README.md

metadata

title: OpenCode Hub
emoji: 🤖
colorFrom: blue
colorTo: indigo
sdk: docker
pinned: false
license: mit
short_description: OpenCode AI coding agent with AirLLM + ChromaDB + turbo

OpenCode Hub — HF Space

Open-source AI coding agent with memory-optimized inference.

Features

• AirLLM — Run 70B models on 4GB GPU via layer-by-layer loading
• ChromaDB — Vector store for RAG (retrieval-augmented generation)
• turbo (turbopuffer) — High-performance vector search index
• OpenCode — Full open-source AI coding agent API
• FastAPI — REST API compatible with the Replit OpenCode Hub frontend

Models Supported

• meta-llama/Meta-Llama-3-70B-Instruct (4GB VRAM via AirLLM)
• Qwen/Qwen2.5-72B-Instruct
• mistralai/Mistral-7B-Instruct-v0.3
• Any HuggingFace model

API Endpoints

GET  /health          — Health check
GET  /models          — List available models
POST /generate        — Generate text with AirLLM
POST /embed           — Generate embeddings
GET  /collections     — List ChromaDB collections
POST /collections/{n}/search — Semantic search
POST /collections/{n}/add    — Add documents
GET  /stats           — Memory and performance stats

Environment Variables

• HF_TOKEN — Hugging Face access token (auto-configured)
• MODEL_ID — Default model (default: meta-llama/Meta-Llama-3-70B-Instruct)
• MAX_GPU_MEMORY_GB — GPU memory limit in GB (default: 4)