Stopping an agent from remembering the same thing twice
A memory that grows by appending is a memory that rots. Six months in, the same decision is recorded four times in four wordings, two of them out of date, and nothing says which one to believe.
Pando answers a new memory with similar, the older ones that say close to the same thing, so the agent folds its note into the one that is already there instead of adding a fourth.
Set it up
- Point the client at https://pando.ink/mcp, which speaks streamable HTTP.
- Authenticate with OAuth 2.1 and dynamic client registration, or paste a key from Agents and API keys as a Bearer token.
The check is in the write, not in a later tidy-up
The answer to a remember call carries similar when older memories are close to it, and it explains what to do about them. A cleanup pass that runs afterwards has to guess which of four notes was meant to win. The write knows, because it is holding the new one.
What to tell the agent, once
- When the answer says similar, update the older memory rather than keeping both.
- Change part of a line with appendText or replace, never by retyping the whole line, because a rewrite is where the half you did not mean to change gets lost.
- A fact that belongs in two places is one bullet mirrored into both. Editing either changes both, because there is only one of it.
Why a mirror beats a copy
A copied bullet is two facts that agree today. A mirrored bullet is one fact in two places, and it cannot drift, because there is nothing to drift from. That is the whole difference, and it is the reason the tool exists.
Deeper
Connect an agent, about two minutes · The twelve tools · Which note apps an agent can reach · Who runs this