ShellBrain
For Humans

Give your agents memory.

AI agents work in sessions. ShellBrain stores what they learn and recalls it in later sessions. Memory stays scoped to each repo.

What ShellBrain stores

Useful work, backed by evidence.

After each session, ShellBrain records problems, solutions, failed tactics, facts, preferences, and changes worth remembering.

When an agent runs shellbrain recall, ShellBrain returns a compact brief for the current task. Every memory links to session evidence.

Memory model

Hierarchical memory.

Episodes

The session trail: prompts, agent steps, tool calls, and outputs.

Evidence

Memories

Concrete records extracted from episodes: problems, solutions, failed tactics, facts, preferences, and changes.

Active Stale Superseded Wrong

Concepts

A graph of reusable ideas linked back to concrete memories and episode evidence.

Grounded Abstraction

Install

Install ShellBrain.

$ curl -L shellbrain.ai/install | bash

The installer

  • Installs the ShellBrain CLI
  • Adds it to your shell PATH
  • Installs the ShellBrain skill for Codex, Claude Code, and Cursor
  • Runs shellbrain init

Choose storage

On first setup, choose where ShellBrain stores data.

Managed local Docker, PostgreSQL, and pgvector on your machine
Your database An existing PostgreSQL database with pgvector