Strata
@agenerationforwordz-tech
Strata is a self hosted AI memory server. Your AI remembers everything across every session, on your own hardware.
Overview
What is Strata?
Strata is a self‑hosted AI memory server that gives any MCP‑compatible AI agent persistent, long‑term memory on your own hardware. It stores thoughts as vector embeddings, provides semantic search, a file vault, and a 3D constellation viewer — and runs on a Raspberry Pi.
How to use Strata?
Clone the repository, create a Python virtual environment, install dependencies (pip install -r requirements.txt), and run python server.py. The web dashboard is accessible at http://localhost:4320/dashboard. Per‑agent API keys can be managed from /admin/agents, and the server can be configured with environment variables such as STRATA_DEMO_MODE=true for a quick demo.
Key features of Strata
- Semantic search by meaning, not keywords.
- File vault with attached source files.
- Per‑agent API keys with granular permissions.
- PostgreSQL pgvector backend for production scale.
- 3D Constellation viewer with live activity.
- Optional AES‑256 encryption at rest.
Use cases of Strata
- Provide persistent memory for AI coding assistants like Claude Code.
- Self‑host a personal knowledge base with semantic recall.
- Audit and trace AI agent actions with daily CSV logs.
- Attach files to thoughts for project context retrieval.
- Run entirely on low‑power hardware like a Raspberry Pi 4B.
FAQ from Strata
What MCP servers or AI tools does Strata integrate with?
Strata works with any MCP‑compatible AI, including Claude Code, Codex CLI, and other tools that support the Model Context Protocol.
Does Strata run on a Raspberry Pi?
Yes, the entire system — semantic search, file storage, and embeddings — runs on a Raspberry Pi 4B with 4 GB of RAM.
Where is my data stored?
Data is stored locally in a SQLite or PostgreSQL database (with optional pgvector). You can also enable AES‑256 encryption at rest via SQLCipher.
What are the main dependencies?
Strata requires Python, ONNX Runtime, FastEmbed, and optionally PostgreSQL 17 with pgvector for production deployments. Initial install downloads about 700 MB of binary wheels.
How do agents authenticate?
Each agent receives a unique API key with granular read/write/delete/admin/kill permissions, managed from the /admin/agents panel. A global MCP kill switch can lock all agents out instantly.
