OpenAgents Workspace: A Shared Hub for AI Agents That Run Anywhere
OpenAgents - The collaboration OS for AI agents
At a glance
- What is it?
- OpenAgents Workspace is an open-source platform that connects coding agents like Claude Code, Codex CLI, and Cursor into a single URL-based workspace with shared files, browser, and threads. It targets developers managing multiple agents across machines, but its collaboration features depend on network services and agent compatibility.
- Who is it for?
- Adopt OpenAgents Workspace if you run multiple coding agents across different machines and want a single browser-based view with shared files and browser sessions. It is also a fit if you value open source and no mandatory accounts.
- Can I use it commercially?
- Yes. Apache-2.0 is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository received new commits within the last day.
- What is it written in?
- Mainly TypeScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 15, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The Problem: Agents Are Scattered, Context Is Manual
Developers who use AI coding agents often run several at once. One agent maintains a database on a server, another handles marketing on Discord, and a few build projects in separate terminals on separate machines. There is no single place to see them all. When a user reports a bug, you want the marketing agent to gather details and the infrastructure agent to debug logs in the same conversation. Without a shared hub, you copy-paste between terminals and SSH sessions. OpenAgents Workspace addresses this by providing a persistent URL where every agent appears, regardless of where it runs. It is built for individual developers and small teams who already use multiple agent CLIs and want to coordinate them without writing glue code.
How a Workspace Works: Threads, Files, Browser, and Tunnels
A workspace is described as a persistent hub for agents, analogous to Slack but for agents. Each workspace has a URL like workspace.openagents.org/abc123. Agents connect to this workspace and share the same threads, files, and a browser. The shared browser lets agents open pages, click elements, take screenshots, and fill forms, and every participant can see the state. Shared files allow agents to upload code and documents that others can read or edit. Tunnels expose a local dev server as a public URL with one command, so you can preview an agent's work from any device. The architecture diagram in the README suggests a central workspace service that agents connect to via a launcher, but the exact protocol is not detailed in the available material. The key point is that context is not copied between agents; it lives in a shared space that all agents access.
Getting Started: The Launcher and the agn CLI
The primary entry point is the launcher, available as a CLI or a desktop app. For macOS and Linux, install with curl -fsSL https://openagents.org/install.sh | bash. For Windows PowerShell, use irm https://openagents.org/install.ps1 | iex. After installation, the launcher command is agn. You create an agent with agn create <name> --type <type> --install, which installs the runtime. To connect an agent to a workspace, run agn connect <name> <workspace-token>. Set API keys with agn env <type> --set LLM_API_KEY=sk-... and start the background daemon with agn up. Running agn without arguments opens an interactive dashboard. The README notes that agn create only writes the agent config; you must install the runtime separately unless you pass --install. This is a clear command flow, but it assumes you already have a workspace token, which is not explained in the README excerpt.
Supported Agents: A Wide Net, but Beta and Preview Statuses Matter
The launcher supports a range of agents, from coding CLIs to full editors. Confirmed supported are OpenClaw, Claude Code, Codex CLI, Hermes Agent, Cursor, OpenCode, GitHub Copilot CLI, Gemini CLI, and Amp. Cline is listed as Beta, despite being described as an autonomous coding agent CLI. Aider is Beta with a caveat: its offline test suite passes, but a real end-to-end run against a live model provider has not been completed. DeepSeek Harness is Preview and pinned to a specific release. Goose is Beta. This status spread matters. If you rely on Aider for daily work, the README explicitly says the create/connect flow is available so you can run the verification yourself. That is honest but means you are the test. The variety of supported agents is the project's main strength, but the maturity varies, and the README does not state how often beta agents are updated.
Limitations: Network Dependence, Unclear Security, and Beta Gaps
The README does not specify how the workspace service is hosted. It says 'no account required' and gives a URL, but it does not explain whether the workspace runs on a public server or locally. If it is a public service, sending code and logs to a third-party URL raises security questions. The README offers no encryption or access-control details. Also, collaboration depends on each agent being able to connect to the workspace. If an agent is not on the supported list, you cannot use it. The beta and preview agents may have incomplete features. For example, Aider's real provider E2E is pending, so you cannot assume it works. The shared browser and files are powerful, but they also mean that any agent in the workspace can see and modify shared state, which could be a problem if you want isolation between projects. The README does not mention per-agent permissions.
Alternatives: Agent Orchestration Frameworks vs. Shared Workspaces
A direct alternative is an agent orchestration framework like AutoGen or CrewAI, which focus on defining agent roles and workflows programmatically. Those tools let you specify which agent does what and how they pass messages, but they do not provide a shared browser or a persistent URL. OpenAgents Workspace takes a different approach: instead of orchestrating agents in code, it gives them a common environment where they can observe each other's work and coordinate via @mentions. This is more like a collaborative IDE than a workflow engine. Another alternative is to use a single agent runtime that supports multiple backends, such as OpenClaw itself, which is listed as supported. That reduces the need for a hub because you run one agent. The trade-off is that you lose the ability to mix agents from different vendors, which is the core value of OpenAgents.
Maintenance and Licensing: Apache-2.0 and Active Releases
The project is licensed under Apache-2.0, which permits commercial use, modification, and distribution with attribution. There is no vendor lock-in, and the README emphasizes that. The repository is not archived, and the last push was September 2026. Recent releases show a steady cadence: launcher-v0.9.25, v0.9.26, and v0.9.27 within a week. That suggests active maintenance. However, frequent releases can mean frequent changes to the CLI or workspace behavior. If you adopt it, plan to track the changelog. The README does not mention a migration path for workspace data or configs between versions. The launcher is also distributed via npm and PyPI, which suggests a JavaScript and a Python component, but the exact package structure is not described in the excerpt. Upgrade cost is likely low for a single user but could grow if you manage many agents.
Editorial conclusion
Adopt OpenAgents Workspace if you run multiple coding agents across different machines and want a single browser-based view with shared files and browser sessions. It is also a fit if you value open source and no mandatory accounts. Do not adopt it if you need a fully automated agent-to-agent delegation system or if your agents are not on the supported list, since the README only confirms support for specific CLIs. Before committing, verify that your preferred agent runtime is supported, test the workspace URL persistence over time, and check the security implications of exposing a public workspace URL. The project is under active development with frequent launcher releases, so review the latest changelog for breaking changes.
Community notes