What Forseti is
Forseti sits at the action boundary. An agent submits a typed intent instead of touching a privileged tool directly. Forseti evaluates policy, asks humans when needed, releases approved execution, and preserves evidence about the decision.
Core objects
- Intent envelope: the structured request the agent submits.
- Policy decision: allow, deny, or requires approval.
- Approval route: single approver, quorum, hierarchy, or other configured tree.
- Evidence bundle: trace, votes, callbacks, model/tool context, and replayable records.
Good first pilot
Pick one workflow where an agent wants to touch production, money, customer communication, or a privileged system. The pilot should prove that the organization can answer: what did the agent intend, what did policy decide, who approved it, what executed, and what evidence remains afterward?
Current status
Forseti works best when the first rollout is narrow. Start with one action type, one policy, one approval route, and one execution callback. Expand only after the evidence bundle is useful during review.
Trust boundary
Forseti should sit between agents and privileged systems. Agents ask for a typed action; they should not receive direct production credentials just because the action might be approved. Policy and approval decisions must be recorded before execution release.
How it fits Caimeo
- Brainstack remembers what happened and turns useful evidence into proposals.
- Forseti decides whether a proposed action is allowed, denied, or needs approval.
- Tyche can replay or simulate action workflows before they reach production.
- Valhalla can use Forseti as the governed execution boundary for delivery rooms.