Getting Started

Prerequisites

• .NET 10 SDK — Download
• RavenDB 7.x — either:
  • Embedded mode (zero setup, Eidet manages it)
  • External instance at http://localhost:8080 (Download)
• Ollama (optional) — for AI-powered enrichment (Download)

Installation

From GitHub Releases

Download the latest release for your platform from GitHub Releases:

Platform	Asset
Windows x64	eidet-win-x64.zip
macOS ARM	eidet-osx-arm64.tar.gz
Linux x64	eidet-linux-x64.tar.gz

Extract and add to your PATH.

From Source

git clone https://github.com/stevehansen/eidet.git
cd eidet
dotnet build
dotnet run --project src/Eidet.Service

First-Time Setup

Run the interactive setup wizard:

eidet setup

This will:

1. Detect or start RavenDB (embedded or external)
2. Create the Eidet database
3. Deploy search indexes (vector + full-text)
4. Configure bge-micro-v2 embeddings
5. Optionally detect and configure Ollama

For non-interactive setup (CI/Docker):

eidet setup --non-interactive

Embedded vs External RavenDB

Mode	When to use	Setup
Embedded	Zero-setup, single user	eidet setup --embedded
External	Already running RavenDB, need Studio access	Point to http://localhost:8080

Start the Service

eidet serve

Output:

Eidet v1.0.0
  RavenDB: Connected at http://localhost:8080
  API:     http://127.0.0.1:19380
  MCP:     http://127.0.0.1:19380/mcp
  Health:  http://127.0.0.1:19380/api/health

Install as System Service

To run Eidet on startup and register with your AI clients in one shot:

eidet install

This installs as a Windows scheduled task / macOS launchd agent / Linux systemd user unit, then walks the MCP client registry and registers eidet with every detected client (Claude Code, Claude Desktop, Codex, Gemini). Already-configured clients are left alone — re-running is safe.

Store Your First Memory

Via CLI

eidet store "The auth module uses JWT with RS256 signing" --type observation --repo /path/to/project
eidet store "Always run db:migrate before db:seed" --type heuristic --tags database,setup
eidet store "Deploy: push to main, wait for CI, tag release" --type procedure --tags deployment

Via REST API

curl -X POST http://localhost:19380/api/eidet \
  -H "Content-Type: application/json" \
  -d '{
    "repo": "/path/to/project",
    "content": "The auth module uses JWT with RS256 signing",
    "type": "observation",
    "tags": ["auth", "jwt"],
    "importance": 0.7
  }'

Via MCP (from an AI agent)

The MCP server exposes eidet_store as a tool. AI agents call it automatically when they discover something worth remembering.

Recall Memories

# Search by query
eidet recall "how does authentication work" --repo /path/to/project

# Get compact context (what agents see at session start)
curl "http://localhost:19380/api/eidet/context?repo=/path/to/project"

The context endpoint returns a compact summary (under 600 tokens) that agents load at session start — no manual configuration needed.

Connect to AI Agents

The fastest path is eidet install — it registers eidet with every MCP-capable client it finds on your machine. To work one client at a time, use eidet mcp install:

eidet mcp install               # interactive picker over detected clients
eidet mcp install claude-code   # specific client
eidet mcp install --all         # everything detected
eidet mcp list                  # see current registration status

Supported clients:

Client	How it’s registered	Config file (fallback)
claude-code	claude mcp add --transport stdio -s user eidet -- eidet mcp	~/.claude.json
claude-desktop	direct file write (no CLI)	claude_desktop_config.json (platform-specific)
codex	codex mcp add eidet -- eidet mcp	~/.codex/config.toml
gemini	gemini mcp add eidet "eidet mcp" -s user	— (CLI only)

When the client’s own CLI is on PATH, eidet rides on it so you get whatever schema upstream is using; otherwise it writes the config file directly.

Any other MCP client

Eidet supports both stdio and HTTP MCP transports:

• stdio: point your client at the command eidet mcp
• HTTP: POST http://localhost:19380/mcp

REST API

Any HTTP client can use the REST API at http://localhost:19380/api/. See the API Reference for all endpoints.

Verify Everything Works

# Health check
eidet doctor

# Quick status
eidet status

# Open Web UI
# Navigate to http://localhost:19380/ui

Next Steps

• Concepts — understand memory types, scoring, and retrieval
• Configuration — tune settings for your workflow
• Hooks — run custom scripts on memory lifecycle events