Grit
Git for LLM reasoning. Best use of Cerebras API.
[ Visit project โ ]Summary
"Git for LLM reasoning" โ a live, branching visualization of an AI's chain-of-thought that lets you fork off any reasoning step into its own side conversation.
Problem you're solving
AI chat interfaces hide reasoning behind a single linear answer, training people to consume conclusions passively instead of following โ or challenging โ the logic behind them.
Target user
Educators and students who want to use AI as a teaching tool for reasoning rather than an answer machine, with a path into Educaide AI's ~700,000 educators.
Competitors
ChatGPT and Claude's built-in "thinking" panels are visible but strictly linear and read-only, with no way to branch, question, or replay a specific step.
Insight
Reasoning is a graph, not a transcript. If you can click any step and ask "wait, why that?" without losing the original thread, the reasoning becomes something you actually learn from.
Solution
Streams an LLM's chain-of-thought into an interactive D3 graph in real time; clicking a node opens a persistent side conversation scoped to that step, with multimodal rendering for code, math, and diagrams, powered by Cerebras for fast inference.
- Next.js
- TypeScript
- D3.js
- WebSockets
- Node.js
- Supabase
- Cerebras
Distribution
Built for and pitched at a hackathon as a path into Educaide AI's educator platform โ winning Best Use of Cerebras API is the current distribution proof point ahead of that integration.
Adapting to user feedback
Pilot users found early graphs unreadable once a conversation had more than a dozen branches, so branches were given persistent context and the layout was reworked to keep the main thread visually anchored.