Audrey

Name: Audrey
Author: Evilander

@Evilander

Repository Homepage

About Audrey

Persistent memory and continuity engine for Claude Code and AI agents.

#memory #ai-agents #mcp #developer-tools #preflight #local-first #sqlite #agent-safety

Basic information

Config

Add this server to your MCP-compatible client using the configuration below.

{
  "mcpServers": {
    "audrey-memory": {
      "command": "npx",
      "args": [
        "-y",
        "audrey",
        "mcp"
      ],
      "env": {
        "AUDREY_EMBEDDING_PROVIDER": "mock"
      }
    }
  }
}

Tools

No tools detected

We auto-extract tools from the README. The maintainer can list them under a ## Tools heading to populate this section.

Overview

What is Audrey?

Audrey is a local-first memory firewall for AI agents. It provides a durable memory layer that agents check before touching tools, so they avoid repeating past mistakes, forgotten rules, and failed commands. All memory lives in a single local SQLite file—no cloud, no account, no code leaves your computer.

How to use Audrey?

Install via npm (requires Node.js 20+). Run npx audrey doctor to verify your setup, then npx audrey demo --scenario repeated-failure to see memory-before-action in action. Use npx audrey guard --tool Bash "npm run deploy" from the terminal, or install into host agents with npx audrey install --host codex (or claude-code, vscode, generic). For programmatic use, import Audrey from the audrey package in JavaScript, or install the Python client (pip install audrey-memory) and connect to the REST sidecar.

Key features of Audrey

Local‑first, no cloud or account required.
SQLite‑backed memory with semantic recall via sqlite‑vec.
Preflight checks (allow/warn/block) before tool execution.
Supports Codex, Claude Code, Cursor, Ollama, and custom agents.
MCP stdio server, REST API, CLI, and JavaScript/Python SDKs.
Memory model with episodic, semantic, procedural, decay, and contradiction handling.

Use cases of Audrey

Preventing AI coding assistants from repeating the same failed commands.
Remembering project‑specific rules across sessions.
Blocking dangerous actions based on past failures.
Providing durable memory for local or sidecar agent loops.
Enforcing team guidelines without storing data in the cloud.

FAQ from Audrey

What infrastructure does Audrey require?

Audrey requires Node.js 20+. It uses a local SQLite database; no hosted database or cloud account is needed.

How does Audrey ensure isolation between agents?

Set a distinct AUDREY_DATA_DIR per tenant, agent identity, or concurrent host. Isolation is a hard requirement for multi‑agent setups.

Can Audrey work with Ollama and local models?

Yes. Start Audrey as a REST sidecar with AUDREY_AGENT=ollama-local-agent npx audrey serve and use routes like /v1/preflight from the agent loop.

How does Audrey handle memory conflicts?

Contradiction handling tracks competing claims instead of silently overwriting. Stale or low‑confidence memories decay over time.

What transports or authentication does Audrey support?

Audrey supports MCP stdio for local agents and a REST API with optional Bearer token authentication via AUDREY_API_KEY. For non‑loopback binding, an API key is required.