Helix

Helix: The first Autonomous Living Repository. Agentic Graph RAG and WASM execution to simulate, predict, and autonomously evolve codebases.

Flow Guide

If you feel lost on first open, use the in-app Flow Guide to follow the recommended path from ingestion → exploration → analysis → swarm actions. It auto-opens once per device and can be reopened from Settings in the Ghost Dock.

LLM Providers (Gemini or Qwen)

Helix supports Google Gemini and OpenAI-compatible endpoints (Qwen, vLLM, Ollama, LM Studio).

Switch to Qwen (OpenAI-compatible)

Set these environment variables in .env:

LLM_PROVIDER=qwen
QWEN_BASE_URL=http://localhost:8000/v1
QWEN_MODEL=Qwen/Qwen3.5-397B-A17B  # or your locally served Qwen model id
QWEN_API_KEY=dummy

Embeddings with Qwen/OpenAI-compatible

EMBEDDING_PROVIDER=qwen
EMBEDDING_MODEL=your-embedding-model
QWEN_BASE_URL=http://localhost:8000/v1
QWEN_API_KEY=dummy

If you leave LLM_PROVIDER unset, Helix defaults to Gemini (set GOOGLE_API_KEY and optionally GEMINI_MODEL).

System Architecture

Helix is organized into four layers:

1. Ingestion & Indexing: parses repositories, builds AST metadata, and populates the graph/vector stores.
2. Knowledge Core: Neo4j holds dependency graphs; Qdrant stores embeddings for semantic search.
3. Agentic Reasoning: the Immune Swarm runs multi-step analysis, proposes mutations, and streams logs over WebSockets.
4. Spatial UI: the 3D graph, overlays, and Ghost Dock provide progressive disclosure and guided workflows.