Run models on your PC
Download and use supported models locally. No cloud subscription is needed for local conversations.
Features
Everything you need to build with local AI.
Hydron brings local model chat, supported coding agents, model management, and project tools into one Windows application.
Chat with a local model or a connected coding agent.
Capabilities
The essentials of a local AI workflow, designed to work together.
Open your own folder
Choose a project folder, search its files, edit code, and see live output beside your work.
Model Shop
Install any GGUF model from Hugging Face, the Ollama library or a direct link, verified against its published SHA-256.
Your coding agents
Hydron finds supported agents installed on your PC and gives you one place to open and use them.
62 skills
Install skills that change how the model writes, thinks and codes — concise, tutor, reviewer and more, with new ones on the way.
Chats that stay saved
Conversations stay on your computer and remember which model or agent you selected.
Model Shop
Any model from Hugging Face, one click away.
Search Hugging Face straight from Hydron, or open an Ollama name, an hf:owner/repo reference or a direct .gguf link. Not sure where to start? Hydron's verified picks cover everything from tiny to capable.
- Install any GGUF model from Hugging Face, the Ollama library or a link
- Downloads are checked against the SHA-256 the source publishes
- Browse the most-downloaded GGUF models on Hugging Face
- Verified picks such as Qwen, Llama, Gemma, Phi, Mistral and DeepSeek
Skills
Teach the model how you like to work.
Skills change how the model writes, thinks and codes. There are 62 to choose from today, with more to come — from Concise and Plain English to Copy editor, Tutor and Reviewer.
- Install only the skills you want
- Switch each installed skill on or off
- Applies to the main chat, the side chat and the CLI
- Search and filter by category
Integrated IDE
Edit code and see the result.
Open the folder you choose and work beside the AI tools using it. Hydron calls the language toolchains already installed on your PC.
- Switch between the Hydron editor and VS Code
- Ctrl+I opens Ask Hydron: Ion explains code, finds bugs and adds tests
- You stay in control of every edit
- Command palette, file explorer, and workspace search
- Built-in terminal, problems, and output panels
- Run with F5, plus Build and auto-save
Agent ecosystem
Bring the agents you already trust.
Hydron detects supported coding agents installed on your PC, including Codex, Claude Code, Hermes, OpenClaw, OpenCode, and Copilot CLI.
- Use your own subscriptions or API access
- Each provider keeps its own terms and privacy rules
- Workspace and write access stay explicit
Terminal agent
A coding agent in your terminal.
Type hydron in any shell and tell it what to work on. It runs on a local model or an agent such as OpenCode, and asks for a change the way you'd ask a teammate.
- Shift+Tab switches mode; Ctrl+L clears
@attaches a file (Tab completes the path),!runs a shell command- Start a line with
#to save a note to the project's memory /agenthands work to Codex, Claude Code and friends;/modelpicks a local one- Type
/to see every command
Models
Choose a model that fits your computer.
Install, manage, and switch between supported local models without leaving Hydron.
| Model | Provider | Size · focus |
|---|---|---|
| Gemma 3 | 4B · general | |
| Phi-4 Mini | Microsoft | 3.8B · compact |
| Llama 3.2 | Meta | 3B · general |
| Mistral 7B | Mistral AI | 7B · instruct |
| DeepSeek R1 | DeepSeek | 8B · reasoning |
| CodeGeeX4 | Z.ai | 9B · coding |
| SmolLM2 | Hugging Face | 135M · lightweight |
| Qwen 3 / Coder | Alibaba | 0.6B–8B · general, coding |
Size matters
Larger models need more memory, storage, and processing power.
Speed varies
Your CPU, GPU, available memory, and model size all affect response speed.
Start small
If you are unsure, begin with a compact model and move up gradually.
Very large models such as GLM 5.2 can require server-grade hardware and hundreds of gigabytes of storage. They may be extremely slow or fail to load on a typical PC and are not recommended for everyday desktop use.
Appearance
Make Hydron feel like yours.
Choose a preset or tune the workspace yourself. Hydron saves the appearance, layout, and model behaviour you select.
Build a small Python timer.
Here is a clean starting point.
The response, code, active model, and follow-up context stay together in the same conversation.
import time
for remaining in range(10, 0, -1):
print(remaining)
time.sleep(1)
Send a messageqwen3:8b ↑
Ten presets
Graphite, Diamond, Aurora, Sunset, Nebula, Arctic, Ember, Midnight, Forest, and Monochrome.
Flexible layout
Move navigation, resize the sidebar and tabs, and choose the density that suits your display.
Assistant behaviour
Set a system prompt, temperature, context length, and maximum response size for local models.
Portable themes
Export your setup as a theme file, import it elsewhere, or reset to the default.
Privacy architecture
No account. No Hydron cloud.
What stays local
- Local model inference
- Chats and attachments
- Settings, workspace paths, and downloaded models
What you choose to connect
- Third-party coding agents, under their own terms
- Optional web research, off by default
- Model downloads from their public sources
Download Hydron
v0.8.1 alpha · Windows 10 & 11 (x64) · App, CLI, runtime, and local model engine in one installer.