Generate, score against real data, critique, repeat — a measured search, not a grid guess.
Any backend that publishes a pipeline definition is optimizable — read-only.
PromptPotter treats your prompt as a program with tunable axes — and breeds, scores and critiques a population of candidates each round until your numbers move.
PromptPotter evolves your prompts and pipeline params with a critique-guided loop — generate, score against real data, critique, repeat — plus posterior elimination and self-healing rails. It runs until your numbers move.
Get early accessNo license keys, no seats, no paywall — the full source is public. Three steps stand between your dataset and a better prompt.
Drop in labelled examples — the input/output pairs every candidate prompt is scored against.
Point PromptPotter at the pipeline you want to improve. It reads the tunable params and never edits your code.
Watch the loop generate, score and critique — round after round — until your numbers move.
When the loop is running, a read-only console streams the live pipeline, the candidate being scored and every round’s verdict — no extra tooling, no setup.
PromptPotter isn't public yet. Leave your email and we'll bring you in when it opens — and tell us what you'd point it at, so we build the right thing first.
