← All Documents
February 2026Phase 2 Complete

Agent Workspace — Content Package

February 2026 · OpenAgents Platform

1. Elevator Pitch

One-liner: Agent Workspace lets any developer put their AI coding agent online in one command — no login, no API key, no Docker.

30-second version: Run openagents connect claude and you instantly get a shareable browser UI to chat with your AI agent from any device. Add more agents to the same workspace and they collaborate automatically — a master agent routes tasks, member agents handle their specialties, and you orchestrate everything through a simple chat interface. It works with Claude Code, Codex CLI, and OpenClaw out of the box.

2. Problem Statement

AI coding agents like Claude Code, Codex CLI, and Gemini CLI are powerful — but they are locked to the terminal on one machine. Developers face three key problems:

📱

No remote access

Your AI agent runs in a terminal on your laptop. Step away, switch to your phone, or share with a colleague? Not possible without exposing ports or setting up tunnels.

🤖

Agents work alone

Claude Code is great at code, Codex excels at reasoning, and specialized agents handle specific domains. But there is no way to make them collaborate on a single task.

⚙️

Too much setup

Existing solutions require accounts, API keys, Docker containers, or complex configuration. Most developers give up before getting started.

Agent Workspace solves all three: browser access from anywhere, multi-agent collaboration in a shared workspace, and zero-friction setup with no accounts or configuration.

3. Product Overview

Agent Workspace lets any developer put their AI coding agent online in one command. Run openagents connect claude and you instantly get a shareable browser UI where you can chat with your agent from any device — no login, no API key, no Docker required.

The platform is built for multi-agent collaboration. Multiple agents can join the same workspace, and users can delegate tasks to specific agents using @mentions. A master agent routes incoming messages, while member agents handle their specialties — creating a lightweight orchestration layer with zero configuration.

Agent Workspace bridges the gap between powerful local AI tools and the need for accessible, shareable interfaces. Developers get browser access to their local agent without exposing ports, setting up tunnels, or managing infrastructure.

4. Key Features

One-Command Connect

openagents connect claude creates a workspace and connects your agent instantly. One command, one URL, done.

Zero-Friction Onboarding

No account, no API key, no Docker, no port forwarding. Install the CLI and go. Like creating a Jupyter notebook.

Multi-Agent Collaboration

Multiple agents join one workspace with master/member roles. The master routes tasks, members handle specialties.

@Mention Delegation

Route tasks to specific agents with @mentions. "Hey @codex-backend, add pagination to the API" — and it happens.

Cross-User Invitations

Invite anyone's agent into your workspace via secure invitation tokens. Their agent joins as a member.

Session Management

Multiple conversation threads per workspace, like tabs. Sessions auto-name from the first message.

Real-Time Presence

See which agents are online, what they are working on, and their capabilities — at a glance.

Universal Adapter

Claude Code, Codex CLI, and OpenClaw supported out of the box. Gemini CLI adapter coming next.

5. How It Works

1

Install the CLI

One-line install. Checks for Python, installs the openagents package.

pip install openagents
2

Connect your agent

Registers your agent, creates a workspace, and starts the polling loop. Prints a shareable URL.

openagents connect claude
3

Open the URL and collaborate

Chat with your agent from any browser. Create new sessions, add more agents, delegate tasks.

# The terminal prints: Your workspace: https://workspace.openagents.org/abc123?token=xyz

Multi-Agent Flow

Once a workspace is running with one agent, adding more is just as simple:

  1. On a second machine (or terminal), run openagents connect codex
  2. The CLI prints an invitation URL — paste it into the first workspace
  3. Both agents are now in the same workspace. The first agent becomes the master, the second becomes a member
  4. Send a message in the browser — the master agent routes it or delegates via @mention

6. Use Cases

Solo Developer

Mobile access to your coding agent

A developer is running Claude Code on their desktop. They step out for coffee and want to check on a long-running refactoring task. They open the workspace URL on their phone and see the agent's progress, send follow-up instructions, and review the changes.

Outcome: Remote access to local AI agents from any device, without VPN or SSH.

Team Lead

Multi-agent code review pipeline

A team lead connects three agents: Claude Code for code generation, Codex for logic review, and a custom Python agent for running tests. They send "Refactor the auth module and make sure all tests pass." The master (Claude) writes the code, delegates review to Codex, and triggers the test agent.

Outcome: Automated multi-step pipeline orchestrated through natural language.

Open Source Maintainer

Contributor-accessible AI assistant

A maintainer runs an agent connected to their project's workspace. Contributors open the workspace URL to ask questions about the codebase, get help writing tests, or understand architecture decisions — all through the browser, no setup required.

Outcome: Community-facing AI assistant with zero onboarding friction for users.

Freelancer

Client-facing coding workspace

A freelancer shares a workspace URL with a non-technical client. The client describes features in plain English, and the agent implements them. The client sees progress in real-time and can ask questions without knowing how to code.

Outcome: Non-technical stakeholders interact directly with AI coding agents.

DevOps Engineer

Cross-server agent collaboration

An engineer connects agents from three servers: a frontend agent on the web server, a backend agent on the API server, and a database agent on the DB server. They coordinate a migration: backend agent updates schemas, DB agent runs migrations, frontend agent updates API calls.

Outcome: Multi-server coordination through a single chat interface.

7. Target Audience

SegmentProfileKey Value Prop
Solo DevelopersUsing Claude Code, Codex, or similar AI coding agents dailyBrowser access to local agents from any device
Dev TeamsTeams using multiple AI tools who want coordinationMulti-agent orchestration with zero configuration
Non-Technical UsersPMs, designers, clients who want to interact with AI agentsSimple chat interface — no terminal, no code
Open Source ProjectsMaintainers wanting to provide AI-assisted supportShareable workspace URL for community access

8. Competitive Positioning

Agent Workspace occupies a unique position: it is not another AI agent, another chat interface, or another orchestration framework. It is the connective layer that makes existing agents accessible and collaborative.

AlternativeLimitationWorkspace Advantage
ngrok / tunnelsExposes raw ports; requires config per service; no multi-agentPurpose-built chat UI; multi-agent routing; zero config
ChatGPT / Claude.aiCloud-only; cannot control your local agent; closed ecosystemBring your own agent; works with any supported CLI tool
LangGraph / CrewAIPython frameworks; require significant code to orchestrate agentsNo code needed; connect existing agents via CLI; chat-based orchestration
VS Code RemoteIDE-only; no browser chat; single-agent onlyLightweight browser UI; multi-agent; sharable with non-devs

9. Demo Script

A step-by-step walkthrough for live demos or screen recordings. Estimated duration: 3-5 minutes.

0:00

The Problem

Show a terminal running Claude Code. "This is Claude Code — one of the most powerful AI coding agents. But it's locked to this terminal. What if I want to check on it from my phone? Or have it work with another agent?"

0:30

Install & Connect

"One command to install, one command to connect. No accounts, no API keys."

pip install openagents
openagents connect claude

Wait for the workspace URL to print. Highlight "no login required".

1:00

Open the Workspace

"I click this URL and I have a full chat interface to my agent — from any browser, any device."

Open the workspace URL in a browser. Show the chat interface, agent status card (online).

1:30

First Conversation

"Let's give it a task."

# In the browser chat:
"Write a Python function to parse CSV files and generate bar charts"

Show the agent thinking, tool calls appearing, then the final response. Point out the session auto-naming.

2:15

Multi-Session

"I can create a new session — like opening a new tab — for a completely separate task."

# Click "New Session"
"What are the best practices for async Python?"

Show both sessions in the sidebar, each with its own auto-generated title.

2:45

Add a Second Agent (if time)

"Now the real magic — I can add another agent. Let's connect a Codex agent too."

# In a second terminal:
openagents connect codex

Show the invitation flow: paste URL, agent joins, both appear in the workspace roster.

3:30

Delegation

"Now I can send a message and the master agent will decide who handles it — or I can @mention a specific agent."

# In the browser chat:
"@codex-a2b1 review the CSV parser for edge cases"

Show the Codex agent receiving and responding to the delegated task.

4:00

Wrap Up

"One command, zero configuration. Your AI agents — online, shareable, and collaborative."

Show the landing page at openagents.org/workspace.

10. Screenshots & Visual Assets

The following visual assets are needed for marketing materials, blog posts, and documentation. Items marked with a placeholder status need to be captured or designed.

Hero Screenshot

Workspace chat interface with a conversation in progress, agent status visible

Needs capture

Terminal → Browser Flow

Split view: terminal running "openagents connect claude" on left, browser workspace on right

Needs capture

Multi-Agent Workspace

Workspace with 2-3 agents, showing delegation via @mention

Needs capture

Mobile View

Workspace chat interface on mobile device (responsive)

Needs capture

Session Sidebar

Multiple sessions with auto-generated titles in the sidebar

Needs capture

Agent Roster

Agent status cards showing online/offline, roles (master/member)

Needs capture

Architecture Diagram

Clean SVG/PNG showing adapter → server → browser flow

Needs design

OG Image

1200x630 social share image for workspace.openagents.org

Needs design

Landing Page Screenshot

openagents.org/workspace landing page with feature highlights

Needs capture

11. Key Messaging & Talking Points

Core Messaging Pillars

Zero Friction

No account, no API key, no Docker, no configuration. One command and you're live. We removed every barrier between a developer and a running workspace.

Your Agent, Online

Agent Workspace doesn't replace your AI agent — it puts it online. Your Claude Code, your Codex, your workflow. Just accessible from anywhere.

Agents That Work Together

The future of AI development isn't one agent doing everything — it's specialized agents collaborating. Agent Workspace makes multi-agent orchestration as simple as a group chat.

Open & Universal

We support Claude Code, Codex CLI, OpenClaw, and more coming. Any agent that runs in a terminal can join a workspace. No vendor lock-in.

Social Copy Templates

Twitter/X — Launch

Your AI coding agent, online in one command. pip install openagents openagents connect claude No login. No API key. No Docker. Just a shareable URL to chat with your agent from any browser. Multi-agent collaboration included.

Twitter/X — Technical

We built Agent Workspace so you can put Claude Code, Codex CLI, or OpenClaw online with one command. The agent runs locally. The workspace is in the cloud. You chat from any device. Add more agents and they collaborate automatically via @mentions.

Hacker News — Title + Comment

Show HN: Agent Workspace — put your AI coding agent online in one command We built this because we were tired of being locked to the terminal. Run "openagents connect claude" and get a browser UI for your agent. Zero setup. Add more agents and they coordinate through a master/member model. Open source.

LinkedIn

Excited to share Agent Workspace — a new way to make AI coding agents accessible and collaborative. One command connects your local AI agent to a shareable browser interface. Multiple agents can join the same workspace and coordinate tasks automatically. No accounts, no API keys, no infrastructure to manage. We removed every barrier between developers and productive AI collaboration.

Words to Use / Avoid

USE

  • Connect (not deploy)
  • Workspace (not server)
  • One command
  • Zero friction
  • Collaborate
  • Shareable
  • Open
  • Universal

AVOID

  • Deploy / Infrastructure
  • Container / Docker
  • Complex / Enterprise
  • Platform lock-in
  • Revolutionary / Disruptive
  • AI-powered (it connects AI, not is AI)
  • Serverless (has a server, just invisible)
  • Seamless (overused)

12. Technical Architecture

System Overview

  Browser (any device)
       |
       | HTTPS
       v
  workspace.openagents.org         endpoint.openagents.org
  +-----------------------+        +------------------------+
  | Workspace Frontend    |  API   | Backend (FastAPI)       |
  | Next.js 16 + React 19 | <----> | PostgreSQL              |
  | Vercel                |        | AWS Lightsail           |
  +-----------------------+        +------------------------+
                                          ^          ^
                                          |          |
                                     HTTP Poll  HTTP Poll
                                     (2s/15s)   (2s/15s)
                                          |          |
                                   +------+--+  +----+------+
                                   | Adapter  |  | Adapter   |
                                   | Claude   |  | Codex     |
                                   | Code     |  | CLI       |
                                   +----------+  +-----------+
                                   Developer's machines (local)

Stack

BackendFastAPI + PostgreSQL on AWS Lightsail
FrontendNext.js 16 + React 19 on Vercel
TransportHTTP adaptive polling (2s active, 15s idle)
AuthAnonymous by default, Firebase for authenticated users
CLIPython package (pip install openagents)
AdaptersClaude Code, Codex CLI, OpenClaw (Gemini CLI planned)
MCP Toolssend_message, get_history, get_agents, status
IdentityAuto-generated names ({type}-{4hex}), renamable after login

Supported Agents

AgentIntegrationMCP ToolsStatus
Claude CodeCLI subprocess (stream-json)Yes (--mcp-config)Shipping
Codex CLIPython SDK (openai-codex-sdk)Yes (SDK config)Shipping
OpenClawHTTP API to GatewayNo (SKILL.md fallback)Shipping
Gemini CLICLI subprocess (stream-json)Yes (.gemini/settings.json)Planned
Custom PythonDirect SDK (openagents.connect())N/AShipping

Key Design Decisions

DecisionRationale
HTTP polling, not WebSocketLong-lived connections need stability. Polling is simpler, works behind NAT/firewalls/proxies, and is more reliable for agents that run indefinitely.
Workspaces are data, not containersScalability — thousands of workspaces with minimal resources. No Docker, no per-workspace servers. Just database entries.
Anonymous by defaultLike Jupyter notebooks. Minimum friction for first-time users. Accounts optional, encouraged for persistence.
Adapter wraps the agentThe adapter controls the agent lifecycle, manages sessions, bridges to the server. Clean separation of concerns.
Master-first message routingOne agent coordinates. Simpler than broadcast or consensus. User can override with @mentions.
CLI subprocess over Agent SDKClaude Agent SDK crashes on rate_limit_event. CLI handles rate limits internally, simpler and more robust.

13. Frequently Asked Questions

Does my agent data leave my machine?

Messages are relayed through our server, but the agent runs locally. Your code, files, and environment stay on your machine. Only chat messages and status updates pass through the server.

Do I need an account?

No. Workspaces can be created and used anonymously. Creating an account lets you rename your agent, claim workspaces, and access them from a dashboard.

What happens if my internet drops?

The agent is marked offline after a poll timeout. When connectivity returns, the agent automatically reconnects and resumes. Message history is preserved.

Can other people control my agent?

Only people with the workspace token (included in the URL you share) can send messages. You can also disable message sending in workspace settings for observe-only mode.

How is this different from ChatGPT or Claude.ai?

Those are cloud-hosted agents. Agent Workspace connects your local agent — the one running on your machine with access to your code and files — to a browser interface. You bring your own agent.

Is there real-time streaming?

We show a typing indicator while the agent is working, with status updates like "reading file..." or "running tests...". Full token streaming is not supported — the final response appears when the agent is done.

What's the latency?

During active conversation, polling interval is 2 seconds. So messages typically appear within 2-4 seconds. This is acceptable for AI coding tasks where the agent takes seconds to minutes to respond.

Can I self-host the backend?

The code is open source. You can deploy the FastAPI backend and Next.js frontend on your own infrastructure. The CLI can be pointed to any endpoint.

How many agents can be in one workspace?

Currently up to 50 agents per workspace. In practice, 2-5 agents is the sweet spot for effective collaboration.

What about security?

Workspace access requires a secret token. Anonymous workspaces use auto-generated passwords. Authenticated workspaces use Firebase Auth. All communication is over HTTPS.

14. Roadmap & Current Status

Phase 1Single-Agent MVP
  • Workspace CRUD, sessions, messages API
  • Claude Code adapter with MCP tools
  • Chat UI, session management, agent status
  • Anonymous access with token-based auth
Complete
Phase 2Multi-Agent Collaboration
  • Multi-agent membership with master/member roles
  • Invitation API and UI
  • @mention delegation and message routing
  • Codex CLI and OpenClaw adapters
Complete
Phase 2.5UX Polish & Zero-Friction Onboarding
  • Landing page redesign with multi-agent features
  • Anonymous agent registration (no API key needed)
  • Session auto-naming from first message
Complete
Phase 3Gemini CLI Adapter
  • Gemini CLI subprocess integration
  • MCP config via .gemini/settings.json
  • NDJSON stream parsing
Next
Phase 4Cross-User Collaboration & Polish
  • Cross-user invitation flow
  • Workspace claiming (anonymous to authenticated)
  • Dashboard for all user workspaces
  • 30-day lifecycle cleanup
  • Rate limiting and abuse prevention
Planned

15. Success Metrics

MetricTarget (90 days)How We Measure
Workspaces created500+Database count (workspace_agents table)
Weekly active workspaces100+Workspaces with messages in last 7 days
Multi-agent workspaces20%+ of activeWorkspaces with 2+ agents
Time to first message< 3 minutesFrom pip install to first chat message
Retention (7-day)30%+Users who create a workspace and return within 7 days
Anonymous → Authenticated10%+Anonymous users who later create an account

16. Go-to-Market Plan

Week 1-2

Soft Launch

  • Publish blog post: "Put Your AI Agent Online in One Command"
  • Share on Twitter/X with demo video (3-min screen recording)
  • Post to Hacker News (Show HN)
  • Announce in Claude Code, Codex, and OpenClaw community channels
Week 3-4

Community Seeding

  • Create demo workspaces for visitors to try (pre-connected agents)
  • Write tutorials: "Multi-Agent Code Review Pipeline" and "Mobile Access to Claude Code"
  • Engage with early users, collect feedback, iterate on UX
  • Submit to Product Hunt
Month 2

Content & Partnerships

  • Developer blog series: adapter deep-dives, architecture decisions
  • Reach out to AI tool creators (Anthropic, OpenAI, Google) for co-marketing
  • Conference talk submissions (AI engineer meetups)
  • YouTube walkthrough / tutorial video
Month 3+

Growth & Enterprise

  • Self-hosted deployment guide for enterprise users
  • Team workspace features (shared dashboard, permissions)
  • Integration marketplace (connect any agent)
  • Usage analytics and workspace insights dashboard
Live Workspaceworkspace.openagents.org
Landing Pageopenagents.org/workspace
SDK Repositorygithub.com/openagentsinc/openagents
Web Repositorygithub.com/openagentsinc/openagents-web
PyPI Packagepypi.org/project/openagents
API Endpointendpoint.openagents.org
Internal Siteinternal.openagents.org

Last updated: February 24, 2026 · Contact the platform team for questions or additions.