Full Mode Deployment¶

Full Mode provides all features including Code Graph, Memory Store, Knowledge RAG, and LLM-powered auto-extraction.

Complete Feature Set¶

All Features Enabled¶

✅ Code Graph: Repository indexing, search, impact analysis
✅ Memory Store: Project knowledge with vector search
✅ Knowledge RAG: Document processing and intelligent Q&A
✅ Auto-Extraction: LLM-powered memory extraction from:
Git commits
Code comments (TODO, FIXME, NOTE)
AI conversations
Knowledge base queries

Use Cases¶

Full-featured AI coding assistant
Intelligent documentation systems
Automated knowledge capture
Enterprise code intelligence platform

System Requirements¶

With Local LLM (Ollama)¶

CPU: 8+ cores (16+ recommended)
RAM: 16GB minimum (32GB recommended)
GPU: Optional but highly recommended (8GB+ VRAM)
Disk: 100GB SSD

With Cloud LLM¶

CPU: 4 cores
RAM: 8GB
Disk: 50GB SSD
API Access: OpenAI, Gemini, or OpenRouter

Quick Start¶

1. Choose LLM Provider¶

Ollama (Local, Private)OpenAI (Cloud, Best Quality)Google Gemini (Cloud, Cost-Effective)OpenRouter (Multi-Provider)

# Install Ollama
curl -fsSL https://ollama.com/install.sh | sh

# Pull models
ollama pull llama3.2          # 8B parameter model
ollama pull nomic-embed-text  # Embedding model

# For better quality (requires more RAM)
# ollama pull mistral:7b
# ollama pull qwen2.5:14b

# Get API key
# Visit: https://platform.openai.com/api-keys
export OPENAI_API_KEY=sk-proj-...

# Get API key
# Visit: https://makersuite.google.com/app/apikey
export GOOGLE_API_KEY=AIza...

# Get API key
# Visit: https://openrouter.ai/keys
export OPENROUTER_API_KEY=sk-or-v1-...

2. Configure Environment¶

# Copy full template
cp docker/.env.template/.env.full .env

# Edit configuration
nano .env

Example with Ollama:

# Neo4j
NEO4J_URI=bolt://neo4j:7687
NEO4J_USER=neo4j
NEO4J_PASSWORD=your_secure_password
NEO4J_DATABASE=neo4j

# Deployment Mode - Enable all features
DEPLOYMENT_MODE=full
ENABLE_KNOWLEDGE_RAG=true
ENABLE_AUTO_EXTRACTION=true

# LLM Configuration
LLM_PROVIDER=ollama
OLLAMA_BASE_URL=http://host.docker.internal:11434
OLLAMA_MODEL=llama3.2

# Embedding Configuration
EMBEDDING_PROVIDER=ollama
OLLAMA_EMBEDDING_MODEL=nomic-embed-text

3. Start Services¶

With Bundled OllamaWith External Ollama

# Start with Ollama container included
make docker-full-with-ollama

# Or
docker-compose -f docker/docker-compose.full.yml --profile with-ollama up -d

# Start without Ollama (use system Ollama)
make docker-full

# Or
docker-compose -f docker/docker-compose.full.yml up -d

4. Verify Deployment¶

# Check all containers
docker ps
# Should see: mcp, neo4j, (optionally ollama)

# Test LLM
curl http://localhost:11434/api/generate -d '{
  "model": "llama3.2",
  "prompt": "Hello, how are you?",
  "stream": false
}'

# Test embedding
curl http://localhost:11434/api/embeddings -d '{
  "model": "nomic-embed-text",
  "prompt": "test embedding"
}'

# Check service health (if using FastAPI)
curl http://localhost:8000/api/v1/health

Available MCP Tools¶

Full mode provides 30 tools across 6 categories:

Code Graph Tools (4)¶

code_graph_ingest_repo
code_graph_fulltext_search
code_graph_impact_analysis
code_graph_pack_context

Memory Management Tools (7)¶

add_memory
search_memories
get_memory
update_memory
delete_memory
supersede_memory
get_project_summary

Auto-Extraction Tools (5) - New!¶

extract_from_conversation
extract_from_git_commit
extract_from_code_comments
suggest_memory_from_query
batch_extract_from_repository

Knowledge RAG Tools (8) - New!¶

knowledge_add_document
knowledge_add_directory
knowledge_query
knowledge_search
knowledge_list_documents
knowledge_delete_document
knowledge_update_document
knowledge_get_stats

Task Queue Tools (4)¶

task_submit
task_status
task_cancel
list_tasks

System Tools (2)¶

health_check
system_info

Advanced Features¶

Auto-Extraction from Git Commits¶

Automatically extract decisions and learnings:

{
  "tool": "extract_from_git_commit",
  "input": {
    "project_id": "myapp",
    "commit_sha": "abc123...",
    "commit_message": "feat: implement JWT authentication\n\nAdded JWT middleware for API auth",
    "changed_files": ["src/auth/jwt.py", "src/middleware/auth.py"],
    "auto_save": true
  }
}

Mine Code Comments¶

Extract TODOs and decisions from code:

{
  "tool": "extract_from_code_comments",
  "input": {
    "project_id": "myapp",
    "file_path": "src/api/routes.py"
  }
}

Conversation Analysis¶

Extract memories from AI conversations:

{
  "tool": "extract_from_conversation",
  "input": {
    "project_id": "myapp",
    "conversation": [
      {"role": "user", "content": "Should we use Redis or Memcached?"},
      {"role": "assistant", "content": "Redis is better because..."}
    ],
    "auto_save": false
  }
}

Knowledge RAG¶

Process and query documents:

{
  "tool": "knowledge_add_document",
  "input": {
    "file_path": "/docs/architecture.md",
    "metadata": {"type": "architecture", "version": "1.0"}
  }
}

{
  "tool": "knowledge_query",
  "input": {
    "query": "How does the authentication system work?",
    "max_results": 5
  }
}

Batch Repository Extraction¶

Comprehensive analysis:

{
  "tool": "batch_extract_from_repository",
  "input": {
    "project_id": "myapp",
    "repo_path": "/repos/myapp",
    "max_commits": 100,
    "file_patterns": ["*.py", "*.js", "*.go"]
  }
}

LLM Provider Comparison¶

Ollama (Local)¶

Pros: - Free and private - No API limits - Works offline - Full control

Cons: - Requires powerful hardware - Slower than cloud - Manual model management

Recommended Models: - llama3.2 (8B) - Good balance - mistral (7B) - Fast - qwen2.5 (14B) - Better quality (needs 16GB+ RAM)

OpenAI¶

Pros: - Best quality - Fast responses - No infrastructure needed

Cons: - Costs money - Requires internet - Data sent to OpenAI

Cost (Nov 2024): - GPT-4o: $5/$15 per 1M tokens (in/out) - GPT-4o-mini: $0.15/$0.60 per 1M tokens - Embeddings: $0.02 per 1M tokens

Google Gemini¶

Pros: - Cost-effective - Good quality - Fast

Cons: - Requires internet - Data sent to Google

Cost: - Gemini 1.5 Flash: Lower cost - Gemini 1.5 Pro: Higher quality - Free tier available

Performance Optimization¶

Ollama GPU Acceleration¶

# Add to docker-compose.full.yml
services:
  ollama:
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]

Neo4j Performance for Large Scale¶

# In docker-compose.full.yml
NEO4J_server_memory_heap_initial__size=4G
NEO4J_server_memory_heap_max__size=8G
NEO4J_server_memory_pagecache_size=4G
NEO4J_dbms_memory_transaction_total_max=2G

LLM Context Optimization¶

# Use context packing to stay within token limits
tool: code_graph_pack_context
input: {
  "entry_points": ["src/main.py"],
  "task_type": "implement",
  "token_budget": 8000  # Adjust based on model
}

Cost Estimation¶

Local Deployment (Ollama)¶

VPS: $40-80/month (32GB RAM, 8 cores)
GPU VPS: $100-200/month (with GPU)
LLM: $0
Embeddings: $0
Total: $40-200/month

Cloud Deployment (OpenAI)¶

VPS: $10-20/month (8GB RAM)
LLM: $20-100/month (depends on usage)
Embeddings: $1-5/month
Total: $31-125/month

Hybrid (Ollama Embeddings + OpenAI LLM)¶

VPS: $10-20/month
LLM: $20-100/month
Embeddings: $0 (local)
Total: $30-120/month

Production Deployment¶

High availability configuration
Backup strategies
Monitoring setup
Security hardening
Scaling considerations

Troubleshooting¶

LLM Generation Fails¶

# Check Ollama
curl http://localhost:11434/api/generate -d '{
  "model": "llama3.2",
  "prompt": "test"
}'

# Check model is pulled
ollama list

# View Ollama logs
docker logs codebase-rag-ollama

Out of Memory Errors¶

# Check memory usage
docker stats

# Reduce model size
ollama pull llama3.2:3b  # Smaller 3B model

# Or increase Docker memory limit
# Docker Desktop: Settings → Resources → Memory

Slow Response Times¶

# Enable GPU acceleration (if available)
# Check GPU is detected
nvidia-smi

# Or switch to smaller model
OLLAMA_MODEL=mistral  # 7B instead of 13B

# Or use cloud LLM for faster responses
LLM_PROVIDER=openai

Next Steps¶

Knowledge RAG Guide - Document processing
Auto-Extraction Guide - Automated memory capture