Introducing Titan
Titan sits between agents and the systems they act on. Every call gets a verified identity, a fresh policy decision and a short-lived grant, and any of it can be stopped.
Agents are starting to act on real systems: opening issues, posting messages, editing spreadsheets. Titan is how you let them, without handing over the keys or losing track of what happened.
Titan is in private preview with design partners.
Decided on every call
Each request is checked against your policy at the moment it is made. Rules can look at the workspace, environment, agent, capability, resource, destination and time of day. The answer is one of three: allow, require approval, or deny.
People where it matters
Actions that need a person wait for one. Approval profiles set who can approve, how many approvals are needed, and that the requester cannot approve their own action.
Grants, not keys
When an action is allowed, Titan issues a signed grant that lasts at most 30 seconds and is bound to the exact arguments. Tool Gateway checks the grant again, adds the credential and makes the call. The agent never holds a key.
Stop anything
A kill switch works at nine levels, from a single session to the whole organization, and is checked before the decision, before the grant and before dispatch. Every step is written to a tamper-evident ledger you can verify.