Logo of Pivvi

Pivvi

Save AI work and route to free models with available capacity

User reviews

Log in to leave a review

Genuinely useful problem to pick. The part I'd want to understand is the handoff itself: when a chat moves to a different model, the accumulated context often exceeds what the next one will accept, and each model treats system instructions and formatting differently. Does it truncate, summarise, or refuse when the target window is smaller than what has built up? That decision is close to invisible to the user, but it decides whether the next model actually picks up mid-task or quietly loses the thread.

View
pivvi-3b1f's avatar
pivvi-3b1fAuthorAug 27, 2026

You’ve gone straight to the hardest part of it.

Right now the Rescue Prompt does [TRUNCATE / SUMMARISE / REFUSE — fill in]. [One line on how it decides.]

Where I know it’s weak: [fill in — e.g. long transcripts, code-heavy context, anything where the middle matters as much as the ends].

The part I keep coming back to is that summarising costs a model call, which is awkward for a product whose whole premise is not spending anything. So there’s a real tension between a faithful handoff and a free one.

How have you seen this handled elsewhere?