samvallad33/vestige ? reverse-engineered prompt

Build me a local first memory server for AI agents that works over MCP and runs entirely on my machine. I want it to remember decisions, catch contradictions when a new claim conflicts with an old one, and be able to trace a failure back to the earlier memory that likely caused it, not just return similar text.

It should store everything in a SQLite file, work offline after setup, and feel like a single small Rust binary. Please include a simple dashboard I can open in the browser to see that it is running and to inspect what it has stored. I also want a backfill mode that starts from a failure and walks backward to the most relevant earlier memory, showing the contrast between what was believed then and what went wrong now.

If you need to check current MCP or agent integration details, look up the latest docs online first. Keep it privacy friendly, no cloud dependency, and make it easy to connect to tools like Cursor or Claude Code.