Loom AI Labs

live Governed actions

Account Resolver

A support workflow that separates model interpretation from validation, confirmation and application-owned account operations.

Account Resolver live demo interface
Live application capture · verified 2026-07-25
Status
live
Last verified
2026-07-25
AI Fabric
0.4.0
Data
Synthetic

Question

Hypothesis

Can confirmation stay useful when user-provided parameters are imperfect, while trusted resource identifiers remain protected?

What this demonstrates

  • Validation warnings can be shown without silently discarding user-confirmed values.
  • Trusted resource identifiers remain blocking when provenance is not established.
  • The final operation is performed by application code.

What this does not demonstrate

  • Real billing-provider integration
  • Unattended financial operations
  • Suitability for high-risk account changes

Architecture and data flow

  1. 01 Synthetic source data
  2. 02 Host application policy
  3. 03 AI Fabric capability
  4. 04 Observable UI result

Guided steps

  1. 01

    Open a synthetic account scenario.

  2. 02

    Ask for a supported account change in natural language.

  3. 03

    Review validation evidence and confirm the proposed operation.

Observable evidence

The running application and its health surface are the public evidence for this experiment. The source repository exposes the implementation path behind the visible result.

Security and data boundary

The application owns subscription state, validation policy, trusted identifiers and execution. The model proposes structured intent; it does not bypass those checks.

All account, subscription and billing records are synthetic.

Known limitations

  • The supported action catalog is deliberately narrow.
  • No external payment provider is contacted.

Technologies and products used

  • Spring Boot
  • AI Fabric 0.4.0
  • Governed actions
  • OpenAI

Related research

The investigation behind this proof.

Actions and governance Reproducible demo

A Governed Lifecycle for AI-Proposed Actions

This investigation follows an action from model interpretation through validation, warning, confirmation and application execution. It distinguishes advisory validation from trusted target checks.

Artifact
Reference implementation
Maturity
Demo verified
Read investigation

Connect

Inspect the live system and its source together.

The experiment is useful when the visible result, application boundary and implementation can all be questioned.