MindMeld

Projects span tools, sessions, and models — and the reasoning behind choices quietly disappears. You're left with a decision but no trail: not the alternatives considered, not who pushed back, not what changed the outcome. The next AI you open knows nothing about why the last one decided what it did.

MindMeld is a user-owned reasoning archive. It stores decisions as structured records — each capturing the choice made, the alternatives rejected, any dissent, the provenance (who or what influenced the outcome), and enough context to reconstruct the thinking later. Decisions live in a local archive that no model owns and no session can delete.

The archive was queried live from four independent AI systems — Claude Code, Codex, ChatGPT Workspace, and Claude Workspace — each able to retrieve prior decisions and reconstruct reasoning without access to the original conversations. That cross-surface query capability is the key proof: the archive lives outside any single model, and any sufficiently capable AI can read from it.

Most systems capture what was decided. This one captures why — so reasoning can survive model changes, tool switches, and time. The archive preserves deliberation, not just outcomes: the alternatives that were seriously considered and rejected are often more useful than the choice that won.

The prototype runs as a local MCP server backed by a structured archive. I designed the record schema (decision, alternatives, dissent, provenance, context), the query interface, and the cross-surface registration approach that lets different AI tools connect to the same store. Next: ingestion from prior conversations, provenance signing, and artifact reconstruction across model boundaries.