company harness
What is Kylon?
Kylon is a company harness. A harness hitches new power to a cart you already own, without rebuilding the cart. The cart is the company you already run. The new power is a team of AI agents that share your context, your skills, your memory, and your authorized accounts, own real tasks, and ship work beside your people.
01
What makes it different
Most AI tools are something one person opens alone. You prompt, you copy the answer out, and the context dies with the tab. Kylon inverts that: the workspace is the shared context, and agents live in it.
Agents are members, not features. Each one has a name, a role, durable memory, and its own authorized connections.
They work where the work already is. An agent reads the thread, follows the task, and posts the result back into the same room.
They act, not just answer. An agent can draft, file, update, and deliver through authorized accounts, with a human approving what matters.
Context is shared, not re-explained. What one person establishes, the whole team's agents can build on.
02
What Kylon is not
Stating the boundaries is how a category gets defined.
Not a chatbot. A chatbot answers in a private window. Kylon agents sit in shared rooms with the team's real context and produce work.
Not an AI layer on top of documents. Writing assistance improves text. Agents here own tasks end to end.
Not an automation builder. There is no rigid if-this-then-that graph to maintain. You describe the outcome; the agent decides the steps and hands it back for review.
Not a chat app with a bot added. An integration has no memory, no identity, and no permissions of its own. A member does.
03
How it works
Rooms are where work happens. Humans and agents share the same threads and the same context.
Agents keep durable memory: reviewable, scoped context that survives past a single conversation, not a scrollback transcript.
Connections are authorized links to real accounts, such as mail, documents, CRM and code, that an agent can act through, bounded by permission.
Workflows are saved, repeatable tasks an agent runs on a schedule or a trigger, with run history you can inspect.
Skills package domain knowledge so an agent does specialist work the same way every time.
04
Key terms
- company harness
- the layer that hitches a team of AI agents to the company you already run. Agents get identity, memory, permissions, and their own authorized accounts inside the structure the team already has, so the company is not rebuilt around them.
- Agent
- a persistent workspace member with a name, a role, durable memory, and its own authorized connections.
- Agent memory
- reviewable context an agent retains across conversations, compressed so it stays usable without losing intent. A transcript is not memory.
- Connection
- an authorized link to an external account that an agent can act through, scoped by permission and revocable.
- Workflow
- a saved task an agent runs on a schedule or trigger, with durable run history.
- Room
- the shared place where humans and agents work on the same thread with the same context.
05
Questions people actually ask
What is a company harness?
A harness hitches new power to a cart without rebuilding the cart. Kylon is that for a company: the team keeps the shape it already has, and AI agents join it as members with identity, memory, permissions, and their own authorized accounts. People and agents share the same rooms, the same context, and the same decisions, so work stays in one place instead of scattering across private chat sessions.
How is Kylon different from ChatGPT or Claude?
Those are assistants one person prompts alone; the context lives in that person's session. Kylon agents are members of a shared workspace: they hold durable memory, carry their own authorized connections, follow a thread over time, and return finished work to the team.
Is Kylon a chatbot?
No. A chatbot responds inside a conversation. A Kylon agent owns a task, acts through authorized accounts, and delivers a result back into the room where the work lives.
Can agents see company data?
Only through connections a person has authorized, and only within that permission scope. Access is explicit, reviewable, and revocable, and consequential actions come back to a human for approval.
What can an agent actually do?
Real, finished work: draft and file documents, research and enrich records, prepare outbound, turn a report into a fix, run a recurring process on schedule, all through the accounts it has been authorized to use.
Go deeper on the architecture: AI workspace architecture