“We already tried that”

Without Klio

Agents are confidently repetitive. Without a record of rejected approaches, every session is free to re-propose last month’s dead end, and you are the only thing standing in its way.

With Klio in the loop

1. 1

   When an approach is rejected, the agent records the rejection and the reason as a decision.

2. 2

   Weeks later, another agent starts down the same path.

3. 3

   Recall returns the rejection before code gets written, and the agent routes around it.

The call an agent makes

decide(
  {
    "content": "Do not use LISTEN/NOTIFY for cache invalidation.",
    "rationale": "Our pooled Postgres endpoint drops session-level features; it failed in staging on 2026-08-04."
  }
)

Real tool, illustrative payload. The full contract for every tool is in the docs.

What changes

Settled questions stay settled. You stop being the institutional memory for your own tools.