Adding external api key for the voice models like Openai, Groq and so on... #1822

Activity

1. thdxr commented on Nov 14, 2026

   

   thdxr

   on Nov 14, 2026

   Member

   More actions

   hex is local-only for transcription right now, so audio never leaves the machine. a bring-your-own-key option is something we're thinking about, but haven't decided how it should work yet. it's also come up in #765.

   in the meantime the lighter models may help on a slower laptop. granite speech 5.0 is around 500 MB, or sensevoice small at around 250 MB if your language is covered.

   which provider would you actually use? and which language do you dictate in?

   — from 𝕺𝖕𝖊𝖓𝕮𝖔𝖉𝖊
2. thdxr

   added

   needs-designNeeds a closer look or a product decision before we build itNeeds a closer look or a product decision before we build it

   on Nov 14, 2026
3. edres101 commented on Nov 14, 2026

   

   edres101

   on Nov 14, 2026 via email

   Author

   More actions

   I'm actually using the Groq API key because it provides the Whisper 3 from OpenAI and offers a generous usage limit. I've used it with other apps, but when I use Hex, Hex is more powerful and smooth, yet it doesn't support external API keys. In terms of the the quality of the small model is not good enough especially when used with other languages. Also, using the whisper 3, it has a latency especially with the eco laptops. Adding an external API key is a good idea because it will help many people work with it. In the future when Hex becomes an app on Google Play or Apple Store, the API will already be there. For sensitive data, I think 90 present of the use it will not involve sensitive it is only prompt, if there there is any sensitive data, this point should told to the user to use the local model instead.

   …

   On Sat, Nov 14, 2026 at 5:27 PM Dax ***@***.***> wrote: *thdxr* left a comment (anomalyco/hex#1822) <#1822 (comment)> hex is local-only for transcription right now, so audio never leaves the machine. a bring-your-own-key option is something we're thinking about, but haven't decided how it should work yet. it's also come up in #765 <#765>. in the meantime the lighter models may help on a slower laptop. granite speech 5.0 is around 500 MB, or sensevoice small at around 250 MB if your language is covered. which provider would you actually use? and which language do you dictate in? — from 𝕺𝖕𝖊𝖓𝕮𝖔𝖉𝖊 — Reply to this email directly, view it on GitHub <#1822?email_source=notifications&email_token=AIT3FCVWXMFMXUTGQHAEJQ35ZAQP3A5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTMOBXGUYTSOBWGYZKM4TFMFZW63VGMF2XI2DPOKSWK5TFNZ2KYZTPN52GK4S7MNWGSY3L#issuecomment-6875198662>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AIT3FCRZ7QMXYYXHJ5A2NCL5ZAQP3AVCNFSNUABGKJSXA33TNF2G64TZHMYTGMBUGQ3DSNZVGE5US43TOVSTWNRTHA4DANBXHE2DNILWAI> . You are receiving this because you authored the thread.Message ID: ***@***.***>