Give Claude Code or Codex persistent memory of past conversations and project decisions.
Build a searchable knowledge base of verified notes across multiple codebases.
Automatically capture and distill AI coding session transcripts into reusable project knowledge.
| venomd846/holocore | 0marildo/imago | 0xdfi/glm-5.2-1m-4x-dgx-spark | |
|---|---|---|---|
| Stars | 3 | 3 | 3 |
| Language | Python | Python | Python |
| Setup difficulty | moderate | easy | hard |
| Complexity | 4/5 | 2/5 | 5/5 |
| Audience | developer | general | researcher |
Figures from each repo's GitHub metadata at analysis time.
Installed via uv tool install, requires choosing a shared Home folder and trusting hooks in Claude Code or Codex.
HoloCore Context Engine is a local, offline AI knowledge and memory system meant to sit alongside AI coding assistants like Claude Code and Codex, giving them useful context about your projects without sending anything to an outside server. It combines three separate parts: a library of verified notes and decisions called the Archive, a structural map of your project's files, functions, and dependencies called the Atlas, and a memory of past work, conversations, and errors called Animus. Rather than tracking one project at a time, HoloCore is built around a single shared Home folder that holds one Obsidian compatible knowledge vault for everything. Each individual project becomes a World inside that vault, with its own section of notes plus its own privately generated data. When you ask HoloCore a question, it runs a fixed sequence: first it checks the project's structural map to narrow down what is relevant, then it searches the verified notes for that project along with anything explicitly marked as shared, and only pulls in memory of past conversations when the question is actually about earlier work or past errors. The stated goal is reducing how much unrelated context gets sent to the AI model, saving on the number of tokens used. Once installed, HoloCore automatically captures conversations from supported AI coding tools at the end of a session, then distills useful facts, decisions, and preferences out of that transcript and files them into the project's Archive automatically, avoiding duplicate entries along the way. Installation happens through a command line tool, after which a command line interface and a local browser based Console are both available for searching, browsing memory, and reviewing exactly where each piece of information is stored on disk. It also runs a MCP server so AI clients like Claude Code, Codex, Gemini, Cursor, and OpenCode can connect to it directly. This project is aimed at developers who work across multiple codebases with AI coding assistants and want those assistants to retain useful project context and history between sessions, without relying on a cloud based memory service.
A local-first AI context engine that gives coding assistants like Claude Code and Codex a structural map, verified notes, and memory of past sessions across all your projects.
Mainly Python. The stack also includes Python, SQLite, MCP.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.