Smriti

I built Smriti out of frustration with how brittle context becomes during serious multi-model work. You can reach clarity, switch tools or directions, and suddenly spend half an hour reconstructing your own thinking.

The project treats reasoning state as something explicit and restorable rather than something you hope a long chat history still conveys correctly.

• Checkpointing decisions, assumptions, open questions, and artifacts as structured state.
• Restoring a clean reasoning context without leakage from later turns.
• Branching from checkpoints to compare how different lines of reasoning diverge.
• Treating the model as a renderer of reasoning state rather than the owner of that state.

The more I work with agents and long-horizon problem solving, the more it feels like state is a fundamental systems problem rather than just a prompt problem.