Open source  ·  Windows

Describe it. Vizio executes it.

Turn plain language into automated media pipelines.
No scripts, no flags — just say what needs to happen.

Download for Windows View on GitHub →
Windows (v0.1.0)  ·  macOS & Linux coming soon

The Vizio workspace — media library, preview, and AI chat in one window.

◂ ▸
crake ↓
Workspace
Tools
Settings
Vizio
Media 4 files
Drop media files here
They'll be copied into your project
interview_raw.mp4
4.2 MB · .mp4
MP4
clip_04_three_...
3.9 MB · .mp4
MP4
output_final.mp4
320 MB · .mp4
OUT
providerOllama
modelgemma4:31b-cloud
statuslocal
preview
output files
session log
Before
No file selected
After
No file selected
Tasks
No tasks yet — describe an edit in the chat
New chat
+ New chat
+
What do you want to do?
Describe an edit, ask a question, or mention a file with @
Trim 30s
To WebP
Extract MP3
Compress
720p scale
Mute audio
Enter to send · Shift+Enter for new line · @ to mention files
How it works

Plain words.
Real pipelines.

No syntax to learn. No flags to memorize. Vizio turns your intent into an executed multi-step workflow.

Describe your goal
Type what you need in plain language. Mention files with @ to give Vizio context. It reads the whole project — file names, sizes, formats — and works with it automatically.
"Compress the interview, add subtitles, mix in the soundtrack at 30%."
Agent plans the workflow
The AI creates a structured multi-step plan — not a one-liner. Every step is tracked, named, and visible in the session log.
Tools execute each step
ffmpeg, Whisper, yt-dlp, ImageMagick — the real tools, running real commands. Full stdout/stderr logged per step.
Self-healing on failure
If a step fails, the AI reads the full error output, identifies the cause, corrects the command, and retries automatically. Up to 3 attempts per step.
01ffmpeg: transcode → 1080pdone
02whisper: transcribe → .srtdone
03ffmpeg: mix audio tracksrunning
04ffmpeg: burn subtitlesqueued
Execution model

The workflow loop

Vizio pipeline: user prompt through self-healing retry to output USER PROMPT "Compress + subtitle" AGENT PLANS JSON workflow EXECUTE STEP ffmpeg / whisper VALIDATE exit code check pass OUTPUT SAVED project/output/ fail → AI diagnoses + retries (max 3×) next step →
Features

Everything in one workspace

AI workflow generation
Describe the goal. The agent plans a structured multi-step workflow — not a one-liner — with every step tracked and logged.
Self-healing execution
On failure, the AI reads stderr, identifies the root cause, corrects the command, and retries — up to 3 attempts per step, without any input from you.
Project management
Projects with dedicated input, output, and session folders. Every command, AI decision, and error fix is saved to the session log.
Media library
Visual file browser with drag-and-drop import. Mention files with @ in chat to pass them as context to any task.
Before / After preview
See source and result side by side, directly in the app. No switching to VLC or QuickTime to check your outputs.
Multi-provider AI
Groq, Anthropic, OpenAI, Ollama, Cerebras, Google AI, and more. Switch models from Settings without touching your workflow.
AI Providers

Every model.
One workspace.

Vizio works with any OpenAI-compatible endpoint. Several providers offer genuinely fast free tiers — no credit card, real throughput.

ffmpeg
ffprobe
Whisper
yt-dlp
ImageMagick
pyannote
Python 3
Groq
llama-3.3-70b
mixtral-8x7b
llama-3.1-8b
Free tier available
⚡ ~315 tk/s · LPU inference
Google AI
Gemini 2.5 Flash
Gemini 2.0 Flash
1,500 req/day free
Multimodal · 1M context
OpenRouter
100+ models
Claude · GPT · Gemini
Free tier available
Model aggregator · unified API
Anthropic
claude-opus-4
claude-sonnet-4
Best for long context
OpenAI
gpt-4o
gpt-4.1-mini
Widest ecosystem
Ollama
Any local model
llama · phi · gemma
100% free · offline
No API key needed
Custom endpoint
Any OpenAI-compatible
LM Studio · vLLM · etc.
Full control
Contributors

Built by

The people who designed, built, and shipped Vizio. Pull requests are open.

T
tomladson
@tomladsn
Creator & Lead Developer
Your name here
Open a pull request →
Security

Your API keys never leave your machine

Keys are encrypted by your operating system the moment you save them. Vizio never stores them in plain text — not in a file, not in a database, not anywhere on disk.

You type your key in Settings
Vizio immediately hands it to Windows — not to any file or database. Windows encrypts it using your user account as the lock. Only your login can ever open it.
When you send a message
Vizio asks Windows to decrypt the key for a fraction of a second — just long enough to attach it to the outgoing request. The instant the request fires, the key is gone from memory.
The response comes back
Your key was never written to disk, never logged, never stored anywhere a process or another app could read it. Vizio itself can't read its own saved keys without your OS unlocking them.

Stop scripting.
Start describing.

Vizio is open source(free) app, and built for Windows.

Windows (v0.1.0)  ·  macOS & Linux coming soon