01 / Governance
Every agent call goes through a gateway
Route LLM and MCP traffic through managed APIs so rate limits, guardrails, cost controls and audit trails apply the same way to every model and tool.
AI governance · Identity · Gateways
Architecture for AI platforms that enterprises can actually trust.
Patterns, reference builds and field notes on governing AI agents: who they act for, what they can reach, and how every call is authorized, metered and audited.
Focus areas
02 / Identity & Auth
Agents act with scoped, delegated identity
OAuth 2.1, token exchange and fine-grained scopes so an agent only ever holds the access its user granted, for as long as the task needs it.
03 / Reference builds
Patterns you can run, not just read
Working sample projects built on open-source API management and identity tooling, with the configuration and trade-offs written down.
Reference flow
The agent never talks to a tool directly.
An identity provider issues short-lived, scoped tokens on the user's behalf. The gateway validates them, applies policy, and only then forwards the call to the MCP server or model.
Identity provider
OAuth 2.1 · token exchange
Client
AI agent / IDE
Gateway
Authorize · rate-limit · audit
Upstream
MCP servers & LLMs
Coming soon
Write-ups and runnable sample projects on MCP governance and agent authorization.