Kylon: Bringing Your AI Agents and Your Team Together
What Kylon actually does
Kylon is a work chat app where AI agents and human teammates share rooms and discussions. The channel had been struggling with this exact problem. Everyone on the team had their own agent, like Hermes, but there was no clean way for agents and people to talk, share documents, or work on threads together. They tried Discord first. It was too gamer-focused and confused the non-technical team members. They also tried monday.com and ClickUp, but those got expensive and didn't have agent-native AI.
Kylon arrived at the right time. It lets you create private rooms for specific projects or departments, add your own AI agents, and invite only the humans who need to be there. The channel's own setup now includes Ultron and Kylo Ren, two agents that handle different tasks and can talk to each other directly through MCP.
The invoice workflow that convinced us
The clearest test was an invoice for Buddy Bites, a deal for the Doc Doc HK side project. The team set up a private accounting room with Kylo Ren, an agent trained on previous invoice samples, company bank details, and styling. Martin gave Kylo the deal information, and Kylo generated the invoice.
But the team didn't let Kylo send it alone. They looped in their accountant, who also does company secretary work. She doesn't want to learn AI and doesn't need to. She just reads the invoice in the chat, checks the details, and approves it. The AI did the creation work, but a human still verified it.
This is the main advantage over Discord or a plain chat app. Non-technical staff can participate by reading messages and approving documents without touching a model dropdown or learning Codex. The person setting up the room picks the agent and handles the prompts. Everyone else just uses a normal chat interface.
Bringing your own agents and the MCP surprise
Kylon has its own cloud-hosted agents, but you can also bring existing agents like Hermes. The setup is simple: copy a prompt to your agent and it links up. The channel did this with Ultron, their super agent, and connected Ultron to a GitHub repository for the Box Mining AI site. Ultron now understands the site's articles, benchmarks, and structure.
Then they added both Ultron and Kylo to the same room and asked Ultron to brief the group on what it had been doing. The two agents communicated with each other via MCP instead of posting long chat messages. They called what they needed, shared responsibilities, and stopped. On Discord, the channel tried the same thing and it created huge, messy threads. Kylon is built for agent-to-agent communication, so it stays clean.
This is the part the channel called a game changer. Before, each team member solo-tanked their own project because there was no easy way for different agents to coordinate. Kylon became the glue.
Connections and cost
Kylon has built-in connections for Gmail, GitHub, Google Analytics, Notion, ClickUp, and MCP servers. Giving an agent access to Google Analytics means it can pull data about what visitors browse on the Box Money AI site, which helps decide what videos to make and articles to write. Setting this up for Hermes was harder; on Kylon it's mostly click and done.
Cost starts at $100 per month with 1,000 usage credits. That's not free, but the channel compares it to ClickUp and monday.com, which charged per seat and added up fast with external accountants and occasional users. Kylon doesn't bill per seat. You can bring your own agents with your own API keys and save on credit costs. For a startup team that already has agents, the channel sees this as a reasonable trade, especially since it replaces a separate project management bill.
Bottom line
Kylon is a work-focused chat app that finally lets AI agents and non-technical teammates work in the same rooms. The invoice workflow and MCP communication are the standout wins. The $100 monthly cost is real, but it replaces other per-seat tools and lets you bring your own agents. For a small team already running agents, it's worth a serious look.