Context
The right information, with a boundary.
Minimal, approved, attributable sources in the exact namespace. Context versions make the inputs inspectable.
THE EVENTUS GOVERNANCE PLATFORM
Governance for
autonomous AI.
Eventus Engine is a domain-neutral governance platform in development for AI agents, tools, and workflows.
Its architecture connects the context a run may use, the process it follows, and the authorization an action requires. Build around your existing agents and orchestration.
In development. Local reference checks exist. Production readiness has not been established. View status
THE ARCHITECTURE
Between an intention
and its consequences.
Agents propose work. Orchestrators coordinate it. Eventus defines the governance contracts around context, review, execution, and learning.
YOUR SYSTEMApplications & eventsDomain logic and user intent
EVENTUS ENGINE · INTENDED GOVERNANCE SCOPEContext → Process → Review → Action → FeedbackExact namespace · Versioned contracts · Attributable decisions
YOUR RUNTIMEOrchestrator & agentsModels, tools, and adapters
Integration boundary diagram. Local reference components cover part of this architecture.
Understand the boundariesnamespace_id: demo.reviewEXACT SCOPE- INPUT BOUNDARYApproved context
Attributable sources · Pinned version
- PROCESS BOUNDARYPropose an action
Versioned workflow · Artifact reference
- ◇AUTHORIZATION BOUNDARYHuman review
Run + action + scope + artifact
- EXECUTION BOUNDARYApproved execution
Idempotency · Bounds · Audit record
Measure outcomes. Propose updates. Review again.
Architecture requirements. This website does not execute workflows or enforce these controls.
SIX PILLARS
Govern the whole loop.
Full governance is the architectural goal. Each pillar names a requirement for a controlled system, with implementation status tracked separately.
Process
A versioned path through the work.
Explicit inputs, outputs, capabilities, resource limits, retries, and recovery behavior.
Knowledge
Evidence before reuse.
Provenance, applicability, and reuse permissions travel with a proposal. Promotion requires review.
Quality Control
A decision before an effect.
Validation and human judgment tied to the run, action, scope, and exact artifact under review.
Triggers
A controlled way to begin.
Authenticated, allowlisted, validated, deduplicated events invoke named workflows.
Feedback Loops
Outcomes that remain attributable.
Measurements link back to the run, versions, artifacts, and approvals that produced them.
A CONCRETE WORKFLOW
A proposed change.
A deliberate decision.
Illustrative example: an agent proposes a configuration change. The reviewed artifact and the requested effect must stay bound together.
Make the change inspectable.
A versioned workflow reads approved context in demo.review and produces a proposed configuration artifact with a content hash.
Authorize the exact effect.
A human reviews the artifact. Authorization identifies this run, the configuration action, its scope, and the reviewed content hash.
Connect the result to its cause.
The intended executor applies bounds and idempotency, records the outcome, and submits any learned update for a new review.
An architectural walkthrough. Production external execution is not validated.
Inspect the review boundarySTART WITH WHAT YOU CAN INSPECT
Read the contracts.
See the boundaries.
Explore the current fields and a pure local reference example, then check the evidence behind the development status.