Privacy & Data Storage

Indexed is designed with privacy as a core principle. This page explains exactly where your data goes and what network calls are made.

Everything Runs Locally

The entire indexing pipeline — parsing, chunking, embedding, and searching — runs on your machine. There is no Indexed server, no cloud service, no API calls to third parties.

The Embedding Model

Indexed uses all-MiniLM-L6-v2, a Sentence Transformers model that runs locally via PyTorch. The model is downloaded once (about 80 MB) and cached on your machine. After that, no network calls are made during embedding or search.

Your documents are never sent to HuggingFace, OpenAI, or any external embedding API.

Search Is Local

When you run indexed index search or an AI assistant searches via MCP, the query is embedded locally using the same model, and FAISS performs the nearest-neighbor lookup on your local index. Nothing leaves your machine.

What Goes Over the Network

The only network calls Indexed makes are to your own infrastructure:

That's it. No telemetry, no analytics, no usage reporting, no phoning home.

Where Data Is Stored

All Indexed data lives in a single directory:

~/.indexed/
├── config.toml                          # Your configuration
└── data/
    └── collections/
        ├── my-docs/
        │   ├── manifest.json            # Collection metadata
        │   ├── documents.json           # Document metadata
        │   ├── chunks.json              # Text chunks
        │   └── index.faiss              # Vector index
        └── eng-tickets/
            ├── manifest.json
            ├── documents.json
            ├── chunks.json
            └── index.faiss

What each file contains

• manifest.json — collection type, source path/URL, creation timestamp, query parameters. No document content.
• documents.json — metadata about each document (title, ID, source path). No full content.
• chunks.json — the actual text chunks that were embedded. This contains excerpts of your documents.
• index.faiss — the vector index. Contains numerical vectors, not readable text, but the vectors could theoretically be used to approximate the original text.

Docker Considerations

When running Indexed in Docker, you typically mount ~/.indexed as a volume:

docker run -v ~/.indexed:/root/.indexed indexed mcp run --transport http

This means your data stays on your host machine. The container doesn't store any persistent data itself.

If you're running the Docker container on a shared server, ensure the volume mount has appropriate filesystem permissions.

No Telemetry

Indexed does not collect any telemetry, usage data, or crash reports. There is no opt-in or opt-out because there is nothing to opt into.

What's Next