Agents that can act
Give an agent instructions, a model, skills, and approved tools. Let it search, read, write, and run tasks in a workspace.
AGENTIC WORK · COMPANY-GRADE CONTROL
Love Claude Cowork, Codex, or ChatGPT Work? Bring that way of working to your whole company.
GO connects your choice of models, company knowledge, and approved tools in a workspace with central access and privacy controls.
GO interface · Demo data shown
THE PLATFORM / 01
GO turns a chat into a working environment. Agents can reason across your files, use approved tools, and produce outputs in the same workspace your people use.
Give an agent instructions, a model, skills, and approved tools. Let it search, read, write, and run tasks in a workspace.
Create shared knowledge spaces, upload documents, and grant access to the people and groups who need them.
Upload files, browse a project, preview documents, and keep work alongside the conversation that created it.
THE CONTROL PLANE / 02
Useful AI needs clear ownership. GO gives administrators a central place to decide who can use which models, tools, skills, and integrations.
See the admin screensOIDC sign-in through an OAuth2 proxy, individual workspaces, and user and group permissions. Model grants apply to admins, too.
File and shell tools run against the selected workspace in a Docker sandbox, with Gondolin available as an alternative. Administrators set the sandbox network policy; trusted extensions run on the app server.
Manage system skills, Pi extensions, MCP servers, and web search settings centrally, then grant them deliberately.
Review usage and active sessions from the admin dashboard, including live sandbox resource data when Docker is available.
HOW IT WORKS / 03
Configure the models and integrations your organization uses.
Set access by user or group and compose agents from approved capabilities.
People chat, attach files, search knowledge, and let agents use tools.
Keep conversations and resulting files together in a workspace.
INSIDE GO / 04
Explore the current GO interface with sample data: a team conversation and central administration for access, agents, skills, and models.
Keep the discussion, model choice, and agent response together in one workspace.
OPEN FOUNDATIONS / 05
GO is built on open-source foundations, including the Pi agent harness. An open-source release of GO is planned. Bring the models your organization approves, extend agents with skills and tools, and give teams a shared place to work.
Capabilities depend on the models, extensions, and permissions your organization enables.
EARLY ACCESS / LIMITED PILOT
We’re looking for mid-size companies to test GO with real teams and real workflows. If you have roughly 50–1,000 people and a use case in mind, tell us what you would try.