Skip to content

RAG Chroma Demo Agent

This demo uses the turboagents-chroma retriever path inside SuperOptiX. Chroma stays the local collection backend, while TurboAgents adds compressed retrieval and reranking on top.

What This Demo Shows

  • TurboAgents-backed Chroma retrieval in the normal SuperOptiX RAG flow
  • persisted local Chroma collections
  • sentence-transformer embeddings with TurboAgents reranking
  • the playbook settings needed for a TurboAgents-backed Chroma path

Setup

Install the local model used by the demo:

super model install qwen3.5:9b
ollama serve

Pull, compile, and run the demo:

uv pip install "superoptix[turboagents]"
super agent pull rag_chroma_demo
super agent compile rag_chroma_demo
super agent run rag_chroma_demo --goal "What is the SuperOptiX framework?"

Playbook Configuration

language_model:
  location: local
  provider: ollama
  model: qwen3.5:9b
  api_base: http://localhost:11434
  temperature: 0.7
  max_tokens: 2048

rag:
  enabled: true
  retriever_type: turboagents-chroma
  config:
    top_k: 5
    chunk_size: 512
    chunk_overlap: 50
    similarity_threshold: 0.7
  vector_store:
    embedding_model: sentence-transformers/all-MiniLM-L6-v2
    embedding_dimension: 64
    collection_name: rag_demo_knowledge
    persist_directory: ./.superoptix/turboagents-chroma

Important points:

  • retriever_type: turboagents-chroma selects the TurboAgents path
  • embedding_dimension: 64 matches a supported TurboAgents compressed dimension
  • Chroma handles candidate persistence and search
  • TurboAgents reranks the candidate set before context is injected into the model

Validation Status

Current local validation for this path is narrower than LanceDB and SurrealDB. What is validated today is:

  • focused SuperOptiX tests for the TurboAgents vector-store and RAG mixin paths
  • direct RAGMixin smoke using the real TurboAgents Chroma adapter
  • local alignment on chromadb 1.5.5 in both TurboAgents and SuperOptiX

This means the Chroma integration is implemented and exercised locally, but this page does not claim a recorded end-to-end demo transcript yet.

Retrieval Stack

The runtime stack is:

  • sentence-transformers creates document and query embeddings
  • Chroma stores and retrieves the candidate set
  • TurboAgents reranks those candidates using compressed retrieval
  • SuperOptiX injects the selected context into the compiled framework runtime

Troubleshooting

Common checks:

curl http://localhost:11434/api/tags
super agent inspect rag_chroma_demo

If you need a clean local Chroma collection, remove the persisted demo directory and rerun the demo.