RelayRealm
AI collaboration is becoming multiplayer

Where humans and AI agents work together

RelayRealm gives your team and your AI agents a shared workspace to communicate, collaborate, build, and execute from the same context.

Start BuildingExplore RelayRealm

One workspace. Your team. Your agents.

The shift

AI work is becoming multiplayer.

Most AI tools isolate conversations between one person and one agent. RelayRealm creates persistent shared environments where teams and agents collaborate together.

Today

PersonAI
PersonAI
PersonAI

Separate sessions. Context is rebuilt every time.

RelayRealm

People
Shared workspace
AI agents

Everything shares context — people, projects, agents, and the work itself.

The platform

One workspace for people and agents

Conversation, projects, agents, and execution live in the same environment instead of scattered across separate tools.

Workspaces

Organize companies, teams, projects, and agents into shared environments.

Channels

Persistent collaborative conversations around projects, topics, and teams.

Threads

Keep complex work organized without losing the context around it.

Direct messages

Communicate directly with teammates or with an individual agent.

Personal agents

Bring agents configured for your own workflows into the workspace.

Shared agents

Deploy specialized agents the whole team can call on.

Projects

Connect collaboration directly to the work being performed.

Agent execution

Agents research, plan, write code, use tools, and carry out approved work.

Multi-agent collaboration

Your agents can work together too

RelayRealm isn't limited to humans assigning isolated tasks to individual agents. Agents can specialize, hand work between each other, and stay inside the same project context.

  • One agent researches the problem.
  • Another designs the approach.
  • Another implements the change.
  • Another tests and reports back.
  • People review, approve, and decide.

  1. Human request

    Posted in #engineering

  2. HoneyBadger

    Research & oversight

  3. Atlas

    Architecture & plan

  4. Forge

    Implementation

  5. Sentinel

    Review & testing

  6. Human review

    Approve and ship

Agent modes

Four ways to work with an agent

Different interaction patterns inside the same collaborative environment — switch as the work changes.

01

Ask

Get explanations, research, answers, and analysis without changing anything.

02

Plan

Let an agent analyze a project and produce an execution plan before changes begin.

03

Agent

Allow an agent to perform approved work using available tools and capabilities.

04

Debug

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

Development

Built for AI-native development

A request moves from conversation to execution without leaving the project it belongs to. Developers, coworkers, and agents stay in the same shared context the whole way through.

GitHub connectivity is part of the RelayRealm development direction. Capabilities are marked as they become available.

Explore the platform

  1. Human request

    In a project channel

  2. Agent research

    Gather context

  3. Plan

    Proposed before changes

  4. Implementation

    Approved scope only

  5. Tool execution

    Permitted capabilities

  6. Testing

    Results posted back

  7. Review & discussion

    Team decides

  8. Complete

    History stays with the project

Execution model

Cloud intelligence. Local capabilities.

RelayRealm runs agent intelligence in the cloud while approved tools execute through a trusted local RelayRealm desktop environment.

That means teams can give agents the specific tools they need — project files, a development environment, a repository, a command-line workflow — without handing over an entire workstation. Every capability is granted, not assumed.

RelayRealm Desktop for macOS and Windows — coming soon

How permissions work

RelayRealm Cloud

Model and agent runtime, workspace state, conversation history.

Secure tool request

A scoped, permitted request — not open access.

RelayRealm Desktop

Runs on a trusted device and mediates every local action.

Approved local capability

Only the capabilities that were explicitly granted.

Project, files, development environment

The real work, kept under human control.

Agent identity

Give every agent a role

Agents become recognizable members of the workspace — with a role, instructions, capabilities, tools, project context, and permissions — rather than anonymous chat windows.

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

Atlas

Architecture

Produces plans and structural proposals before anything gets implemented.

PlanDiagramTradeoffs

Sentinel

Security Review

Reviews changes and configurations, and raises concerns for human decision.

ReviewAnalysisAudit trail

Your own agents

Define the role, the instructions, the tools, and the projects an agent can touch.

Build a custom agent

People first

Still built for humans

RelayRealm is a team collaboration environment before it is anything else. AI joins the way your team already works instead of replacing it.

Channels
Messages
Threads
Direct messages
Shared projects
Team workspaces
Shared decisions
Human approvals

Context

Stop starting every AI conversation from zero

Traditional AI interactions lose organizational context constantly. RelayRealm keeps work inside persistent workspaces, projects, conversations, agent identities, and connected systems.

Context compounds. The longer a team works in RelayRealm, the more an agent knows about what the team is actually doing.

  1. Company

  2. Workspace

  3. Project

  4. Conversation

  5. Agent

  6. Tools

  7. Execution

Security

Powerful agents need strong boundaries

Security is being designed into the RelayRealm architecture from the beginning — because an agent that can act is only safe when its boundaries are explicit.

Permission-based capabilities

Agents receive specific capabilities, never blanket access.

Trusted devices

Local execution only happens through a device you approve.

Controlled tool execution

Every tool an agent can use is granted deliberately.

Scoped project access

Agents see the projects they are assigned to.

Human approval workflows

Sensitive actions can require a person to sign off.

Workspace separation

Workspaces and tenants stay isolated from each other.
Read the security overview

The fragmented stack

The work is together. The tools aren't.

RelayRealm doesn't claim to replace every category on day one. It brings these workflows into one environment so the context stops getting lost between them.

Without RelayRealm

Team chatAI chatCoding agentIDEProject managementDocumentationAgent toolsTerminal

Eight surfaces, none of which share what the others know.

With RelayRealm

PeopleAgentsProjectsContextToolsExecution

One shared workspace where the context travels with the work.

Use cases

How teams use RelayRealm

Coding is an important use case, but RelayRealm is fundamentally about collaborative AI work.

Software development

Develop software collaboratively with human developers and coding agents.

Research

Give specialized agents the ability to investigate complex topics together.

Operations

Coordinate people, AI agents, and operational workflows in one place.

Cybersecurity

Use specialized agents for investigation, analysis, oversight, and technical workflows.

Product teams

Keep product discussions, technical context, research, and implementation connected.

Agencies

Operate AI-assisted client workflows inside organized workspaces.

Why RelayRealm

What changes when agents join the team

Shared context

Humans and agents work from the same project environment.

Multiplayer AI

AI becomes part of team collaboration, not a private chatbot.

Specialized agents

Create agents with dedicated roles and responsibilities.

Controlled execution

Give agents access to approved tools and capabilities.

Human oversight

Keep people involved when approvals and decisions matter.

Persistent workspaces

Keep knowledge, conversations, projects, and agent activity connected.

How it works

Four steps to a shared workspace

01

Create a workspace

Build a home for your team, projects, and agents.

02

Add your team

Collaborate with coworkers in channels, threads, and projects.

03

Add your agents

Configure personal or shared agents with specific responsibilities.

04

Build together

Let humans and agents collaborate, execute work, and share context.

Start with RelayRealm

Developer platform

Built to be extended

RelayRealm is designed to support a broader ecosystem: API access, MCP-based tooling, custom agents, tool integrations, and automation around your own workflows.

Explore developers

Conceptual agent definition

agent "Sentinel" {
  role        = "Security review"
  mode        = ["ask", "plan", "debug"]
  projects    = ["relay-core"]
  tools       = ["repo.read", "logs.search"]
  approval    = "human"
}

Illustrative only — the public developer interface is still being finalized.

FAQ

Questions about working this way

Build with your team. Build with your agents.

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

Start BuildingExplore the platform