New line / numbered list etc

[g33khubber]

1. Can it mimic the features of Wispr, E.g. when I say "home address", it automatically populates my actual address. when it says new line, it should go to a new line.
2. it will be good if I can customize my prompt before the voice gets sent for transcription.
3. it is currently overzealous in correcting words in my directory. For instance, I say "lat" often. While it is similar to let, but it is different from LOT but it still corrects to lat. is there any other more intensive model that can be aware?
4. Appreciate if we can toggle to using fasterwhisper instead too.

[cjpais]

You can probably do most of this using the post processing feature which is in alpha (ctrl/cmd+shift+d) to enable the debug menu. It will allow you to switch on this feature

[g33khubber]

I just tested it. can I confirm this prompt is processed locally? I did not enter any API key and after editing the prompt, it did not seem to do anything to the output at all.

E.g. If number one... number two and so on are mentioned, format it as a numbered list.

[cjpais]

@g33khubber it does not work locally unless you are using the Apple Intelligence

[g33khubber]

I just connected via OpenRouter and using Gemini 3 flash preview, but the post-processing doesn't seem to be happening. How do I verify that it's working?

…

On Tue, 30 Dec 2025 at 17:49, CJ Pais ***@***.***> wrote: @g33khubber <https://github.com/g33khubber> it does not work locally unless you are using the Apple Intelligence — Reply to this email directly, view it on GitHub <#497 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ALU4562H2RTXEDQ4P254GYD4EJDDBAVCNFSM6AAAAACQDCXD2KVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTKMZXGM4TAOA> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[cjpais]

Look at the logs, there should be prints indicating or not. Make sure you have a valid api key, it is required

[g33khubber]

got it! do u have a favourite free model that u use for basic cleaning? I tried plenty and so far The only acceptable one with low latency is Mistral 3.1. Google Flash 2.0 is supposed to be the best for this use case, but it keeps getting rate limited. Do you have any other suggestions?

…

On Wed, 31 Dec 2025 at 12:30, CJ Pais ***@***.***> wrote: Look at the logs, there should be prints indicating or not. Make sure you have a valid api key, it is required — Reply to this email directly, view it on GitHub <#497 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ALU4565TUAUD7LIBXOFAPR34ENGNDAVCNFSM6AAAAACQDCXD2KVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTKMZXHA4TCNY> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[cjpais]

I've found qwen3 30B A3B pretty good, I would guess there are free versions for it but I am not sure

[g33khubber]

I would also like to add that the app acts funky sometimes because it passes the empty transcript to open router, resulting in this output instead. "Sure, please provide the transcript you would like cleaned."

…

On Wed, 31 Dec 2025 at 15:19, Dylan Smith ***@***.***> wrote: got it! do u have a favourite free model that u use for basic cleaning? I tried plenty and so far The only acceptable one with low latency is Mistral 3.1. Google Flash 2.0 is supposed to be the best for this use case, but it keeps getting rate limited. Do you have any other suggestions? On Wed, 31 Dec 2025 at 12:30, CJ Pais ***@***.***> wrote: > Look at the logs, there should be prints indicating or not. Make sure you > have a valid api key, it is required > > — > Reply to this email directly, view it on GitHub > <#497 (reply in thread)>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/ALU4565TUAUD7LIBXOFAPR34ENGNDAVCNFSM6AAAAACQDCXD2KVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTKMZXHA4TCNY> > . > You are receiving this because you were mentioned.Message ID: > ***@***.***> >