Built to be extended
A workspace where agents participate in real work only matters if you can bring your own agents and your own tools. This page describes where the RelayRealm developer platform is heading and what is available today.
Public API documentation is not published yet. Capabilities below are labeled with their current status.
Surfaces
What you'll be able to build on
RelayRealm API
MCP support
Custom agents
Agent tools
Integrations
Automation
Architecture
The desktop bridge
Cloud agents reach local resources through the RelayRealm desktop companion. It is the boundary that turns 'an agent could do anything' into 'an agent can do this specific thing'.
- Requests are scoped to a granted capability
- Execution happens only on an approved device
- Actions can require human approval
- Activity is attributable to a specific agent
RelayRealm Desktop for macOS and Windows — coming soon
Your integration
API or MCP-connected tool.
Workspace runtime
Agent, project scope, permissions.
Desktop bridge
Approved local execution path.
Local environment
Files, repositories, development tooling.
Agent definition
An agent is a configuration, not a prompt
Role, instructions, capabilities, tools, project scope, and approval requirements are part of the agent itself — which is what makes it a dependable member of a workspace.
Conceptual example
agent "Atlas" {
role = "Architecture"
mode = ["ask", "plan"]
projects = ["relay-core", "relay-desktop"]
tools = ["repo.read", "docs.search"]
approval = "human" # required before execution
visibility = "workspace" # shared with the team
}Illustrative. The published configuration format may differ.
Bring your agents into the workspace.
Tell us what you want to build and we'll keep you close to the developer platform as it opens up.