Audrey

Overview

Audrey is a local-first memory firewall for AI agents that provides durable memory continuity. It acts as a SQLite-backed memory layer that agents can consult before taking actions. The server offers 20+ MCP tools for memory operations like preflight checks, recall, encoding, and validation. It integrates with Claude Code, Claude Desktop, Cursor, Windsurf, VS Code, JetBrains, Ollama-backed agents and more. Audrey implements a comprehensive memory model including episodic, semantic, and procedural memory with decay, interference handling, and contradiction management.

Try asking AI

After installing, here are 5 things you can ask your AI assistant:

• memory_preflight

  Check prior failures, risks, rules, and relevant procedures before an action

  Try:Check if there are any known issues with running npm deployTry:Review previous failures before executing this database query
• memory_recall

  Find durable context by semantic similarity

  Try:Recall information about Stripe rate limitsTry:Find previous solutions for authentication errors
• memory_reflexes

  Get trigger-response guidance agents can follow based on remembered evidence

  Try:What should I do when encountering a 404 error?Try:Show me the reflexes for database connection failures
• memory_validate

  Validate whether memory was helpful after an action

  Try:Was the preflight advice useful for this deployment?Try:Rate how helpful the memory recall was for solving this problem
• memory_dream

  Consolidate episodes into principles and apply memory decay

  Try:Consolidate recent memories into principlesTry:Apply memory decay to old information
• encode

  Store a useful observation in memory

  Try:Remember that Stripe returns HTTP 429 above 100 req/sTry:Store this information about our deployment process
• capsule

  Get a turn-sized memory packet

  Try:Get the relevant memory for this conversationTry:Provide context for the current task
• guard

  Check memory before executing a tool

  Try:Guard this bash command against past failuresTry:Check if there are any warnings before running this deployment
• promote

  Promote memories to higher importance

  Try:Promote this solution to a top-level principleTry:Increase the importance of this memory
• impact

  Assess whether the memory system is performing effectively

  Try:How effective has Audrey been at preventing repeated mistakes?Try:Show me the impact metrics for the memory system
• observe-tool

  Record and analyze tool usage patterns

  Try:Record this tool usage for future analysisTry:Observe the patterns in this command execution
• reembed

  Re-embed memories with updated models

  Try:Update embeddings for all memories with new modelTry:Re-process all memories with current embedding model

Comparable tools

Installation

# Install via npm
npm install -g audrey

# Basic setup
npx audrey doctor

# Install for Claude Code
npx audrey install --host claude-code

# Verify installation
claude mcp list

{
  "mcpServers": {
    "audrey": {
      "command": "npx",
      "args": ["audrey", "mcp-config", "generic"]
    }
  }
}

FAQ

What memory models does Audrey support?

Audrey supports episodic memory (specific observations), semantic memory (consolidated principles), and procedural memory (remembered ways to act), along with affect/salience, interference/decay, and contradiction handling.

Can Audrey be used with non-Claude AI agents?

Yes, Audrey can be used with any AI agent that supports MCP or its REST API, including Claude Code, Claude Desktop, Cursor, Windsurf, VS Code, JetBrains, Ollama-backed agents, and custom agent services.

On Hacker News

Recent discussion from the developer community.

• Audrey: Local-first memory guard for AI agents (source)▲ 2💬 1
  Story by evilanders · 2026-05-13

Compare Audrey with