HomeAbout meRésumé
AI Engineer

Aleksandr
Vechenkov

Octopus OS · A self-evolving agentic OS.
↓ SCROLL FOR DETAIL

Octopus OS

Octopus OS is a desktop app for AI agents, built around self-evolution. Talk or type to them, ask them to write code, research something, or work with your files and apps.

The difference is that agents can change the system they work in while a task is still running. They can add a missing tool, reorganize their work, or set up what should bring them back to it later. Those changes stay available for future tasks, within the permissions you give them.

I spent eight months developing Octopus before publishing the open-source Research Preview. I am now actively looking for contributors. It was built as a product from day one and is usable today.

February 2026–present  ·  Research Preview  ·  GitHub ↗
What exists today.
550+
GitHub stars
60+
GitHub forks
43k+
lines in the Rust core — about 5% of the codebase
860k+
lines across the core, tools, integrations, and desktop app

Code counts are from October 10 and include adapted code. They exclude dependencies, tests, and generated files.

An environment agents can change

An agent needs more than a model: somewhere to work, tools it can use, and a way to keep track of what it is doing. That surrounding setup is often called a harness.

In Octopus, that setup is editable data. Agents can change how their work is organized, what information reaches them, and which tools they use, without stopping the task. Self-improvement is the research question: do those changes help them do better work over time?

Keep related work together
Each project has its own goal, files, and permissions. Tasks stay connected to the project they belong to.
Pick up later
Leave a task running and come back to it. Its progress and results are saved, including across restarts.
Use your computer
Agents can write code, read files, use a browser, and work with connected apps, within the permissions you give them.
Work in parallel
Several agents can work on separate parts of a project at once and wait for results they need from each other.
Build a missing tool
With permission, an agent can write and connect a new tool while working. The next task can use it too.
See what happened
Inspect what ran, which tools were used, what they returned, and what is still waiting.

The desktop app

The desktop app combines a full product for everyday work with professional tools for monitoring the system.

Home is the main workspace. Talk by voice or type, create live widgets, discuss ideas, research, and get work done. From there, launch sub-agents and follow their work as the system runs.

Alongside Home, inspect live system activity, connect services through Applications, and choose where tools run through Drivers. Projects, agents, applications, and drivers are different views of the same underlying system. The terminal app shares the same projects, agents, and tasks.

Octopus Research Preview — HomeOctopus split workspaceOctopus task tools and model controlsOctopus Home voice conversationOctopus system topology — execution zonesOctopus layered world topology

What I built

Over eight months, I designed a new architecture for agent execution around a small Rust core. The desktop app took three months to build, using the Codex source for its agent capabilities.

System design
The Rust core keeps track of work, checks permissions, and records activity. Models, tools, and apps connect separately, so adding a new ability does not mean rewriting the core.
Desktop app and integrations
I built the app in Electron and React, with voice, live widgets in Home, file panes, connected models and tools, and system-monitoring views.
Founder
I developed Octopus on my own for eight months before publishing it. I am now actively looking for contributors to help develop it further.
Open-source Research Preview since October 10, 2026. Try it, read how it works, or explore the source on GitHub ↗.