Local AI that keeps your GPU quiet.
Ocadox is a private desktop app for Windows, macOS and Linux, built on Ollama. You get chat and a full coding agent that runs on your own machine. Between agent steps it waits for your GPU to cool down, so the fans stay at normal speed.
Windows and Linux (NVIDIA GPUs), macOS on Apple Silicon (beta). Runs locally, no account needed.
How it keeps things quiet
Running a model hard for minutes at a time heats the GPU, and the fans ramp up to match. Ocadox manages the heat instead of letting it build.
Fan guard Free, on by default
- The agent takes a step. It reads a file, writes code, or runs a command with the model generating.
- Ocadox checks the GPU temperature. If it is above your target (60 °C by default), the agent pauses.
- The GPU cools, then work continues. Fans never need to spin up past their normal speed. No admin rights needed.
GPU power cap Optional
For NVIDIA cards, via NVML.
Set a wattage limit that applies while the model is generating. Less power means less heat in the first place. The live thermal panel shows temperature, fan speed, and watts as you work, so you can see what each setting does.
What you get
A real coding agent
Reads, edits, and searches files, runs shell commands with your approval, uses git, and can browse the web if you turn that on. Each chat works in the folder you set for it.
Any model, in-app
Download from the Ollama library or any GGUF on Hugging Face. Models are labelled by whether they fit your GPU.
Long chats that hold up
Older turns are summarised automatically, so the agent keeps track of the task instead of losing it to the context window.
Live thermal panel
GPU temperature, fan speed, and power draw, updated while you work.
Saved, searchable chats
Every conversation is kept locally and can be searched.
Private by design
Models run on your machine through Ollama. Nothing goes to a cloud service.
Free vs Pro
The free app does everything core. Pro adds tools for people who lean on the agent every day.
| Feature | Free | Pro |
|---|---|---|
| Chat and full coding agent, all tools | Yes | Yes |
| Model downloads (Ollama + Hugging Face GGUF) | Yes | Yes |
| Fan guard and GPU power cap | Yes | Yes |
| Saved chats and chat search | Yes | Yes |
| Checkpoints with one-click undo, per-file diff review | No | Yes |
| Agent branches (own git worktree, merge or discard) | No | Yes |
| Quiet benchmark (energy per token, peak temp, fan, speed) | No | Yes |
| Custom agents (own instructions, model, power cap) | No | Yes |
| Night Shift: queue tasks for quiet hours | No | Yes |
| Semantic code search (local embeddings) | No | Yes |
| MCP servers | No | Yes |
| Auto-approve rules | No | Yes |
| Energy report (kWh, cost, time hot or loud) | No | Yes |
| Export chats to Markdown or HTML | No | Yes |
| Remote Ollama on your network | No | Yes |
Pricing
Ocadox Free
$0
- Chat and the full coding agent
- Fan guard and power cap
- Model downloads, saved chats, search
Ocadox Pro Lifetime
$79
- One-time purchase, lifetime license
- Every Pro feature in the table above
- Use it on up to 3 of your computers
- Works offline after activation
- 30-day refund, no questions asked
No subscription. Checkout is handled by Gumroad.
Your code stays on your machine
Ocadox talks to Ollama on your own computer (or, with Pro, an Ollama server on your own network). Your prompts, files, and chats are not sent to a cloud service.
No cloud model
Every model runs locally through Ollama.
Web is opt-in
Web browsing tools only run when you turn them on.
You approve commands
The agent asks before running shell commands.
Stays in its folder
The agent can only read and change files in the folder you pick for each chat.
FAQ
Do I need Ollama?
Yes. Ocadox is built on Ollama, which runs the models. If you don't have it, Ocadox offers to install it for you on first launch (or get it from ollama.com). Ocadox then lets you download and manage models from inside the app.
Does it need admin rights?
No. Ocadox runs as a normal app, and the fan guard works by pausing the agent until the GPU cools, which needs no special permissions. The optional power cap does need administrator rights; Ocadox asks only if you turn it on.
Which GPUs are supported?
Windows and Linux (AppImage, .deb, .rpm) with an NVIDIA GPU get everything: live GPU telemetry, the fan guard and the optional power cap (via NVIDIA's NVML). macOS on Apple Silicon is in beta: the fan guard reads the chip's own temperature, and there is no power cap. AMD GPUs are not supported yet.
Is it really free?
Yes. Chat, the full coding agent with every tool, model downloads, the fan guard, the power cap, saved chats and search are all free, with no time limit. Ocadox Pro is an optional one-time purchase for extra features.