API Set-Up

One place for your API keys. Set once — every tool page uses them.
API keys
Keys are stored only in this browser's localStorage on this device. They never appear in the page source and are sent only to the provider of the model you pick — OpenAI for gpt- models, Anthropic for claude- models.
Get one at platform.openai.com.
Get one at console.anthropic.com.
Checking…
Models & rough cost
USD per 1M tokens. Actual spend depends on prompt and output length.
ModelProviderInputOutputNotes / best for
gpt-5.4-miniOpenAI$0.75$4.50Current default. Cheapest, fast. The phrase doctrine was tuned against this model at temperature 0.5.
gpt-5.5OpenAI$5.00$30.00Flagship reasoning. Slower and ~7× the cost. Ignores custom temperature.
claude-haikuAnthropic$1.00$5.00Best Claude pick for most work. Closest cost to the current default, and the only Claude model that still honours a custom temperature — so it preserves the low-temperature tuning used for generation (0.5), prioritising (0.3) and conform (0). Good for high-volume batches.
claude-sonnetAnthropic$3.00$15.00Balanced reasoning model. Worth trying for phrase generation when you want more care. Thinks before answering, so it is slower; temperature is ignored (steer via the prompt).
claude-opusAnthropic$5.00$25.00Highest-capability, roughly 8× the cost per build. Best kept for spot checks rather than routine use. Also a reasoning model: slower, no custom temperature.
Why cost varies so much: the phrase doctrine sent with every generation is about 18,000 tokens, so input price dominates — output is a rounding error. That is why a flagship model costs several times more per build than a small one.
Not yet measured: these notes reflect cost, speed and temperature support. Which model best follows the phrase doctrine (identity preservation, no outcome language) has not been benchmarked for Claude yet — treat model choice as provisional until it is.
Running session cost: $0.00000 Reset API cost calculation
This total is shared across all pages and is an estimate only.
Note: claude-sonnet and claude-opus are reasoning models — they think before answering, so they are slower and do not accept a custom temperature. The Classifier's Validate step is always pinned to an OpenAI model and needs an OpenAI key.
Clear all stored data
Erases everything this site has stored in this browser on this device: both API keys, the selected model, accumulated phrases, cached prompts, and handoff data. Permanent.

For personal exploration. Not a medical device and not intended to diagnose, treat, cure, or prevent any condition.

Phrase Player · Phrase Prioritizer · AI Phrase Generator

Session Cost: $0.00000 Reset

Last updated: —