mnemina

Mnemina is a local-first context runtime for AI coding agents. It indexes a repository into a compact map of files, symbols, dependencies, tests, domains, flows, recent changes, and security boundaries, then exposes that context through a CLI and Model Context Protocol (MCP) server.

Your source code and index remain on your machine. Mnemina does not require a hosted service or send repository contents to a Mnemina backend.

Requirements

• Node.js 20 or newer
• Git for recent-change context
• A TypeScript or JavaScript project

Quick start

npm install --global mnemina
cd /path/to/your/project
mnemina init
mnemina index
mnemina doctor

You can also run commands without a global installation:

npx -y mnemina doctor

Connect an AI coding agent

Configure an MCP server named mnemina with:

{
  "command": "npx",
  "args": ["-y", "mnemina", "mcp"],
  "env": {
    "MNEMINA_PROJECT_ROOT": "/absolute/path/to/your/project"
  }
}

Then add an always-on agent instruction:

For every coding task, first call the Mnemina get_task_context tool with the
user's task. Use the returned map before reading broader repository files.

Common commands

mnemina init
mnemina index
mnemina index --watch
mnemina doctor
mnemina version --json
mnemina context "fix session expiration"
mnemina mindgraph serve
mnemina mcp

Run mnemina --help for the complete command list.

Local data and security

Mnemina writes its index and context-run history to .ai-cache/, which mnemina init adds to .gitignore. Agent-visible reads enforce built-in deny rules and .aiignore. Common credential stores such as .npmrc, .aws/credentials, and .ssh/ are denied. Secret redaction is enabled by default.

Review generated .ai/ configuration and .aiignore rules before using Mnemina with sensitive repositories.

Current language support

The initial release indexes TypeScript and JavaScript. Other languages are not yet supported.

License

Apache-2.0