What it is
AEGIS Evidence Desk is a human-in-the-loop professional service for teams deploying AI agents into real workflows. It converts an agent workflow into an evidence package that can be inspected, challenged, and replayed instead of asking a customer to trust a black-box automation.
A customer supplies a workflow, intended tools, data boundaries, and allowed consequences. Evidence Desk produces a bounded delivery package: capability and authority map, policy/admission checks, stale-state and replay tests, action/effect receipt structure, risk register, and an explicit ledger of what is established versus not established.
Why this is a business
The paid unit is a fixed-scope Verified Agent Workflow Audit for small businesses and professional-service teams adopting agentic automation. The service is designed to reduce the practical trust gap between a useful prototype and a workflow a customer is willing to let act on business systems.
Revenue, users, testimonials, and production execution will only be reported when supported by actual records. This project page does not treat planned integrations or internal tests as customer evidence.
XPRIZE build boundary
This is a new commercial application created during the Build with Gemini XPRIZE period. It is built on top of pre-existing AEGIS Omega open-source governance research and runtime components. Those pre-existing components are disclosed rather than presented as newly created for the competition.
The competition-specific build is the customer-facing business workflow: intake -> AI-assisted workflow decomposition -> bounded authority/evidence plan -> reproducible audit bundle -> customer delivery. The intended XPRIZE deployment adds Gemini for at least one production LLM step and Google Cloud for the deployed service. Those competition-specific integrations must be demonstrably live before final submission and will not be claimed before that evidence exists.
Evidence discipline
AEGIS Evidence Desk separates:
- identity from authority;
- capability from permission;
- observation from state-changing authority;
- a signed receipt from proof that a real-world effect occurred;
- private development chronology from public prior art;
- provenance evidence from execution and truth authority.
The goal is useful AI autonomy with bounded consequences and independently inspectable evidence, not an unsupported claim of universal safety or certification.
Existing technical foundation
The pre-existing AEGIS Omega repository includes TypeScript, Rust, and Python runtime surfaces; deterministic authority/admission concepts; execution identity and workspace binding; replay rejection; receipts; and fail-closed governance experiments. The XPRIZE project will disclose exactly which pieces are reused and which customer-facing Gemini/Google Cloud/business components were built during the competition window.
Current state
Pre-submission commercial build. No revenue or customer count is claimed on this page until evidenced.
Built With
- aegis-omega
- python
- rust
- typescript
