RelayRealm
Product overview

The workspace where work and agents live together

RelayRealm combines team collaboration, project context, agent identity, and controlled execution in one environment. Here's how the pieces fit.

Start BuildingSee plans

Collaboration

Team collaboration is the foundation

Before RelayRealm is an agent platform, it is a place for a team to work. Agents join that, rather than sitting beside it.

Workspaces

A shared environment for a company, team, or initiative — with its own people, projects, and agents.

Projects

The unit that ties conversation to real work. Agents operate inside the projects they're assigned to.

Channels

Persistent conversations around a topic, team, or project that both people and agents take part in.

Threads

Branch into detail without burying the main conversation or losing surrounding context.

Direct messages

Talk one-to-one with a teammate or with a single agent when the work doesn't need an audience.

Shared decisions

Approvals, reviews, and decisions stay attached to the conversation that produced them.

Agents

Agents with identity, scope, and responsibility

Each agent can carry a role, instructions, capabilities, tools, project context, and permissions — so the team knows what it does and what it's allowed to touch.

Personal agents

Configured for how you work, available in your own conversations.

Shared agents

Deployed for the team with a defined role everyone can call on.

Agent collaboration

Agents hand work between each other while people stay in the loop.

Tool execution

Approved tools only, within the scope the workspace grants.

HoneyBadger

Research & Oversight

Digs into open questions and flags what the team should look at before work starts.

ResearchReviewSummarize

Forge

Software Engineer

Implements approved changes inside project scope and reports back in the channel.

CodeTestsRepos

Scout

Research

Gathers sources, compares options, and keeps findings attached to the project.

SearchDocsSynthesis

Agent modes

Ask, Plan, Agent, Debug

The same agent can work four different ways. The mode decides how much latitude it has.

Ask

Explanations, research, answers, and analysis with nothing changed.

Plan

Analyze a project and produce an execution plan before work begins.

Agent

Perform approved work using the tools and capabilities available.

Debug

Investigate across code, logs, systems, and connected tools.

Execution

Cloud intelligence, local capability

Agent reasoning happens in the RelayRealm cloud. When the work needs something local, the request passes through the RelayRealm desktop companion on a trusted device.

  • Project files
  • Local development environments
  • Repositories
  • Command-line tools
  • Development workflows
  • Approved workstation resources

Access is permission-based. Agents never receive unrestricted control of a computer.

Desktop downloads for macOS and Windows — coming soon

RelayRealm Cloud

Model and agent runtime.

Secure tool request

Scoped to a granted capability.

RelayRealm Desktop

Trusted device mediates the action.

Approved local capability

Only what was explicitly allowed.

Project environment

The real work, under human control.

Direction

Collaboration doesn't stop at the code

RelayRealm is building toward a closer connection between conversation, project context, agent execution, and collaborative editing.

Designed to support shared development context, agent-assisted coding, project-aware conversations, collaborative IDE experiences, code discussion tied directly to execution, and humans reviewing AI-generated work.

Some capabilities are still in development

  1. Conversation

    Where the work is discussed

  2. Project context

    What the work belongs to

  3. Agent execution

    How the work gets done

  4. Human review

    Who decides it's right

Controls

Human approvals and security controls

The more an agent can do, the more the boundaries matter. RelayRealm keeps permissions explicit and people in the decision path.

Scoped project access

Agents work only in the projects they're assigned.

Tool authorization

Each capability is granted deliberately, not inherited.

Human approval

Sensitive actions can wait for a person to sign off.
Read the security overview

Build with your team. Build with your agents.

Bring people, projects, AI agents, context, and execution into one shared workspace.

Start BuildingSee plans