Installation

You can install Strata with a one-liner, grab it from source, or wait for the PyPI release (whenever that lands on the calendar). Your call.

Quick Install (Recommended)

pip install git+https://github.com/jeremykamber/strata-memory.git

That's it. This installs the strata CLI globally and makes the strata Python package importable. Zero pip dependencies. Nothing to clone. No extra steps. You're done. Go grab a coffee.

Requirements

Python 3.9+ (stdlib only - no pip dependencies, we're not animals)
Node.js (optional, for QMD hybrid search)
OS: macOS and Linux fully supported. Windows hasn't been tested yet - contributions welcome, no pressure.

From Source

If you'd rather clone the repo and poke around locally:

git clone https://github.com/jeremykamber/strata-memory.git
cd strata-memory
pip install -e .

Editable mode (-e) means you can tweak the package and see changes immediately. Handy for contributors, curious tinkerers, or anyone who enjoys living on the bleeding edge.

From PyPI (When Published)

pip install strata-memory

For the cautious among us (hi, that's me):

pip install strata-memory==0.2.0

Version pinning is your friend in production. Trust me on this one.

Verify Installation

strata init
strata status

If both commands run without screaming, you're in business. init creates your data directory and walks you through search backend selection. status shows you the system state - healthy, we hope.

Optional: QMD Hybrid Search

QMD by Tobias Lutke brings you BM25 full-text search combined with vector embeddings. All local. No LLM required. It's pretty neat.

# Install QMD globally via npm
npm install -g @tobilu/qmd

# Configure Strata to use it
strata qmd-setup
strata qmd-embed

If you select QMD during strata init, Strata will try to auto-install it via npx (there's a 30-second timeout, so don't wander off too far). If that fails, just install manually with the commands above. It happens.

QMD Reranker Providers (Optional)

Want even better search results? QMD supports LLM reranker providers:

OpenAI (openai://gpt-4o-mini)
Anthropic (anthropic://claude-3-haiku)
Ollama (ollama://llama3) - local inference, no cloud required
Local GGUF - fully local, no API calls, maximum privacy

Rerankers chew through API credits or your local compute, depending on which route you take. Set one up during strata init, or come back later with strata config set qmd_reranker <provider>.

Fallback: FTS5 Keyword Search

No QMD? No problem. If you choose FTS5 during init - or QMD just isn't installed - Strata falls back to the built-in SQLite FTS5 shadow index for archive searches. Active and cooled strata get searched via trusty old filesystem grep. Works without Node.js, no extra setup, no fuss.

See Search for the full breakdown on search backends and their tradeoffs.

Platform Notes

macOS / Linux: Fully supported. Everything works - daemon mode, background Janitor, signal handling. You're in good hands.
Windows: Not yet tested. Path handling and daemon mode might need some TLC. If you're a Windows user with opinions, we'd love to hear from you.

Upgrading

# From git (recommended)
pip install --upgrade git+https://github.com/jeremykamber/strata-memory.git

# From source
cd strata-memory
git pull
pip install -e .

# From PyPI
pip install --upgrade strata-memory

After upgrading, run strata status to make sure everything's still in one piece. Configuration files and existing data play nice across updates within the same major version - we've got your back.