ai.diy is an open-source, local-first AI workspace that works with your own cloud or local models. Beyond chat, it can research the web, run Python, use MCP tools, compare multiple models, remember context, and let agents work inside a browser-based Linux environment. It is BYOK, self-hostable, and built around one idea: your models, tools, data, and AI workspace should stay under your control.
What's great
I’ve used many AI assistants and agent-style products, and what stands out about ai.diy is that it goes beyond simply generating an answer.
It has access to its own virtual computer in a sandboxed environment, so it can run commands, work with files and actually carry out parts of a task for you. That shifts the experience from “tell me how to do this” to “help me get this done.”
I also like that it reduces the amount of prompt engineering required. You can focus more on the outcome you want rather than figuring out exactly how to instruct the AI.
The combination of a simple interface, action-oriented AI and an isolated environment for getting real work done is what makes it interesting to me.
I’ve also been providing feedback during its development, and it has been great to see some of that feedback make its way into the product.
What needs improvement
It is still an early product, so the biggest opportunity is helping new users quickly understand just how much it can do.
I’d like to see an even broader range of guided examples and real-world use cases, particularly showing tasks where the virtual computer makes a meaningful difference. Making progress on longer or more complex tasks more transparent would also help users understand what the AI is doing and build confidence in handing it more work.
I’ve already seen the product improve significantly based on feedback, so I’m looking forward to seeing how this develops.
Full disclosure: the maker is my son, and I’ve also been one of the people providing feedback during the product's development.
vs Alternatives
ChatGPT by OpenAI
Claude by AnthropicI regularly use products like ChatGPT and Claude and have experimented with several other AI assistants and agent-based tools.
What made ai.diy feel different was the emphasis on execution rather than conversation. With many AI products, I still have to take the answer, move to another tool, run the commands or complete the next steps myself.
With ai.diy, the virtual computer gives the AI an environment where it can actually work through those steps. I also find the experience less dependent on knowing how to write sophisticated prompts, which should make this type of capability accessible to a much wider audience.
Ratings
Ease of use
Reliability
Value for money
Customization
Report
6 views