/database: ^12.17
• laravel-zero/framework: ^12.0.2

Requires (Dev)

• fakerphp/faker: ^1.23
• laravel/pint: ^1.25.1
• mockery/mockery: ^1.6.12
• pestphp/pest: ^3.8.4|^4.1.2
• phpstan/extension-installer: *
• phpstan/phpstan: ^1.0
• phpstan/phpstan-deprecation-rules: ^1.2
• phpstan/phpstan-strict-rules: ^1.6

MIT

• Jordan Partridge

• cli
• ai
• knowledge
• chromadb
• semantic-search

README

AI-powered knowledge base with semantic search and ChromaDB integration.

Vision

Build an intelligent knowledge management system that:

• Semantic Search: Find knowledge by meaning, not just keywords (ChromaDB)
• AI-Powered: Confidence scoring, relevance ranking, smart suggestions
• Git Integration: Automatic context capture, knowledge attribution, evolution tracking
• Test-Driven: 100% coverage enforced via Synapse Sentinel
• Quality-First: Maximum static analysis (PHPStan level 8)

Features

Git Context Integration

Automatically capture and track git metadata for knowledge attribution:

• Auto-detection: Automatically captures repo, branch, commit, and author when adding entries
• Knowledge Attribution: "Git blame" for knowledge - see who documented what and when
• Evolution Tracking: Track how knowledge changes across commits and branches
• Cross-repo Support: Store full repository URLs for knowledge shared across projects

Commands

# Add entry with automatic git context detection
./know knowledge:add "Fix Database Connection" --content="Always check .env configuration"

# Skip git detection for sensitive repositories
./know knowledge:add "API Keys" --content="Store in vault" --no-git

# Manual git field overrides
./know knowledge:add "Config" --content="..." --repo="custom/repo" --branch="main"

# Display current git context
./know knowledge:git:context

# List entries from a specific commit
./know knowledge:git:entries abc123def456

# List entries by author
./know knowledge:git:author "John Doe"

Auto-Detection Behavior

When you run knowledge:add in a git repository:

1. Repository: Detects remote origin URL (falls back to local path)
2. Branch: Captures current branch name
3. Commit: Records current commit hash (full SHA-1)
4. Author: Uses git config user.name

To disable auto-detection, use the --no-git flag.

Manual Overrides

You can override auto-detected values:

./know knowledge:add "Title" \
  --content="Content here" \
  --repo="https://github.com/org/repo" \
  --branch="feature/new-feature" \
  --commit="abc123def456" \
  --author="Jane Smith"

Non-Git Directories

The tool gracefully handles non-git directories:

• No errors or warnings
• Git fields remain null
• All other functionality works normally

Git Hooks Integration

You can integrate with git hooks for automatic knowledge capture:

Post-commit hook example (.git/hooks/post-commit):

#!/bin/bash
# Automatically prompt for knowledge entry after each commit

echo "Would you like to document this commit? (y/n)"
read -r response

if [[ "$response" =~ ^[Yy]$ ]]; then
    ./know knowledge:add "$(git log -1 --pretty=%s)" \
        --content="$(git log -1 --pretty=%B)" \
        --category=debugging
fi

Pre-push hook example (.git/hooks/pre-push):

#!/bin/bash
# Check for undocumented commits

UNDOCUMENTED=$(git log origin/main..HEAD --format="%H" | while read commit; do
    if ! ./know knowledge:git:entries "$commit" | grep -q "Total entries: 0"; then
        echo "$commit"
    fi
done)

if [ -n "$UNDOCUMENTED" ]; then
    echo "Warning: Some commits lack knowledge entries"
    echo "$UNDOCUMENTED"
fi

Confidence Scoring & Usage Analytics

Track knowledge quality and usage over time:

# Validate an entry (boosts confidence)
./know knowledge:validate 1

# View stale entries needing review
./know knowledge:stale

# Display analytics dashboard
./know knowledge:stats

Collections

Organize knowledge into logical groups:

# Create a collection
./know knowledge:collection:create "Deployment Runbook" --description="Production deployment steps"

# Add entries to collection
./know knowledge:collection:add "Deployment Runbook" 1
./know knowledge:collection:add "Deployment Runbook" 2 --sort-order=10

# View collection
./know knowledge:collection:show "Deployment Runbook"

# List all collections
./know knowledge:collection:list

Search & Discovery

# Search by keyword
./know knowledge:search --keyword="database"

# Search by tag
./know knowledge:search --tag="sql"

# Search by category and priority
./know knowledge:search --category="debugging" --priority="critical"

# List entries with filters
./know knowledge:list --category=testing --limit=10 --min-confidence=75

Development

# Install dependencies
composer install

# Run tests
composer test

# Run tests with coverage
composer test-coverage

# Format code
composer format

# Static analysis
composer analyse

# Run single test file
vendor/bin/pest tests/Feature/GitContextServiceTest.php

Quality Standards

• Test Coverage: 100% (enforced by Synapse Sentinel)
• Static Analysis: PHPStan Level 8
• Code Style: Laravel Pint (Laravel preset)
• Auto-Merge: PRs auto-merge after certification

Architecture

Services

GitContextService - Git metadata detection and retrieval

• Detects git repositories using git rev-parse
• Retrieves branch, commit, author information
• Handles non-git directories gracefully
• Supports custom working directories for testing

ConfidenceService - Knowledge quality scoring

• Age-based confidence decay
• Validation boosts
• Stale entry detection

CollectionService - Knowledge organization

• Create and manage collections
• Add/remove entries with sort ordering
• Duplicate prevention

Models

• Entry - Knowledge entries with git metadata, confidence scores, usage tracking
• Collection - Groups of related entries
• Tag - Normalized tags with usage counts
• Relationship - Links between entries (related-to, supersedes, depends-on)

Status

Active development. Core features implemented:

• Database schema and models
• Basic CLI commands
• Collections support
• Git context integration
• Confidence scoring and analytics

Semantic Search with ChromaDB

Advanced vector-based semantic search for finding knowledge by meaning:

Quick Start (Docker)

Start both ChromaDB and the embedding server with a single command:

# Start services
make up

# Check status
make status

# View logs
make logs

# Stop services
make down

This starts:

• ChromaDB on http://localhost:8000 - Vector database for semantic search
• Embedding Server on http://localhost:8001 - Sentence-transformers (all-MiniLM-L6-v2)

Manual Installation (Alternative)

If you prefer not to use Docker:

# Install ChromaDB server
pip install chromadb
chroma run --path ./chroma_data

# In another terminal, start embedding server
pip install flask sentence-transformers gunicorn
python docker/embedding-server/server.py

Configuration

Enable ChromaDB in your .env file:

SEMANTIC_SEARCH_ENABLED=true
EMBEDDING_PROVIDER=chromadb
CHROMADB_ENABLED=true
CHROMADB_HOST=localhost
CHROMADB_PORT=8000
CHROMADB_EMBEDDING_SERVER=http://localhost:8001
CHROMADB_EMBEDDING_MODEL=all-MiniLM-L6-v2

Usage

Once configured, semantic search automatically works with the existing search commands:

# Semantic search will automatically be used
./know knowledge:search --keyword="database connection issues"

# Results are ranked by semantic similarity and confidence
# Falls back to keyword search if ChromaDB is unavailable

How It Works

1. When you add/update entries, embeddings are generated and stored in both:

   • ChromaDB vector database (for fast similarity search)
   • SQLite database (as JSON, for fallback)

2. Search queries are:

   • Converted to embedding vectors
   • Compared against indexed entries using cosine similarity
   • Ranked by: similarity_score * (confidence / 100)
   • Filtered by metadata (category, tags, status, etc.)

3. If ChromaDB is unavailable:

   • Automatically falls back to SQLite-based semantic search
   • Or keyword search if no embeddings are available

Architecture

ChromaDBClient - ChromaDB REST API client

• Collection management
• Document indexing (add/update/delete)
• Vector similarity search

ChromaDBEmbeddingService - Text embedding generation

• Generates embedding vectors using specified model
• Calculates cosine similarity between vectors
• Graceful error handling

ChromaDBIndexService - Index management

• Indexes entries on create/update
• Removes entries on delete
• Batch indexing support
• Automatic embedding generation and storage

SemanticSearchService - Hybrid search orchestration

• ChromaDB search (when available)
• SQLite fallback search
• Keyword search fallback
• Metadata filtering (category, tags, module, priority, status)

Coming soon:

• Export and publishing
• Web interface

License

MIT License. See LICENSE for details.