Geometra's strongest business wedge is not general-purpose frontend replacement. It is regulated operational software where humans and AI agents share a workflow and the business must prove what the agent saw, what it could do, what it clicked, and what changed afterward.
- Claims operations leaders who want AI-assisted review without losing auditability.
- Compliance teams that need structured proof for escalations, approvals, and evidence handling.
- Security and access-governance teams approving privileged actions.
- Platform teams building internal agent workstations for regulated workflows.
Claims and compliance workflows are UI-bound: evidence, status, risk, approvals, and exceptions all appear together on one screen. Backend APIs can prove a mutation happened, but they cannot prove what was visible and actionable when an agent chose that mutation.
Geometra can package that proof in the UI protocol:
- Exact semantic geometry for every relevant node.
- Stable ids for agent targets like
approve-payout. - Risk classes and human confirmation requirements on actions.
- Trace events for request, approval, completion, denial, and failure.
- Replay frames showing before/after UI state.
- Postconditions attached to the action contract.
- A claim enters the review queue.
- The agent calls
/inspectand receives exact semantic geometry and action contracts. - The agent chooses
approve-payout,request-evidence, orescalate-claim. - Gateway policy checks risk and requires approval when needed.
- A human approves or denies the pending action.
- The gateway executes the workflow action.
- Replay stores the frame-before, policy decision, approval, output, and frame-after.
Browser automation infers claims workflows from DOM or screenshots. Geometra-native workstations expose the claims workflow as an auditable UI protocol.
demos/agent-native-opsshows the human workstation and the agent-visible semantic geometry side by side.demo/replay-viewerrenders the generated replay as a visual audit packet for the public demo build.bun run demo:agent-native:httpruns the external-agent HTTP flow and writesexamples/replays/claims-review.json.packages/gateway/openapi.jsondescribes the inspect/actions/trace/replay API surface.bun run benchmark:agent-native:livecompares a live Geometra-native flow with browser-inference-style automation.bun run create:app -- ./claims-compliance --template claims-compliancescaffolds a vertical gateway starter.benchmarks/agent-native-methodology.mdexplains the modeled benchmark assumptions.
Prioritize features that make this vertical easier to sell:
- Replay retention, export, and redaction policies.
- Approval webhooks and integrations with existing workflow systems.
- Tenant/session auth examples.
- Audit packet viewer and downloaded replay artifacts.
- Starter templates for claims/compliance queues with queue filters, evidence panels, and risk-score policies.
- Live benchmarks against equivalent HTML/Playwright workflows.