piLoci: Self-hosted team memory service for AI agents on Raspberry Pi
piLoci is a self-hosted, multi-user LLM memory service that runs on Raspberry Pi 5, designed for teams to share project context across AI agents. It combines an MCP-based interface with LanceDB vector storage and an Obsidian-style workspace, enabling project-scoped memory isolation while keeping data on-premise. Teams use piLoci to maintain persistent context across sessions without relying on cloud-based memory services.
Use cases
- Teams maintaining persistent project context across multiple AI coding assistant sessions
- Research groups sharing and curating domain knowledge accessible to AI agents
- Software teams storing architectural decisions, design patterns, and technical context
- Content teams maintaining briefs, brand guidelines, and creative direction accessible to AI tools
- Open source projects providing maintainers with a self-hosted alternative to cloud memory services
Key features
- Project-Scoped Memory Isolation: Strict separation of context between projects, ensuring team members only access relevant memories
- MCP-Based Interface: Standard Model Context Protocol surface for programmatic memory storage and retrieval across AI clients
- LanceDB Vector Storage: Embedded semantic search with no external database process to manage
- Multi-User Authentication: Per-user accounts with optional 2FA, session management, and audit logs
- Obsidian-Style Workspace: Curation interface for organizing and reviewing shared knowledge
- Low-Power Hardware: Runs efficiently on Raspberry Pi 5, reducing cost and energy footprint
Who Is It For?
- Development teams seeking private, on-premise memory for AI coding assistants
- Research teams needing shared knowledge bases accessible to AI agents
- Privacy-conscious organizations that cannot use cloud-based AI memory services
- Technical teams running AI workflows on Raspberry Pi or low-power hardware
Frequently Asked Questions
- What hardware is required to run piLoci?
- piLoci is designed for Raspberry Pi 5 with at least 4GB RAM. It also runs on standard Linux servers. The alpha version requires approximately 2GB storage for the SQLite auth database and LanceDB vectors.
- How does piLoci differ from cloud-based AI memory services?
- piLoci keeps all data on your own hardware, eliminating subscription costs and data privacy concerns. It provides the same MCP interface that cloud services use, but stores memories locally on a Raspberry Pi or Linux server you control.
- Can piLoci integrate with Claude Desktop and Codex?
- Yes, piLoci exposes a standard MCP surface with tools for memory, recall, listProjects, and whoAmI. It has been tested with Claude and Codex according to the project documentation.
Related
Related
1 Indexed items