Give Claude Code a persistent identity and memory that carries over between coding sessions.
Let Claude Code learn your preferred level of detail and when to push back on decisions.
Track what approaches worked or failed on a project so future sessions build on that history.
| domdemetz/claude-soul | deepelementlab/jupyter-studio | eric248550/comcom | |
|---|---|---|---|
| Stars | 49 | 49 | 49 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | easy | easy | hard |
| Complexity | 2/5 | 2/5 | 4/5 |
| Audience | developer | data | developer |
Figures from each repo's GitHub metadata at analysis time.
Requires Node.js 18 or newer and an active Claude Code Pro or Max subscription.
Claude Soul is an add-on for Claude Code, Anthropic's AI coding assistant, that gives it a kind of persistent memory and an evolving thinking style across sessions. By default, Claude Code starts fresh each time you open it, with no recollection of previous conversations, your preferences, or what worked and what did not. Claude Soul is built to change that, and it goes beyond simply storing notes. The idea is that instead of just remembering facts, such as that a user prefers concise answers, it develops what the project calls cognitive frameworks: patterns of reasoning and behavior that emerge from your actual usage. When you correct a mistake, when something works particularly well, or when you express confusion, the system extracts those signals automatically at the end of each session. Periodically, it runs a reflection process that tests those signals against existing frameworks, updates their confidence, retires frameworks that proved unhelpful, and discovers new ones that fit how you actually work. Over time the claim is that Claude becomes more calibrated to you specifically. It learns when to push back rather than just agree, how much detail you want for a given kind of question, and when to flag uncertainty rather than guess. The project describes three tiers of reflection, using progressively larger models, that run at different signal thresholds as the system matures. Everything runs locally on your machine, using Claude Code's official extension system of MCP tools and hooks, and nothing is sent to a cloud service. It is installed via a command-line tool and requires Node.js and an active Claude Code Pro or Max subscription. The project is written in TypeScript.
An add-on for Claude Code that builds persistent, evolving memory of how you work by learning from corrections and successes across sessions, all stored locally.
Mainly TypeScript. The stack also includes TypeScript, Node.js, MCP.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.