Interactive ToolReady to use
Architecture Decision Aid
An interactive decision aid for framing the decision, weighting the forces that matter, comparing up to three options, and exposing low-confidence assumptions.
Problem: Structure consequential architecture decisions without reducing them to product preference, committee opinion, or an unexplained scoring table.
Useful for: Architects, engineering leaders, platform teams, and senior engineers comparing viable technical approaches.
Evidence available: Transparent weighted scoring with explicit confidence markers; the tool deliberately warns that hard constraints and known trade-offs can override the numeric ranking.
Interactive ToolReady to use
AI Platform Readiness Assessment
A 21-question readiness assessment across identity, platform interfaces, governance, evaluation, observability, cost, and security.
Problem: Understand whether an engineering platform has the foundations to support AI agents safely alongside human developers.
Useful for: Platform leaders, enterprise architects, engineering leaders, and teams planning AI-assisted or agentic software delivery.
Evidence available: Transparent scoring model with no hidden weighting; runs entirely in the browser and links back to the architecture guide and downloadable checklist.
Open-source LibraryPublished package
aeg-intent-gate
A small TypeScript approval gate that turns model tool calls into governed intents and exposes approved, blocked, or approval-required outcomes before execution.
Problem: Keep AI-generated tool requests separate from the authority to execute consequential actions.
Useful for: Agent builders, AI platform engineers, and architects implementing tool-use or human approval workflows.
Evidence available: Executable examples, package tests, a live demo, and AEG Lab scenarios exercise the separation between proposed model intent and runtime authority.
Reference ImplementationOpen source
agent-firewall
A focused CLI implementation of execution-control thinking: proposed commands cross a policy boundary before they are allowed to reach the shell.
Problem: Apply an explicit control boundary before an AI agent or automation executes shell commands.
Useful for: Engineers experimenting with agent execution, command policies, and practical runtime enforcement.
Evidence available: Source, tests, changelog, and releasing documentation are public so the execution boundary can be inspected rather than taken on trust.
Reference ArchitecturePublic reference
AEG Reference Architecture
The public reference repository for AEG, including diagrams, examples, and practical artifacts that complement the canonical model and technical paper.
Problem: Provide reusable architecture artifacts for separating autonomous intent from execution authority.
Useful for: Architects and senior engineers evaluating agentic systems, tool execution, governance, and auditability.
Evidence available: The model is paired with an implementation package and an explicit experiment track rather than being presented as a purely conceptual pattern.
ExperimentPhase A + Phase B v2 measured
AEG Lab #001
A reproducible baseline-versus-governed experiment combining deterministic boundary tests with a published 360-trial real-model evidence run.
Problem: Test whether a governance boundary prevents unsafe execution when model or agent proposals are malformed, escalated, replayed, forged, or otherwise adversarial.
Useful for: Practitioners who want to inspect the methodology and limitations behind a governance claim instead of relying on diagrams alone.
Evidence available: Phase B v2 reports 360/360 exact model and governance expectation passes, all 180 legitimate executions preserved, and 0 governed unsafe executions across 180 non-execution trials. The published limitations keep this scoped as integration evidence, not proof of safety.