Sylog

Your agent gets its own computer. You watch it work.

Sylog is a cloud agent that writes, runs and debugs code and researches the web on a cloud computer made for your project, with a full Ubuntu desktop you can watch. Follow every step live, or take over the mouse at any time.

For developers and small teams who hand work to a coding agent.

  • A full desktop per project
  • Models on Google Vertex AI
  • A plan to run work; plans open at launch
  • One Ubuntu desktop per project
  • Runs on Google Cloud
  • Models on Google Vertex AI
  • Usage windows shown live

Most agents work where you cannot see.

You hand over a task and wait for a result. When something fails, you cannot tell where. Sylog's agent works on a real computer you can open at any moment: you see its screen, its terminal and its files as they change.

Product

What you get with every project

A computer that stays yours

Each project has its own Ubuntu desktop with a terminal, a browser, files and internet access. It sleeps when idle and wakes with your files and installed programs in place.

Watch it, or take over

See the screen live. Take the mouse and keyboard to sign in somewhere or fix something yourself, then hand it back.

Every step on the record

Each command, edit and result shows in the chat as it happens. You can interrupt or redirect the agent while it works.

Choose the model and the effort

Pick a model and an effort setting. The app lists the models it offers, all on Google Vertex AI.

Code first, then everything else

It writes, runs and debugs software in the terminal. It also researches the web and works with the project's files; driving the desktop itself is on the roadmap.

Things to hand to an agent with a computer

  • Build a small web app from this spec and run its tests.
  • Find why this build fails on a clean machine.
  • Research three options and write a comparison with sources.
  • Fill this spreadsheet from these documents.
Example prompts, not customer results.

Three steps

  1. 1

    Create a project

    Its computer is set up for you, with its own disk.

  2. 2

    Describe the work

    Write it in the chat, as you would to a colleague.

  3. 3

    Watch, steer, take over

    Stay in the chat, open the screen, or take the controls.

You write to the agent in the chat. The agent runs commands on your project's own cloud computer. You watch that computer's screen, terminal and files live, and you can take over its mouse and keyboard at any time.

A diagram of how Sylog works, not a screenshot of the product. Real frames from a run follow below.

A real run

Frames from a real run in our development environment on 5 October 2026, as captured: one task, the agent working on the project's computer, and its answer.

  1. The Sylog chat before sending: an empty chat on a project whose computer shows as awake with 4 vCPU and 16 GiB, the task 'Create a file fib.py that prints the first ten Fibonacci numbers, run it, and tell me the output.' typed in the message box, and the Files panel listing fib.py
    1. The task: create a file fib.py that prints the first ten Fibonacci numbers, run it and report the output. The project's computer is awake and its Files panel is open.
  2. The same chat while the agent works: the request, then steps 'write fib.py' marked Failed, 'read fib.py' marked Done and 'bash Run fib.py' still running, with a Working indicator below them
    2. The agent at work. Its first step, writing fib.py, is marked failed; it then reads the file and runs it, and the last step is still running in this frame.
  3. The end of the same chat: the three steps, the failed write left visible and the read and bash run marked Done, then the agent's answer that fib.py was already in the workspace, with the output 0, 1, 1, 2, 3, 5, 8, 13, 21, 34
    3. The answer: the agent says fib.py was already in the workspace, ran it and lists the first ten Fibonacci numbers, 0 to 34. The failed write stays in the chat as it happened.

The product, as it is today

A Sylog chat: the request to write fib.py and its unittest tests, then the agent's steps on the project computer -- a bash command, file writes and reads, and a bash run of the tests -- each marked done
A request in the chat and every step the agent takes on the project's own cloud computer, as it happens.
The end of the same Sylog chat: the agent's last steps and its answer, with the unittest output 'Ran 5 tests ... OK' and the first 15 Fibonacci numbers
The agent ran the tests it wrote on that computer and reports what it verified.
Pricing

Monthly plans with usage windows

Every plan has the whole product. Plans differ in how much work fits in each window and how much runs at once.

Payments are not open yet. Running work needs a plan, and sign-up and plans open at launch.

One person, one or two projects
Plus
$20a month
  • The base five-hour and weekly usage windows
  • 2 projects
  • 80 GiB of computer disk
  • 1 computer awake at once
  • 2 chats working at once
Daily work across many projects
Pro
$100a month
  • 5× the usage of Plus
  • 10 projects
  • 400 GiB of computer disk
  • 5 computers awake at once
  • 10 chats working at once
Heavy, parallel work
Max
$200a month
  • 10× the usage of Plus
  • 20 projects
  • 800 GiB of computer disk
  • 10 computers awake at once
  • 20 chats working at once

How limits work

  • Each plan has a five-hour window and a weekly window. The app shows your usage in each as a percent, with the time it resets.
  • When a window is full, new work waits until it resets. Work already running finishes its step and pauses.
  • You can add a balance to keep going past a window. It is used only if you turn that on.
  • Storage has its own limit per plan and never stops work that is already running.
  • When payments open, you cancel from your account in the app. If a plan ends, your projects are archived at once and deleted after 90 days, with notices 30, 7 and 1 days before. Choose a plan before then and they come back.
FAQ

Questions people ask first

Short answers about the computer, the models, your data and the plans.

Ask a question

Who builds Sylog

Sylog is built and operated by its sole founder, Maksim Mikhnev. It is in early access, and the public is not admitted yet.

About the founder and where Sylog stands

Give your next project its own computer.

Sylog is in early access, and sign-up and plans open at launch. Write to us to join the early-access waiting list, or see where it stands today and what we build next.

Request early accessSee where Sylog stands

A monthly plan with usage windows; your balance is used only if you turn it on.