Speak, listen,
make pictures.
Voice and images run on this computer or with a service you choose. Nothing downloads until you say so.
Where it runs
Models → Voice and Models → Images are one column of settings each. In each group, Runs on says where the work happens:
| Runs on | What it means |
|---|---|
| On this computer | Free and private. Listening uses whisper.cpp, speaking uses Piper, and images use stable-diffusion.cpp on your graphics card. |
| A service | A provider you have a key for, billed by that provider. Each service shows whether its key is saved. One key can cover more than one job: an OpenAI key, for example, listens, speaks and makes images. |
| ComfyUI | Images only: a ComfyUI you run yourself. |
| Off | Voice only. With listening off, the mic button is hidden in chats. With speaking off, answers aren't read aloud. |
A change you pick waits until it can work, for example until its key is saved or ComfyUI answers. Until then, what you had keeps running.
Setting it up
The first time you open Voice or Images, Asyncat offers what suits this computer, with each part's download size. You can change any choice before downloading. Nothing downloads until you press the button, and the setup carries on if you leave the page.
Model files, at the bottom of each page, lists the models on this computer. Use Add from Hugging Face or Add a file… to bring your own.
Listening needs whisper.cpp from your package manager, as there is no ready-made download of it for Linux. Install it, then press Check again. Speaking and pictures set themselves up.
Models → Voice
Voice has two groups:
- Listening turns what you say into text. On this computer you choose a whisper.cpp model, from Base (quick, the recommended start) to Large v3 Turbo (the most accurate, also for accents and many languages). As a service: OpenAI, ElevenLabs or fal.
- Speaking reads answers aloud. On this computer you choose a Piper voice. As a service: OpenAI, ElevenLabs, Gemini or fal.
Before you choose a voice, press the play button next to it to hear it. Hearing a voice never chooses it. With a service, hearing one uses a little of your credit.
Once a group is set up, Try it checks it. Test the mic records you and shows what it heard. For speaking, type some words and press Play. The mic works on its own, but transcribing other audio files needs ffmpeg.
Voice in chats
With listening set up, the mic button in the message box lets you speak instead of typing. With speaking set up, each answer gets a Read aloud player.
When both are set up, turn on Voice mode from the + menu to talk hands-free. Asyncat stops recording when you stop talking, sends what you said, reads the answer aloud, then listens again.
Models → Images
How images are made shows where pictures are made and with what:
- On this computer: Asyncat reads your graphics card and offers a model that fits it, such as Stable Diffusion 1.5 (small and quick) or SDXL Turbo (better pictures in a few steps).
- ComfyUI: enter its address, check that it answers, and pick a checkpoint.
- A service: OpenAI, Gemini, Stability AI or fal. Stability AI and fal make new images but can't change existing ones.
Try it makes a picture from a description. Open Image options to set the shape: Square, Wide or Tall. Depending on where pictures are made, you can also set a seed and things to avoid, and on this computer or with ComfyUI the steps, guidance and sampler. After a picture is made, Use this seed keeps its seed for the next try.
Pictures in chats
Ask for a picture in any chat and the agent makes it with what you chose in Models → Images. To change one, press Change it in a chat under a picture in Try it or in Files, and describe the change. Changing pictures works on this computer, with ComfyUI, OpenAI or Gemini.
Every picture lands in Files → Made by agents, with its prompt and settings, next to the documents, pages and data agents make. See Files.