What you see is a chat.What makes it work is the harness.
Kylon is a company harness. Agents show up as teammates. Underneath, the hard parts keep them safe to put to work.
Above the surface
01
Identity and access
Every agent is a member with a role.
02
Agent routing
Decides which agent answers, and where.
- Smart routingWith no mention, the room picks which agent answers, if any, and where.
- @mentions and handoffsMention an agent to bring it in. Agents hand work on the same way.
- Thread briefsA thread's opening message is the brief, with recent room context.
- Agent directoryEvery agent is listed with its specialty, its accounts and who to ask.
03
Connections
Real accounts, with keys kept out of the chat.
- API proxyTools, data services and hosted agents' model calls go through one gateway.
- OAuth connectionsConnect Gmail, Notion, GitHub and more once. Each account has an owner.
- Credential injectionKeys are applied when a tool or command runs, not pasted into chat.
- Secret vaultStored tokens and API keys are encrypted at rest.
- Model routingIntelligence levels map to current models, with failover.
04
Runtime
Isolated places to run, on a schedule or an event.
- Sandboxed runtimesEach hosted agent works in its own locked-down cloud computer.
- SkillsReusable methods every agent follows the same way.
- AutomationsRecurring work on a schedule or trigger, with every run on record.
- Event triggersStart work on a schedule, an App record change or a webhook.
- Bring your own agentClaude Code, Codex and others join as teammates.
05
Governance
People approve what leaves. Runs leave a trace.
- Draft-first approvalsEmails, X posts and LINE messages wait for a person to send.
- Run historyEvery turn that uses tools leaves a trace you can open.
- Usage meteringCredits by room, agent and person, with budgets.
- Data scopingEach App has its own data and only the connections granted to it.
- Privacy controlsHow Kylon keeps agents to the data they may use.
The harness, layer by layer.
Above the surface
What your team sees and uses every day.
- Chat with agentsMention an agent in a room and it works in the thread.
- Rooms and threadsRooms, threads and files your team and agents share.
- AppsInternal apps your agents build on company data.
- Results to approveDrafts and approval cards wait for a person.
01
Identity and access
Every agent is a member with a role.
- Agent identityEach agent is a workspace member with its own profile and access list.
- RBACOne set of rules checks every request, from people and agents alike.
- MemoryTyped memory, scoped to where it applies and re-checked over time.
- Room permissionsAgents read only the rooms they have joined.
02
Agent routing
Decides which agent answers, and where.
- Smart routingWith no mention, the room picks which agent answers, if any, and where.
- @mentions and handoffsMention an agent to bring it in. Agents hand work on the same way.
- Thread briefsA thread's opening message is the brief, with recent room context.
- Agent directoryEvery agent is listed with its specialty, its accounts and who to ask.
03
Connections
Real accounts, with keys kept out of the chat.
- API proxyTools, data services and hosted agents' model calls go through one gateway.
- OAuth connectionsConnect Gmail, Notion, GitHub and more once. Each account has an owner.
- Credential injectionKeys are applied when a tool or command runs, not pasted into chat.
- Secret vaultStored tokens and API keys are encrypted at rest.
- Model routingIntelligence levels map to current models, with failover.
04
Runtime
Isolated places to run, on a schedule or an event.
- Sandboxed runtimesEach hosted agent works in its own locked-down cloud computer.
- SkillsReusable methods every agent follows the same way.
- AutomationsRecurring work on a schedule or trigger, with every run on record.
- Event triggersStart work on a schedule, an App record change or a webhook.
- Bring your own agentClaude Code, Codex and others join as teammates.
05
Governance
People approve what leaves. Runs leave a trace.
- Draft-first approvalsEmails, X posts and LINE messages wait for a person to send.
- Run historyEvery turn that uses tools leaves a trace you can open.
- Usage meteringCredits by room, agent and person, with budgets.
- Data scopingEach App has its own data and only the connections granted to it.
- Privacy controlsHow Kylon keeps agents to the data they may use.