Architecture
StevenOS: one system, 9 repos, five layers.
Every project in this portfolio rolls up into a single Personal AI Operating System. This map shows how they connect — from infrastructure plumbing to the public proof a visitor can inspect.
Public Proof
The top layer visitors see — proof-chain site, profile, demos.
Evaluation
Quality gates that decide whether agents deserve more trust.
Agent Runtime
Where agents actually work — messaging, tools, skills, gateway.
Knowledge & Identity
What agents know, how they behave, and where knowledge lives.
Infrastructure
Tool and API plumbing — model routing, CLI compat, MCP servers.
How data flows through the stack
Knowledge → Runtime: The digital-twin identity files and knowledge-harness vault provide context to agents before they act.
Runtime → Evaluation: Agent work produces traces. The scorecard inspects those traces for tool use, verification, and durable artifacts.
Evaluation → Public Proof: Passing checks produce receipts, build-log entries, and shipped-receipt pages that visitors can verify independently.
Infrastructure → Runtime: CLIProxyAPI routes model requests. Manage-up provides manager-facing MCP skills. Both feed into the agent runtime.
What is not shown
Production work systems, internal tooling, private vault content, and employer-specific code are intentionally excluded. The map shows only public-safe repos and contributions.