Skip to content

Latest commit

 

History

History
99 lines (65 loc) · 2.39 KB

File metadata and controls

99 lines (65 loc) · 2.39 KB

ShellBrain logo

ShellBrain

Long-term Memory for AI.

ShellBrain uses case-based reasoning and a self-managing concept graph as long-term memory for software engineering.

Install

curl -L shellbrain.ai/install | bash

Works for Codex, Claude Code, and Cursor.

Requirements.

  • macOS or Linux, Python 3.11+, Docker.

Upgrade for Latest Capabilities

shellbrain upgrade

Recall in One Command

ShellBrain recall uses vector search and BM25 to search your memories. An inner recall agent summarizes the search results.


Architecture

ShellBrain stores evidence and two forms of reusable knowledge:

  • Episodic Memory. It stores prompts, responses, and tool calls.
  • Case-Based Memory. Problems, solutions, and failed attempts are structured and stored.
  • Concept Graph. It connects claims, relations, and implementations.

Memories and concepts link directly to supporting evidence. The code is the source of truth.


How Agents Use ShellBrain

Recall

Working agents run shellbrain recall for long-term memory related to their current task.

Recall combines BM25, vector similarity, and explicit graph associations.

shellbrain recall "What is ShellBrain, and how does it help a working coding agent?"

Response:

{
  "status": "ok",
  "data": {
    "brief": {
      "memories": ["Relevant remembered context."],
      "code": ["README.md"]
    },
    "fallback_reason": null
  },
  "errors": []
}

Terminals show styled Memories and Code bullets. Pipes receive JSON. If no useful memory is found, both lists are empty. Provider failures return errors.

Recall Provider

For blazing fast recall for cheap, get an Inception API key and run:

mkdir -p ~/.shellbrain
printf '\nINCEPTION_API_KEY=%s\n' 'PASTE_YOUR_KEY_HERE' >> ~/.shellbrain/.env
chmod 600 ~/.shellbrain/.env
shellbrain admin recall provider inception

Switch back with shellbrain admin recall provider codex, or choose claude.


Docs