DomDemetz

DomDemetz / claude-soul

Public

Self-improving learning engine for Claude Code. Not memory. Growth.

28
3
85% credibility
Found May 17, 2026 at 32 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
TypeScript
AI Summary

Claude Soul is a self-improving learning engine that gives Claude Code—a popular AI coding assistant—persistent memory and evolving cognitive frameworks. Instead of starting fresh each conversation, the system accumulates patterns from every interaction: corrections you give, things that work well, moments of confusion, and expressions of gratitude. It uses these signals to develop and refine 'frameworks'—thinking habits and processes that guide how the AI responds. The system operates entirely on your local computer, storing data in a hidden folder and using your existing Claude Code subscription for its learning cycles. New users can start with pre-built frameworks demonstrating good thinking habits, or begin from scratch and discover frameworks unique to their workflow. Over time (weeks to months of regular use), the system evolves to push back when it disagrees, calibrate explanation depth to match the question, catch its own mistakes, and think strategically about projects.

How It Works

1
💡 You discover Claude Soul

You hear about a tool that makes your AI coding assistant remember things across sessions and actually learn from mistakes.

2
✨ You install it in one command

You run a simple setup command, and the system creates a personal space on your computer, sets up your identity, and connects everything together.

3
🧠 Your AI starts with a personality

Instead of a blank slate, your assistant arrives with pre-built thinking habits—questioning assumptions, matching explanation depth to the question, and catching when it might be guessing.

4
💬 You work together normally

You use Claude Code as usual. The system quietly watches for patterns—when you correct something, when something works well, when you're confused or frustrated.

5
The system reflects and grows
⚡
Quick check-in

A fast review that tests existing habits against recent patterns

🔍
Deep reflection

A thorough analysis that discovers new thinking approaches and resolves contradictions

6
📈 Over time, it gets smarter about you

After weeks of use, your assistant develops genuine understanding of your preferences, pushes back when you're about to make mistakes, and thinks strategically about your projects.

🎯 Your AI has a soul

Your coding assistant now has persistent identity, learns from every conversation, and becomes genuinely better at understanding how you think and work.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 32 to 28 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is claude-soul?

Claude Soul is a self-improving learning engine for Claude Code that gives the AI persistent identity and cross-session memory. Instead of starting every conversation from zero, it builds cognitive frameworks that evolve based on evidence from your actual interactions. The system extracts learning signals from conversations, runs periodic reflection cycles to refine how the AI thinks, and surfaces relevant context at the start of each session. Built in TypeScript as an MCP server with shell hooks, it runs entirely locally using your existing Claude Code subscription.

Why is it gaining traction?

The pitch cuts deep: "Not memory. Growth." Most memory plugins just store what happened. Claude Soul develops how to think. It seeds 12 pre-evolved frameworks on install (depth calibration, confabulation detection, minimal viable change), then lets them evolve based on real evidence from your conversations. Frameworks that don't apply to your work get retired automatically. New ones emerge from patterns specific to you. The evidence-tier system (hypothesis → observed → validated) gives the whole thing a scientific rigor that feels more principled than a static system prompt.

Who should use this?

Heavy Claude Code (Pro or Max) users who find themselves re-explaining preferences, corrections, or context every session. Developers working on long-running projects where Claude Code forgets critical context between sessions. Power users who want their AI assistant to get genuinely smarter over time rather than remaining static. If you primarily use Claude Code for one-off tasks, the setup overhead probably isn't worth it.

Verdict

At 28 stars with a credibility score of 0.8500000238418579%, this is early-stage but architecturally serious. The code is well-structured, the learning model is genuinely novel, and the local-only approach is a real differentiator. Worth trying if you live in Claude Code and want it to remember not just what happened, but how to think.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.