MeshMind
Local AI flows Open the app ↗

MeshMind

Text, image, sound, tools.
Wired into one flow.

On your own graphics card.
No server, no account, no key.

Your machine is the server.

Wire AI models into flows that think, see, listen and speak.

Every model runs on your own graphics card and processor, inside a browser tab. No server, no account, no API key. The models download once and stay with you.

Text
Qwen 2.5 · Llama 3.2
Image
Janus-Pro 1B
Sound
Whisper · Kokoro
Server
None. It runs in your tab.
01

The canvas

Drop nodes, drag a wire, press enter. Three experts answer in parallel, a fourth merges them, and you watch the signal travel through every step.

  1. 01 MessageWhat you type, a photo or a voice note enters the flow.
  2. 02 SplitThree assistants get the same message at once.
  3. 03 ThinkThey think in parallel, on your own graphics card.
  4. 04 MergeA fourth reads all three and merges them into one.
  5. 05 AnswerOne clear reply. Nothing left your machine.
02

Every sense

Each wire carries a type — text, list, image, sound — and the canvas only lets you plug what fits.

01 / 08 Graphics card Think Assistants with roles
Qwen 2.5 · Llama 3.2
02 / 08 Instant Sort Split into items, judge each one
Join and decide by condition
03 / 08 Graphics card See Describes and reads photos
Janus-Pro 1B
04 / 08 Graphics card Paint Turns a prompt into an image
Janus-Pro 1B
05 / 08 Graphics card Listen Transcribes your voice notes
Whisper
06 / 08 GPU or processor Speak Reads answers in a natural voice
Kokoro, five voices
07 / 08 Online Film Text to video, optional
fal.ai with your own key
08 / 08 Online Reach Tools from MCP servers, optional
an agent picks which to call
03

Your whole machine

Every model gets the best place on your machine: the graphics card when it can, every processor core when it can't. The next model loads while the current one works, so nothing waits.

0 servers reading your prompts
250 KB to open the app
4 kinds of model in one flow
GPU+CPU
working at the same time
04

Start from a template

01 Single assistantMessage, assistant, answer. 02 Panel of expertsThree answer in parallel, a fourth merges them. 03 Draft & reviewA reviewer approves it or sends it back. 04 Brainstorm & rankIdeas, judged one by one, ranked. 05 Illustrate & narrateA short story with a picture and a voice. 06 Photo remixYour photo, described and painted again. 07 Voice assistantTalk to it, hear it answer.
05

Try it here

Not a video: the real app, with the panel of experts already wired. Pick a model, type a message and watch it run on your own machine.

Live · this is the real app, running in your browser

The live demo runs in a desktop browser with a mouse or trackpad (Chrome or Edge): there it works right here, on your own graphics card.

Open the app ↗

Open MeshMind ↗ Chrome or Edge on a desktop. Nothing to install.