The company look-up (done by itself when an employer or business name is entered) and the LinkedIn search ask Claude to search the web. They need an Anthropic API key. The key is kept only in this browser on this computer. Only the company name is sent, and for the LinkedIn search the client's name, employer and occupation.
Enter the date shown as "Expires" next to the key in the Anthropic console. The top bar then counts down the days. Open the console ↗
Used only for the company look-up and the LinkedIn search. Each one reads several web pages, so this is the dearest thing the form does: Sonnet costs half of Opus, and Haiku a small fraction. The company look-up runs by itself when an employer or business name is entered; the LinkedIn search runs only when you tap its button. The running cost is in the top bar.
Voice mode asks this model what was said and which box each fact belongs in. Only the last few sentences are sent, as text. ID numbers, phone numbers and email addresses are never sent: the model sees a marker such as [N1] and says which box it belongs in, and the real number stays on this computer.
Type the date this OpenAI key stops working, or the date you plan to replace it. The top bar then counts down the days, next to the Anthropic key's. Open the OpenAI keys page ↗
Used only for the cost shown in the top bar. OpenAI lists about US$0.006 a minute for gpt-4o-transcribe and whisper-1 and about US$0.003 for gpt-4o-mini-transcribe, but it really charges by audio token, so this is an estimate. Type the price from your OpenAI usage page here to replace it. Leave it empty to use the listed price.
The microphone hears only you when you wear a headset: the client's voice comes out of the headset, not out of the room. With Yes, pressing Listen opens Chrome's share window. Choose Entire screen and tick Also share system audio (the Teams app), or choose the Teams tab and tick Also share tab audio (Teams in Chrome). The form then hears you through the microphone and the client through the call's sound, and so knows who said what. Only the sound is used: the picture is thrown away and nothing is recorded. The call's sound goes to OpenAI together with yours, so the same care applies. Wear a headset: with loudspeakers the microphone hears the client as well.
With the OpenAI speech source each spoken line is cut out on this computer and sent, as sound, to OpenAI for transcription, together with the names on the form and the terms of the trade so that names are heard properly. The sound goes to OpenAI and not to Anthropic; Claude only ever reads the words. Use it only when your client terms allow a speech service to hear the call. The key is kept only in this browser. Transcription costs a few cents a meeting; the OpenAI account shows the real figure. Make the key under API keys in the OpenAI console, and keep a few dollars of credit on the account (Billing). Open the OpenAI console ↗
The Listen button in the top bar uses the microphone. Whisper keeps the sound on this computer: the speech model is downloaded once (about 40 MB for Tiny) and kept in this browser. Chrome's own recognition is instant, but Google hears the audio, so use it for practice talk only, never with a client. A change here applies the next time Listen is pressed.
Keep the keys safe. Both keys are saved in this browser, on this computer, as plain text, and this web address is shared with every page published under pavlos1972.github.io. Set a spending limit on each key in its console, and press Remove saved keys before leaving a shared or borrowed computer.
