wezzel.com — Documentation

Install Cicerone

Bash

curl -L https://idm.wezzel.com/crab-meat-repos/ \
  cicerone-goclaw/-/releases/latest/downloads/ \
  cicerone_linux_amd64 -o cicerone
chmod +x cicerone && sudo mv cicerone /usr/local/bin/
                    

Install ChromaDB

Bash

# pip install
pip install chromadb
chroma --host 0.0.0.0 --port 8000 &

# or Docker
docker run -d -p 8000:8000 -v chroma_data:/chroma/chroma \
  chromadb/chroma
                    

Configure

config.yaml

gateway:
  listenAddr: "0.0.0.0:14170"
plugins:
  allow: [web-search, browser, tts, rag]
  env:
    RAG_CHROMA_URL: "http://localhost:8000"
    RAG_COLLECTION: "my_notes"
    RAG_TOP_K: "5"
    RAG_MODULES: "search,retrieve,memory,fusion,adapt,predict"
                    

Commands

RAG Commands

# Index a directory into ChromaDB
cicerone rag index ~/notes

# Query your knowledge base
cicerone rag query "How do I set up TLS?"

# List indexed collections
cicerone rag list

# Show pipeline status
cicerone rag status

# Remove a collection
cicerone rag remove old_collection
                

Injection Modes

Control how RAG context enters your LLM conversations.

Auto

Always inject RAG context when a collection exists. Zero-friction intelligence.

RAG_INJECT_MODE: "auto"

On-Demand

Only inject when triggered: "ask my notes", "check my documents", "from my notes".

RAG_INJECT_MODE: "on-demand"

Off

Disable RAG injection entirely. Still available via explicit commands.

RAG_INJECT_MODE: "off"

Architecture

Package Structure

rag/
├── rag.go              # Version, module registry
├── config.go           # RAGConfig, defaults, validation
├── types.go            # Result, Chunk, PipelineState, PipelineResult
├── store.go            # Store interface (ChromaDB abstraction)
├── chroma_client.go    # ChromaDB REST client
├── embedder.go          # Embedder interface + ChromaEmbedder
├── chunker.go           # 3 chunking strategies
├── walker.go            # Directory walker
├── indexer.go            # Walk → Chunk → Store pipeline
├── modules.go           # Module interface + registry
├── module_search.go     # Query classification
├── module_retrieve.go   # ChromaDB query + MockStore
├── module_memory.go      # LRU query cache
├── module_fusion.go      # Dedup + sort by relevance
├── module_adapt.go      # Format selection per query type
├── module_predict.go    # Structured answer generation
├── pipeline.go           # Orchestrator: Execute(), partial results
├── plugin_rag.go        # RAGPlugin (cicerone Plugin interface)
├── context_injector.go  # LLM context injection
├── collection_manager.go # Multi-collection management
└── resilience.go         # QueryCache + CircuitBreaker
                