Paste a list of phrases you're considering. The AI scores each phrase against your current state — you pick the top ones to play.
Score your phrases
Describe yourself · Find areas to dig into or score your own phrases · Send selected to Phrase Player
Beta · experimental. The AI's read is exploratory — useful as a starting point, not as a diagnosis.
Set your OpenAI API key first. Phrase Prioritizer shares the key you set on the
AI Phrase Generator page. Open it, scroll to API Settings, save your key, then come back here.
Type a few sentences
Write whatever's on your mind — what you're noticing, what you've been doing, or anything at all. The words can feed the AI as your description, and your typing rhythm is captured as a signal while you write. The content matters less than simply typing naturally for a bit.
Tip: When Your Sentences is checked, the AI reads the text and weighs it heavily. When it's off, only the typing-rhythm signal is used.
Just type naturally for a few lines. Your words (when Your Sentences is on) and your typing rhythm are signals which will inform your phrase scoring.
Independent input channels — check any combination to control what the AI uses to score your phrases:
Keyboard — your typing-rhythm signals (keys/sec, backspace rate, gaps, jitter) captured from the description box. Your Sentences — the written note about your current state. Mouse — pointer-movement signals (speed, jitter, pauses). Voice — vocal dynamics from a short recording (pitch movement, energy, pace, silence). Captures how your voice moves, not what you say — no transcription. Click "Record voice", speak, click Stop (auto-stops at 10s).
Each channel is independent. With Your Sentences off, scoring leans on signals alone — experimental and noisy; useful for testing whether signals carry usable info on their own.
Pick which channels feed the AI. Each is independent.
Find areas to dig into experimental - Testing — not proven
Not sure which phrases to consider yet? This scores whole areas of the library against your current state, so you can see where to focus before picking individual phrases.
Click Select Areas to Score to open the picker. Choose a level using the tabs — Type, Family, Focus, or Tag — then check the specific areas you want, or leave none checked to score every area at that level. It's one level per run: switching tabs starts a fresh selection. Your choice shows as a summary on the page (e.g. "Family · 6 selected"), and Reset clears it.
Hit Score areas and each area is scored by sampling the phrases inside it (up to 15 per area), scoring that sample with the AI, and rolling up to a single area score (the average of its top 5 phrases). Areas come back ranked. On any area, ← Load phrases pulls all of its phrases into the "Phrases to score" box (to the left), where you can score them individually as usual; or check several areas and use ← Load selected.
This uses one AI call per area and reads your description + signals the same way phrase scoring does. Scoring many areas (e.g. all tags) means many calls — you'll get a warning first. Experimental: results are exploratory, not a diagnosis.
Score whole areas of the library against your state to decide where to focus. Loads from preset.tsv.
Paste a list of phrases you're already considering — could be from your saved configurations, an AI Generator session, a custom TSV, or hand-typed. You can also fill this box from Find areas to dig into (right) using ← Load phrases. The AI will score each one against your description and the inputs you've selected. One phrase per line. Empty lines are ignored. Duplicates (case-insensitive) are dropped silently and a note appears under results. Special characters are stripped on paste.
Paste the phrases you're considering, one per line. Empty lines and duplicates are ignored.
Each click of "Score these phrases" pushes a new run card here. Each run preserves its mode, description, captured signals, full results table, threshold, and which phrases you selected. Use this to compare A/B/C runs against each other — see if signals actually move the ranking, see how the same input scores under different modes, copy a structured data block back to chat when you want to share with the AI.
Newest run auto-expands; older runs collapse. Click any header to toggle. Runs persist in memory only — they clear on page refresh or via "Reset all runs."
Each run keeps its signals, mode, and results. Most recent run is expanded; click an older header to expand it.
For personal exploration. Not a medical device and not intended to diagnose, treat, cure, or prevent any condition.