Work by voice, from anywhere

Speak.
Your computer
gets to work.

Sorema turns a conversation into real work on your own machine. Start a task, close the browser, and come back when it is done.

Your machine. Your OpenAI account. Your call.

Sorema is listening 09:41
You Run the tests in my website project and fix what is failing.
Sorema I’ll start a coding task on your computer. You can leave — I’ll let you know when it’s finished.
Task finished

Tests passing. Two failures fixed.

12m
Conversation active

One conversation

Real work

Your own computer

Still working when you leave

The problem with assistants

Most assistants stop
when the chat stops.

Real work does not fit inside a conversation. It takes tools, access to the right machine, and time.

Sorema separates the request from the work. Your conversation can end. The task keeps running, its progress is saved, and the result is waiting when you return.

How it works

From thought
to finished work.

Three steps. No remote desktop. No watching a progress bar.

01

Say what you need

Talk naturally in your browser. Sorema understands the task and asks when something matters.

02

Your machine picks it up

A small, auditable agent on your computer receives only the typed operations it supports.

03

Leave. Come back to done.

The task continues after the call ends. Sorema keeps the result and tells you when it lands.

Built for work that takes time

Talk at the speed
you think.

Let a coding agent handle the waiting, the commands and the context switching.

01

Run, test and diagnose

“Run the full test suite and tell me why it fails.”

02

Change an existing project

“Add the onboarding flow we discussed yesterday.”

03

Pick up where you left off

“What happened with the last task I started?”

04

Work across your machines

Choose the computer and project that have the right files.

Trust is part of the product

The cloud can ask.
Your machine decides.

The component that reaches your files is open source. It connects outward, sees only folders you choose, and offers no arbitrary remote command tool.

Read the local agent source

Outbound only

No inbound port, router setup or public address on your computer.

Folders you choose

Without an explicitly allowed workspace, the agent can see nothing.

Your API key

Voice usage runs on your own OpenAI account, not a shared credential.

No arbitrary exec

Every operation is named, structured and validated at both ends.

Good questions

Before you let software
near your computer.

Does Sorema upload my projects?

No. Work happens through the coding agent on your machine. Sorema coordinates the task and stores its status and summary; it does not mirror your project into its cloud.

What happens when I close the browser?

The conversation ends, but an accepted task keeps running. Progress and completion are saved so you can ask about them later.

Which coding agents can it use?

The local agent currently supports Codex CLI and Claude Code, using the tools and permissions already configured on your machine.

Can it reach every file on my computer?

No. You choose the workspace folders the local agent may access. Paths outside them, traversal attempts and symlink escapes are refused.

A shorter distance between intent and action

Say it.
Then let it run.

Connect your computer once. Start the next task with your voice.

Open Sorema