LIVE
+++ A skill that tells coding agents to skip the paperwork +++ useAgent gives each AI agent its own cloud workstation +++ Texas freezes funding for more Flock AI cameras +++ Skills Hub syncs agent skills across 40-plus AI tools +++ Musicians turn detective on AI music grifters +++ ima2-gen runs image generation on your own machine ++++++ A skill that tells coding agents to skip the paperwork +++ useAgent gives each AI agent its own cloud workstation +++ Texas freezes funding for more Flock AI cameras +++ Skills Hub syncs agent skills across 40-plus AI tools +++ Musicians turn detective on AI music grifters +++ ima2-gen runs image generation on your own machine +++
All news ›
AI IN LIFE AI IN LIFENEWS
DAILY
DE EN
TOOLS

useAgent gives each AI agent its own cloud workstation

One event contract for Claude Code, Codex and OpenCode, plus an isolated Linux workstation for every thread. The project is still alpha.

useAgent gives each AI agent its own cloud workstation

Illustration: a row of empty desks with glowing screens in an open-plan office at dusk.

useAgent is an open-source orchestrator that runs coding agents such as Claude Code, Codex and OpenCode inside an isolated Linux sandbox and hands back finished deliverables.

At a glance

  • License: GNU AGPL v3.0; the maintainers call it alpha software and warn that APIs and schemas may still change.
  • Snapshot at retrieval: 129 stars, 2 forks, 1,356 commits, 0 open issues and 0 open pull requests.
  • Supported engines: Claude Code, Codex and OpenCode behind one shared, provider-neutral event contract.
  • Requirements: Bun runtime and Postgres 16+ with pgvector; backend on port 3201, web UI on port 3400.
  • Per-thread sandbox: terminal, repositories, browser and a noVNC desktop with MP4 recording; credentials stay outside.

useAgent is an open-source orchestrator that runs coding agents such as Claude Code, Codex and OpenCode inside an isolated Linux sandbox and hands back finished deliverables. The code sits at useagenthq/useagent and ships under the GNU AGPL v3.0. Its maintainers describe the current state as alpha.

One contract, three agent CLIs

The center of the project is a shared, provider-neutral event contract. Behind it, Claude Code, Codex and OpenCode become interchangeable, so the calling application does not have to know which engine did the work. Each run is stored as an event timeline in a Postgres database, which makes a completed job replayable step by step.

The listed deliverables are live websites, decks, spreadsheets, research reports and tested pull requests. The agents are meant to run on whatever subscription a team already pays for, and the project publishes no price list of its own.

A separate machine for every thread

Rather than sharing one working directory, every thread gets its own walled-off Linux workstation. That includes a terminal, repositories, a browser and a visible desktop served over noVNC, recorded to MP4. According to the maintainers, credentials never reach that sandbox: knowledge, memory, skills, GitHub, web search, desktop control and artifact publishing are exposed as typed tools served by the backend.

The design separates what the agent can touch from what its operator can touch. Whether that boundary holds under real workloads cannot be judged from the project description alone.

Setup, channels and self-hosting

The quick start expects the Bun runtime and Postgres 16 or newer with the pgvector extension; a Docker command using the pgvector/pgvector:pg16 image is included, since stock Postgres images do not carry the extension. The backend listens on port 3201 and the web interface on port 3400.

Work can arrive through a web app, Slack, a REST API or scheduled runs. Slack and GitHub are wired in natively, while Gmail, Linear, Notion and HubSpot arrive through connectors. For self-hosting, the project points at any Linux host — AWS, Google Cloud, Azure, Hetzner or bare metal — with two pluggable sandbox backends, Daytona and CubeSandbox.

What could not be verified

Only one independent source was available for this article: the repository itself. No second newsroom had examined the project at the time of research. Everything stated here about architecture, the credential boundary and the feature set is therefore a maintainer claim, not an independently measured result.

Also unresolved: no version tag or release date is listed, and no first-publication date is given. The figures of 129 stars, 2 forks and 1,356 commits are a snapshot taken at retrieval. The maintainers themselves caution that interfaces and schemas may shift between releases.

◈ AI-GENERATED REPORT · SOURCES LINKED

FAQ

Is useAgent free?

The repository lists no pricing. The code is licensed under the GNU AGPL v3.0, and the agents are meant to run on an existing Claude Code, Codex or OpenCode subscription.

What do I need to self-host useAgent?

The Bun runtime and Postgres 16 or newer with the pgvector extension. The backend starts on port 3201 and the interface on port 3400; any Linux host is said to be sufficient.

Is useAgent production ready?

By its own description, no. The project is labeled alpha software, and the maintainers warn that APIs and schemas may change between releases.