cqs 1.15.2

Code intelligence and RAG for AI agents. Semantic search, call graphs, impact analysis, type dependencies, and smart context assembly — in single tool calls. 52 languages + L5X/L5K PLC exports, 90.9% Recall@1 (BGE-large), 0.949 MRR (296 queries). Local ML, GPU-accelerated.
Documentation
# Privacy


## Data Stays Local


cqs processes your code locally by default. With `--llm-summaries`, function code is sent to Anthropic's API for one-sentence summary generation. With `--improve-docs`, LLM-generated doc comments are written back to your source files. With `--hyde-queries`, function descriptions are sent to Anthropic's API for synthetic search query generation. See [Anthropic's privacy policy](https://www.anthropic.com/privacy). Without these flags, nothing is transmitted externally and no source files are modified.

- **No telemetry by default**: Optional local-only command logging when `CQS_TELEMETRY=1` is set. Stored in `.cqs/telemetry.jsonl`, never transmitted
- **No analytics**: No tracking of any kind
- **No cloud sync**: Index stays in your project directory

## What Gets Stored


When you run `cqs index`, the following is stored in `.cqs/index.db`:

- Code chunks (functions, methods, documentation sections)
- Embedding vectors (dimension depends on configured model; 1024 for BGE-large default, 768 for E5-base/v9-200k presets)
- File paths and line numbers
- File modification times

This data never leaves your machine.

## Model Download


The embedding model is downloaded once from HuggingFace:

- Default: `BAAI/bge-large-en-v1.5` (BGE-large, ~1.2GB, 1024-dim)
- Preset: `intfloat/e5-base-v2` (E5-base, ~438MB, 768-dim)
- Preset: `jamie8johnson/e5-base-v2-code-search` (v9-200k LoRA, ~417MB, 768-dim)
- Custom: any HuggingFace repo via `[embedding]` config section, `--model` CLI flag, or `CQS_EMBEDDING_MODEL` env var
- Size varies by model
- Cached in: `~/.cache/huggingface/`

HuggingFace may log download requests per their privacy policy. Custom model configurations cause downloads from the specified HuggingFace repository. After download, the model runs offline.

## CI/CD


If you fork or contribute to the cqs repository:

- GitHub Actions runs tests on push/PR
- Code is processed on GitHub-hosted runners
- No index data is uploaded (only source code)
- See GitHub's privacy policy for runner data handling

## Deleting Your Data


To remove all cqs data:

```bash
rm -rf .cqs/                          # Project index
rm -rf ~/.local/share/cqs/refs/       # Reference indexes
rm -rf ~/.config/cqs/projects.toml    # Project registry
rm -f ~/.config/cqs/config.toml       # User configuration
rm -f .cqs.toml                       # Project config
rm -f docs/notes.toml                 # Project notes
rm -rf ~/.cache/huggingface/          # Downloaded model
```