Persistent Memory for AI Agents

A local SQLite memory layer for MCP tools and terminal workflows.

Echo lets agents store, recall, search, and refine context across sessions while keeping memory inspectable from the CLI.

Memory Infrastructure for AI Agents

  • Echo provides an external long-term memory layer for AI agents, overcoming the limitations of stateless sessions.
  • Instead of relying on repeated prompts, memory tools retrieve stored context so agents can process requests without the developer re-typing project instructions.
  • It supports two primary interfaces: an automated MCP server for AI tool calls and a terminal CLI for direct human curation of the memory state.
  • The integrated DuckDB analytical engine tracks knowledge ROI and resource impact so memory usage can be reviewed, refined, and kept high-signal.

Core Capabilities

🔗

Shared MCP & CLI Access

Agents and humans work with the same local memory state through MCP tools or terminal commands.

🧠

Persistent Store & Recall

Save instructions, facts, and artifacts, then recall them by context with benchmarked sub-millisecond local retrieval.

🔍

FTS5 Search & Curation

Find, update, and delete memories precisely with SQLite FTS5 search and ID-based governance.

📊

Usage Analytics

Use embedded DuckDB telemetry to review knowledge ROI, cost-efficiency, and environmental impact.

Driving Smarter, More Capable AI

  • Enhanced Context: AI agents can retrieve relevant context across extended interactions.
  • Continuous Learning: Knowledge gained is persistent, improving agent performance over time.
  • Reliable Persistence: Leverages SQLite for persistent, local data storage.
  • Developer Experience: Simple JSON-RPC interface for easy integration into agent architectures.
  • Local Control: Memory stays inspectable, editable, and portable through shared MCP and CLI workflows.