---
post_type: "project"
title: "Agent Mind"
blurb: "A persistent, shared memory for AI coding agents — so every session with Claude Code, Cursor, or Codex starts with context instead of cold."
org: unbody-lab
tags:
    - AI agents
    - memory systems
    - developer tools
    - MCP
    - HAC
    - open source
year:
    - 2026
current: true
stack:
    - TypeScript
    - Node.js
    - Adapt runtime
    - MCP
---
Agent Mind is the first product built on the Adapt runtime. The problem: every coding agent session starts from zero — no memory of what was decided last week, how you like to work, what's already been tried and failed. Agent Mind watches your agent sessions, ingests them, and distills them into a durable brain: decisions, preferences, project context, constraints, failed approaches, successful fixes, open questions.

It ships as a standalone repo with a CLI (`agent-mind start | status | ask | recent | stop | mcp`). The CLI drives a long-running process: source plugins watch session transcripts, an ingestion pipeline compacts them into memory chunks, and an LLM-backed brain organises them by subject — preserving how decisions changed over time rather than flattening them. It auto-registers an MCP server and a skill into Claude Code so agents can read and write memory directly.

The relationship to Adapt: the runtime provides the generic memory engine; Agent Mind is the productised user-facing layer — handling the CLI, onboarding, observability, and the agent-facing MCP surface. It's the answer to the same question I've always worked on, one layer further in: not how humans interact with software, but how agents do.