How do you find out why your AI agent "remembers" something?

Hi Product Hunt, I'm Rama, maker of Titen. We launch here on October 20.

The problem I keep hitting with agent memory isn't retrieval, it's the write side. Stores quietly fill up with duplicates, copies of the agent's own output, stale facts and the occasional pasted secret. One public audit of a production store found 97.8% of 10,134 entries were junk after 32 days. When I ran the same kind of audit on Titen's own store, it found 17.9% duplicates and 96.7% stale, so I'm not pointing fingers.

Titen is built around receipts: every memory keeps its source, who may read it, and the evidence that contradicts it. It runs with no API key, no LLM and no embedding provider (Bun + SQLite, or Cloudflare Workers + D1), and it's a drop-in for the MCP reference memory server: same nine tools, your existing memory.json imported on first run.

You can audit any store today, even one Titen doesn't own. It's read-only, no network, no upload:

npx titen-memory audit ./memory.json

Two questions for anyone running agents with memory:

1. How do you find out today why your agent believes something?

2. What would a memory layer need to show you before you'd trust it enough to switch?

Blunt feedback before launch day is very welcome. Repo: https://github.com/RamaAditya49/...