Eidet
Long-term memory for AI coding agents — local-first, privacy-absolute, works everywhere.
From "eidetic" — relating to extraordinarily vivid, detailed recall.
Fun coincidence: in Antwerp dialect, "èdde 't?" ("hebt ge het?" / "do you have it?") sounds a lot like "eidet" — fitting, for a memory system.
What Is This?
A persistent, semantic memory system that gives AI coding agents the ability to learn and remember across sessions. Built as a standalone local service that any MCP-compatible AI client can use.
Core philosophy: The local service is the product. It just works. Complexity is for the app, simplicity is for the user.
Quick Start
# Install (requires .NET 10 SDK)
dotnet tool install -g eidet
# Interactive setup — configure RavenDB (embedded or external), Ollama, embeddings
eidet setup
# Register as system service + auto-configure every detected MCP client
# (Claude Code — including the desktop app's Code tab — Codex, Gemini)
eidet install
# Verify everything is running, including which MCP clients picked up eidet
eidet status
After eidet install, the service autostarts at login and your AI clients can use Eidet's memory tools immediately. Need a specific client only? Use eidet mcp install <client> (claude-code, codex, gemini) or eidet mcp list to see registration status.
Key Properties
- Local-first: Fully functional offline. RavenDB (embedded or external) with built-in vector search. No cloud required.
- Universal: Works with Claude Code, Claude Desktop, Cursor, Windsurf, TerminalHost, or any MCP client.
- Typed memories: Observations, Insights, Procedures, Heuristics — each with distinct lifecycles and decay curves.
- Docker-like layers: Local (read-write) + Shared (read-only, team) + Base (read-only, from packs).
- < 600 token wake-up: L0 identity + L1 top-K context at session start. Minimal overhead.
- Hybrid search: Vector + full-text + metadata in a single round-trip.
- Self-improving: Echo/fizzle feedback loop tunes recall quality over time.
- Privacy-absolute: Secret scanning gate, localhost-only API, no data leaves the machine.
- Always running: System service (scheduled task / launchd / systemd) with managed updates.
- Rich TUI: Interactive setup, connection testing, troubleshooting — built for developers AND AI agents.
Updates & Feedback
# Check for updates
eidet update --check
# Update (restarts the service; open AI sessions keep working and switch on their next restart)
eidet update
# Report an issue (opens GitHub with version pre-filled)
eidet feedback
Distribution
MVP Scope
The initial release is local-only:
- Local RavenDB (embedded or connect to existing instance)
- Local optional Ollama enrichment (with CoT reasoning stripping)
- MCP server (stdio for AI clients, HTTP for network)
- REST API (for tools and custom integrations)
- Rich TUI for setup, configuration, troubleshooting
- Docker container integration (devcontainers, custom containers)
- Full 13-tool MCP surface
- System service with always-on scheduling (maintenance, consolidation, enrichment)
- Local Web UI at
http://localhost:19380/ui(dashboard, memory browser, knowledge graph) - Client SDKs (TypeScript, Python, C#)
Team sync, remote backup, and collaboration features are designed but deferred to a future release. See SyncSpec for the planned architecture.
Documentation
Project Structure
eidet/
├── src/
│ ├── Eidet.Core/ # Core library (domain, services, indexes)
│ ├── Eidet.Service/ # System service (MCP, REST, scheduler)
│ └── Eidet.Sync/ # Sync adapters (future)
├── sdk/
│ ├── typescript/ # @eidet/sdk
│ ├── python/ # eidet-sdk
│ └── dotnet/Eidet.Sdk/ # Eidet.Sdk
├── tests/
│ ├── Eidet.Core.Tests/
│ ├── Eidet.Service.Tests/
│ └── Eidet.Integration.Tests/
└── docs/
└── specs/