hellohal2064/cca-embedding-server repository overview

⁠CCA Embedding Server

Produces 4096-dimensional vector embeddings using Qwen3-Embedding-8B⁠ via vLLM pooling mode. Part of the CCA Full Stack⁠.

⁠What This Does

Provides an OpenAI-compatible embeddings API for CCA's code intelligence system:

• Codebase indexing — embeds source files into Qdrant for semantic search
• Note-taker memory — embeds CCA's observation notes for retrieval
• User profile matching — semantic similarity for user identification

Built on vllm-dgx-spark-gb10 with vLLM's pooling runner configured for embedding generation.

⁠Quick Start

docker run -d \
  --gpus all \
  --name qwen3-embedding \
  -v /data/models:/models:ro \
  -p 8200:8000 \
  hellohal2064/cca-embedding-server:latest

⁠Model Required

Download Qwen3-Embedding-8B⁠ (~8GB) to your models directory.

⁠API

# Generate embeddings (OpenAI-compatible)
curl http://localhost:8200/v1/embeddings \
  -H "Content-Type: application/json" \
  -d '{"input": "semantic search query", "model": "Qwen/Qwen3-Embedding-8B"}' 

# Health check
curl http://localhost:8200/health

⁠Configuration (baked into image)

⁠Part of CCA Stack

This is one of 11 containers in the CCA full stack. See:

• CCA agent⁠
• Full deployment guide⁠