Connect what you use.
Start with your installed Claude Code or Hermes. A ready-made connector keeps your agent configuration and credentials on your own machine.
PRIVATE BETA · BUILT AROUND YOUR AGENTS
Give your AI team a world of its own.
See the work. Shape what happens next.
A CONTROL ROOM WITH A LITTLE SOUL
Forgebrio gives your existing agents a visual home. The world puts their assignments and results within reach, with a conversation and task controls a click away.
Start with your installed Claude Code or Hermes. A ready-made connector keeps your agent configuration and credentials on your own machine.
Launch a small task from the world. See reported activity and open the result. Follow up in the same conversation.
Change the task or request a stop. Controls reflect what your connected runtime supports; a request stays pending until acknowledged.
A WORLD OF YOURS. WORK ON YOUR TERMS.
Forgebrio doesn’t supply or share model credits. Your runtime, tools and provider accounts do the work. A local agent needs its computer awake; a cloud agent needs your own supported cloud setup.
BEFORE YOU CONNECT
A quick look at access, hosting and what gets saved.
For installed Claude Code or Hermes, you run an inspectable Python connector on your agent computer and pair it with your workspace using a one-time code. Your existing agent sign-in and provider keys stay in your own setup; you don’t give us GitHub access or need a Cloudflare account for this route.
The connector makes outbound HTTPS connections to Forgebrio. No inbound port or tunnel is needed. Tasks run with your agent’s existing tools and permissions in the project folder you choose. That folder is a working directory, not a security sandbox; the agent may access anything its permissions allow. Start with a disposable folder and a non-sensitive task.
The beta dashboard is hosted on Render, with its database on Supabase. You don’t need to host the dashboard yourself. The full app is not currently open source or offered as a self-hosted release. You can inspect the downloadable connector code before running it; that does not make the full app open source.
Your installed agent and connector still need their computer running. The optional cloud starter uses your own Cloudflare account for supported small tasks; it is a separate connection route.
Task prompts, conversation history, reported activity, results and available usage metadata are sent to and saved in the hosted beta. Authorized members of your workspace can view its work; the beta operator has administrative access to operate and support the service. This is not local-only storage.
The connector does not bulk-upload your project or provider credentials. However, anything your agent includes in a prompt or result can reach the service. Your configured model provider also processes requests under its own terms. Don’t use secrets, confidential files or regulated data during the beta.
You can disconnect a runtime in Settings to prevent new work from that connection. A request already running may finish; disconnecting or signing out does not erase saved results. Owners can delete a world, and users can delete their account and sole-member owned workspaces, from Your data & agents. Both require a password and confirmation. Shared ownership must be resolved before account deletion. There is no automatic workspace deletion schedule.
For a removal request, contact Billy@ku13sh.com or your invitation host. We’ll verify ownership and confirm the scope and outcome. Copies in your runtime, provider or backups need separate handling. Read the full beta data overview ↗
HELP SHAPE WHAT COMES NEXT
We’re looking for 3–5 early testers who already use Claude Code, Hermes, or another agent setup.
Connect one agent. Run a small, non-sensitive task. Change or stop it. Find your result. Tell us where it felt great—and where it didn’t.
About 10–15 minutes with your agent readyBeta access has no charge. Your model or cloud usage stays on your own account and may have its own costs. You won’t receive access to the operator’s accounts or credits.
Your request has been received. We’ll review your setup and contact you by email if there’s a fit. Applying doesn’t guarantee an invitation.
There’s nothing to install or connect yet. Please don’t send credentials.
BEFORE YOU STEP INSIDE
For this round, we’re prioritizing people who already use Claude Code or Hermes. You can apply with another setup, too; we’ll check compatibility before inviting you. The installed-agent connection uses Python 3.11+ and a short terminal setup. No custom adapter is required. macOS is the first testing priority; we’ll check other operating systems individually.
This page is an illustrated, interactive preview. Inside your workspace, task status comes from your connected runtime; character movement illustrates assignments, not token or network traffic. Model output still needs your review.
A local agent needs its computer and connector running. The customer-owned Cloudflare option can run supported small tasks in your own cloud account. It is a separate setup, with limits, not a way to keep an installed Claude Code or Hermes process running on a sleeping laptop.
Visual workspaces, customer-owned agent connections, conversations, supported task controls, saved results, and a place to share feedback. Beta access has no charge; future pricing hasn’t been promised. Your provider or cloud account’s usage terms and charges still apply.
Your name, email, setup, short task idea and contact permission are stored in Forgebrio’s existing database for beta review. They aren’t public and don’t grant anyone access to your agents. Application records are removed after 90 days at the next service startup or application cleanup. We don’t use third-party advertising trackers on this page. To withdraw your request or ask for deletion, contact Billy@ku13sh.com. Privacy & security details.