How It Works
Gvner is the decision layer between agent intent and production execution. Agents propose actions, Gvner evaluates policy and authority, then returns ALLOW or DENY with immutable evidence.
Canonical flow
1. Agent generates intent
The agent emits a structured intent with actor, action, target, and context. See Agent Inventory and Tool Policy.
2. Intent sent to Gvner
The runtime calls Gvner before side effects happen: use /adapters/http/intent for generic integrations, or /openclaw/intent and /temporal/intent on compatibility paths. Execution is blocked until a signed decision is returned.
3. Gvner evaluates
Gvner checks policy, identity authority, risk, budget, and compliance constraints. Related docs: Policies, Role Mapping, Identity Changes, Control Mapping.
4. Gvner returns ALLOW or DENY
The decision is deterministic, time-bound, and tied to policy basis. Every result is written to evidence and can be exported.
5. Execution only proceeds if allowed
ALLOW permits execution for the defined scope and time window. DENY blocks execution and records reason codes for review.
Integration surfaces
Unified adapter (recommended)
Use one contract for LangGraph, Semantic Kernel, Bedrock, Vertex ADK, automation, and worker queues.
OpenClaw compatibility
Existing OpenClaw runtimes keep dedicated compatibility endpoints.
Temporal compatibility
Existing Temporal workflows keep dedicated activity-gating endpoint.
Full integration hub
All integration families, payload mappings, fail-closed behavior, and wrapper patterns.
What gets evaluated
Policy constraints
Deterministic rule checks on action type, scope, and environment.
Identity authority
SSO/SCIM, role mapping, session policy, and approval requirements.
Operational limits
SLA budgets, readiness state, and incident conditions.
Security/compliance controls
Control coverage, least privilege, and fail-closed behavior.
Governance OS layers
Execution authority
Explicit ALLOW/DENY decisions gate all execution.
Identity governance
SSO, SCIM, roles, sessions, and change approvals are policy-bound.
Evidence exports
Ledger, packet, receipt, and regulator exports with verification.
Operational oversight
Readiness, observability, failover, restore drills, and SLA tracking.
Example flow
Below is a representative example showing the intent → decision → audit cycle.
Operate this in production
Observe
Start with observability, performance targets, and SLO burn tracking.
Harden
Use readiness checks, maintenance controls, and restart advisories.
Recover
Run support packs, launch packs, failover rehearsal, and restore drills.
Scale
Use org-level dashboards for cross-tenant incidents, errors, and SLA.
Key guarantees
Call to Action
Use Gvner to move from agent experimentation to controlled production execution.