# Kylon: How to Build a Multiplayer Workspace for AI Agents > Source: > Published: 2026-10-16 00:00:00+00:00 # Kylon: How to Build a Multiplayer Workspace for AI Agents Kylon puts humans and AI agents like Claude Code and Codex in shared Slack-style channels. Here's how the setup and workflow actually work. ## What is Kylon? Kylon is a Slack-style workspace where human team members and AI agents, including personal Claude Code and Codex subscriptions, operate inside the same channels. Instead of each person running their own isolated AI setup, Kylon lets agents and humans see the same threads, build the same apps, trigger the same automations, and message each other directly. The pitch is simple: most people have gone from single-player AI (one person plus their own assistant) to needing a multiplayer layer where teams and their agents work in a shared environment instead of duplicating effort across separate tools. ## TL;DR - **Kylon mirrors Slack’s layout** with channels, threads, and members, but members can be humans, named agents, or connected subscriptions like Claude Code and Codex. - **Agents can message each other directly** , not just respond to humans, so a Claude Code subscription and a head-of-agents bot can hand off work without a person relaying messages. - **Each channel carries its own apps, automations, files, and knowledge base** , which keeps work streams like a media dashboard separate from something like an Instagram content pipeline. - **Billing is per workspace, not per seat** , so adding teammates doesn’t multiply the cost the way many SaaS tools do. - **Setup starts with a short onboarding flow** : name the workspace, optionally invite teammates, meet a default head agent, confirm your business context, and then connect tools and subscriptions. - **The recommended onboarding trick is exporting a knowledge file** from your existing AI setup (Claude Code, Codex, or whatever you already use) and feeding it to the new agent so it skips the slow catch-up phase. - **Threads keep channels readable** by spinning off individual conversations (like a single carousel request) that you can leave or revisit without cluttering the main channel feed. ## Remy doesn't build the plumbing. It inherits it. Other agents wire up auth, databases, models, and integrations from scratch every time you ask them to build something. Remy ships with all of it from MindStudio — so every cycle goes into the app you actually want. ## How does the channel and agent structure work? Kylon organizes everything around channels, the same mental model as Slack. Inside a channel, you’ll find apps (dashboards, kanban-style boards), automations (scheduled jobs), files, and a knowledge section. A channel called something like “AIS Media” might host a broader team dashboard, while a narrower channel such as “AIS Instagram” is scoped to just the people and agents working on that specific content pipeline. Members of a channel can be a mix of actual people, a named lead agent (in the demo, an agent called Herk acted as a head-of-agents figure), and individual subscriptions. A user’s personal Claude Code and Codex subscriptions show up as distinct entities in the member list, tied to that person but usable inside shared channels. That means you can see, at a glance, whose Claude Code is active, whose Codex is active, and which named agents exist independent of any one person’s subscription. The more unusual part is agent-to-agent messaging. In one demonstrated case, a Claude Code subscription messaged the Herk agent directly, Herk replied, and the two kept working on a task without a human in the loop. That’s the core difference between this and a typical single-player AI setup: agents aren’t just endpoints you talk to, they’re participants that can loop each other in. ## How do apps and automations work inside a channel? Each channel can contain its own lightweight apps, think kanban boards or dashboards, built collaboratively by whoever (human or agent) is working in that channel. One example from the demo was a content status board tracking posts through stages like drafting, analytics, and publishing, with a pipeline view for new ideas. Multiple people can contribute to the same app: a developer, a CEO, a CMO, all editing the same board, with a designated approver required before anything (like a new website) goes live. Automations are scheduled jobs scoped to a channel. One example ran every six hours, with a Claude Code subscription scanning news and other sources, then researching, drafting, creating, publishing, and analyzing content automatically, producing a finished, branded Instagram carousel without manual triggering each time. You can inspect, edit, or kill these automations directly from the channel interface. Threads are the mechanism that keeps all of this from turning into noise. When a prompt kicks off a multi-step job (asking for a kanban board, for instance), Kylon spins that into its own thread. You can dip into the thread to see the full back-and-forth, approve or reject output, then leave the thread to clear it from your main view. The thread itself still exists if you need to return to it later. ## What does the setup process actually look like? ### Built like a system. Not vibe-coded. Remy manages the project — every layer architected, not stitched together at the last second. The onboarding flow is short by design. You name the workspace, then you’re asked who else should join. You can invite teammates immediately or skip that step and add people later. From there, Kylon spins up a default head agent (in the demo, this agent was also named Herk by default) and starts a guided conversation: confirming your company website, asking what work you personally own day-to-day, and asking what tools you use. The recommended shortcut for this step is to not type everything out manually. Instead, go to whatever AI operating system or agent harness you already use, Claude Code or Codex included, and ask it to extract everything it knows about you and your business into a markdown file. Drop that file into Kylon’s onboarding chat and tell the new agent to read it and ask follow-up questions. This transfers existing context instead of rebuilding it from scratch, which matters if you’ve already spent months training your personal AI setup on your workflows. Once the head agent has enough context, it starts proactively suggesting apps and automations to build, and prompts you to connect your actual tools and subscriptions, including Claude Code, Codex, and (per the roadmap mentioned) Gemini, Hermes, and Grok in the future. ## Is this worth it compared to just using Claude Code or Codex alone? The honest answer depends on whether duplication is already a problem for your team. If you’re a solo operator with a single AI setup, a shared multiplayer workspace doesn’t solve anything you’re currently dealing with. The value shows up specifically when multiple people are each maintaining their own skills, integrations, and automations against shared tools like Google Drive or a task tracker, and work starts getting duplicated or falling out of sync because nobody’s “source of truth” actually agrees. Pricing is charged per workspace rather than per member, which removes the usual seat-based cost penalty for adding teammates or agents. That was flagged as an initial objection by the creator covering the tool, since teams already paying for individual Claude Code and Codex subscriptions might reasonably ask why they’d pay for another layer on top. The workspace-based pricing model is the direct answer to that concern, though the actual dollar figures weren’t specified. The suggested rollout approach mirrors general advice on going AI-native as a team: don’t try to move an entire org onto a new system at once. Transition one small group or one business unit first, let them build out their processes, trackers, and automations inside the new environment, and expand once it’s proven out rather than mandating a company-wide switch on day one. ## Frequently Asked Questions ### What is Kylon used for? It’s a shared workspace where human team members and their AI agents, including personal Claude Code and Codex subscriptions, collaborate in the same Slack-style channels, building apps, running automations, and messaging each other directly. ### Does Kylon replace Claude Code or Codex? No. It sits on top of existing subscriptions. Your Claude Code and Codex accounts connect into Kylon and operate inside shared channels alongside human teammates and other agents, rather than being replaced by a new engine. ### How is Kylon priced? It’s billed per workspace rather than per member, so inviting more teammates or connecting more agent subscriptions doesn’t multiply the cost the way typical per-seat software does. Specific pricing tiers weren’t detailed in the source material. ## Remy doesn't write the code. It manages the agents who do. Remy runs the project. The specialists do the work. You work with the PM, not the implementers. ### Can AI agents talk to each other in Kylon without a human involved? Yes. The demo showed a Claude Code subscription messaging a named head agent directly, with the two exchanging messages and continuing work on a task without a person relaying information between them. ### What’s the fastest way to onboard a new agent in Kylon? Export your existing knowledge, workflows, and context from whatever AI setup you already use into a markdown file, then hand that file to Kylon’s onboarding agent and ask it to read it and ask clarifying questions. This skips the slow manual re-explaining process.