Launching today
NonBioS.ai
The AI Software Dev with its own Computer
30 followers
The AI Software Dev with its own Computer
30 followers
NonBioS is an AI software engineer that runs on its own Linux machine. Every user gets a dedicated machine the agent fully controls - 4GB RAM, a public IP, and root access. NonBioS can install anything, run anything, open ports, and deploy. We have obsessed over our context engineering architecture - called Strategic Forgetting, which helps keep focus on long running tasks and handle complex codebase. NonBioS shares its plan upfront, streams every step, and lets you interrupt/redirect anytime.
Hey Product Hunt 👋
I'm Nishant, founder of NonBioS.
They hand you a sandbox. We hand you a server.
A few things that make NonBioS different:
1. The agent actually finishes.
Most agents lose the plot on anything complex - they forget what they were doing, repeat themselves, or drift. We've spent most of our engineering effort on a context approach we call Strategic Forgetting, which keeps NonBioS focused across multi-hour work. In practice: give NonBioS access to your largest codebase and ask it to implement a feature - it will get it done.
2. It's just a Linux box.
No proprietary stack, no lock-in. NonBioS has root access and can set up any stack - Node.js, Python, LAMP, React, whatever. If it runs on Linux, NonBioS can build and deploy it. If you ever want to leave, just take it with you - nothing to export, nothing to rebuild. Free to try, paid starts at $9 - a real plan you can actually launch a SaaS on.
3. Every step is visible.
NonBioS is a terminal agent - it ONLY runs linux commands, and shows you every one. NonBioS operates on top of a linux shell - anything a human developer can do, NonBioS can do. Even file edits happen through a linux command. You can trace progress on long-running tasks, interrupt anytime, or redirect it mid-flow. Everything happens though a browser based chat interface.
4. It works on your existing code too.
Pull in a repo from GitHub, ask NonBioS to deploy it and add a feature, then check it back in when you're done. Not just for greenfield projects.
We built this for developers, but the people who love it most are turning out to be non-technical and semi-technical builders who just have an idea and want to see it running.
I'll be here all day. Thanks for taking a look
Fun fact: As NonBioS has its own VM, and can deploy anything - you can ask NonBioS to setup OpenClaw for you - or for that matter any open source software. We made a video with the screengrab:
Strategic Forgetting is doing a lot of work here! What actually makes it different from rolling summarization or hierarchical memory, both of which everyone else is also calling something branded? Asking because most context engineering claims fall apart when you look at a 6-hour trace
@artstavenka1 I think you hit the nail on the head - "most context engineering claims fall apart when you look at a 6-hour trace". This is the exact problem we are looking to solve and I would argue that nonbios goes farther than anyone else. How much of that 6 hour actually runs successfully, is very hard to predict - on some problems it will run very reliably, on others it might not. But we are extending that limit every day.
I have written a very long blog post which goes into the weeds about Strategic Forgetting if you want to take a look. At a high level Strategic Forgetting seeks to replicate exactly what the brain does when it is targetting a very large problem - it is a structured process in which the agent periodically pauses, reflects on its recent experience, and extracts the essential knowledge from that experience into a dense, organized format. Then it releases the raw history and continues working from the distilled understanding. The phrase that we like to use to describe it is the agent curating its own brain.
Couple of things that make it different from a summarization pass:
1. It is not a single summarization call - it is a multi-pass reflection. NonBioS spends roughly the same 'tokens' in this reflection than it spends on the actual work. So this is a very 'intense' process that we have architected over thousands of agent runs - and we continue to improve over time.
2. This might be surprising to you, but if you know about long running agents then key deprecation happens because of the increase in token counts. NonBioS almost never crosses the 20k token mark because of Strategic Forgetting - what this means is that it remains 'sharp' no matter how large the code base is or how large the problem it is trying to solve.
3. Strategic Foregetting is continuous, not triggered by context pressure. Rolling summarization usually fires when you're running out of room. Strategic Forgetting fires because curation is the point, not because the buffer is full.
But coming back to your point - would love for you to give a shot at the 6 hour trace. In our experience we have only come across one very complex React bug that nonbios was not able to solve - over thousands of sessions.