Smriti

Git for reasoning state across coding agents

Open Source

Developer Tools

Artificial Intelligence

GitHub

Visit WebsiteSee on Product HuntGithub ⧉

Hi Product Hunt, I’m Himanshu, and I’m launching Smriti early because I want this idea shaped by real agentic workflows, not only by my own usage. Smriti started as a different product. The first version was a chat app around "Git for reasoning state" - checkpointing, branching, comparing and restoring reasoning. It worked conceptually, but I realized the UX was wrong. Most people do not want a version-control system just to ask a question or do lightweight research. The idea became much sharper when I started using coding agents seriously. When Claude Code, Codex, Cursor or other tools work on the same project, Git versions the code, but the reasoning behind the work is still trapped inside individual sessions. If I switch tools because one model gets better, I have to re-explain the project. If two agents work in parallel, they can duplicate work or miss what the other one already decided. Markdown handoff files help, but they are not a reliable state layer. That is the problem Smriti is trying to solve. Smriti is a local-first, open-source reasoning-state layer for coding agents. It lets agents: - attach to the same project - read shared state at session start - claim work before starting - checkpoint decisions and open tasks - detect repo drift - continue from another agent’s checkpoint - coordinate without a central orchestrator The project itself was built using Smriti, with Claude Code and Codex coordinating through the same state. I want to be transparent: this is not a polished SaaS product yet. It works, and the idea is real, but I have mainly tested it with Claude Code and Codex on macOS. I’m launching it now because the agentic development space is moving fast, and I would rather learn from real workflows than keep polishing it in isolation. What I’d love from this launch: 1. Try it on a real coding-agent project. 2. Tell me where the setup breaks. 3. Tell me if the “Git for reasoning state” framing makes sense. 4. Tell me which agent tools or workflows it should support next. 5. If it solves a real pain for you, consider starring the repo, opening issues, or contributing integrations. My belief is simple: If software development becomes increasingly agentic, reasoning state cannot remain trapped inside individual chat sessions. I’d love honest feedback, especially critical feedback.