LLM
Local or hosted models
The desktop AI companion
A voice.
A personality.
A place on your desktop.
Talk naturally. Interrupt freely. Sherka gives your AI a transparent, expressive presence on your desktop — connected to your own models, voices and image tools.
- Windows · Linux
- Local voice
- 7 languages
- GPLv3
Eight-minute overview
See Sherka
in motion.
Voice, companions, expressions, scenes and character creation in one short tour.
The connected AI ecosystem
One companion.
Every capability.
Intelligence, voice and creation — connected directly to Sherka.
Fully local.
Models, voice and images can run on your own machine.
Remote LLM APIs are optional
VoiceStudio
Speech recognition + synthesis
ComfyUI
Image & avatar generation
- Your models
- Your voice
- Your creations
Meet your companions
Five characters.
Seven expressions each.
Choose your company, then give each character its own voice and model. Expressions follow the tone of a reply and blend with a gentle half-second crossfade.
Use the bundled images or assign your own, alongside each character's VoiceStudio voice. Persistent moods can let reactions linger and soften over later exchanges.
A conversation that keeps up with you
More than a chat window.
Interrupt a spoken reply — it fades out instead of cutting. Pause, then continue. Streaming and speech prepared ahead keep the exchange flowing.
A personality, image, voice and provider for every companion. Each keeps its own history, memory and desktop layout.
An independent transparent avatar. Move it anywhere; controls appear on hover and follow your theme. Double-click an illustration for full screen.
Persistent memory per companion. Review, edit or delete facts, and preview memory optimization before applying it.
Search the web with sources — DuckDuckGo without an account, or Brave, Tavily and SearXNG. Drop text files and images onto your companion.
French, English, Spanish, Italian, German, Simplified Chinese and Japanese — for the interface and the built-in Sherka.
Turn conversations into scenes
Your request
sets the scene.
Connect ComfyUI, pick an Illustrious or SDXL model and your favorite LoRAs. Images start preparing as soon as the reply text is ready, while the voice continues.
- Up to eight variations, looping with gentle crossfades
- Live mode follows the latest exchange — and pauses for speech
- Portrait, landscape or square, with an optional ×2 upscale
New in 1.3 · Character creation
Bring your own
companion to life.
Your connected LLM proposes an editable personality and appearance. ComfyUI does the rest, in one guided flow.
- 01
Approve its portrait
A front-facing head-and-bust portrait with SDXL or Illustrious, on a solid gray background.
- 02
Give it six emotions
Flux Klein 4B edits the approved portrait. Generate or retry each expression individually.
- 03
Make it transparent
Automatic background removal on all seven portraits, then choose its VoiceStudio voice.
Local voice. Your choice of intelligence.
Connect what you already use.
LM Studio
Your downloaded models. Conversation can work fully offline.
API providers
OpenRouter, Z.ai, OpenAI, Anthropic or any compatible API. Keys stay in your system's credential store.
Hermes
Hermes keeps its personality, memory and tools; Sherka handles the avatar and voice.
From download to your first conversation
Ready when you are.
- Pick your download — Windows installer, no-install ZIP, or Debian package.
- Connect voice and intelligence — VoiceStudio in Settings, then LM Studio, an API or Hermes.
- Choose a companion, turn on the mic, and talk.
From 8 GB of VRAM with small quantized models. Headphones recommended for voice interruption. Models and voices are separate downloads.