Architecture Hub

High-level flow: You deploy an agent. It communicates strictly through governed model and tool gateways, emitting a trace.

Zelkor compiles your intent into a distributed system that sandboxes the agent you already wrote. The agent cannot break out, reach unauthorized data or networks, its prompts are verified, budget controlled, and it is under observation; tenants stay isolated.

This hub maps the hops and boundaries that enforce that sandbox.

---
config:
  theme: neutral
---
flowchart LR
  laptop[Your laptop / Client]
  subgraph cluster [Your cluster]
    model[Model]
    tools[Tools]
    agent[Your agent]
    trace[Run trace]
  end
  laptop -->|run| agent
  agent --> model
  agent --> tools
  agent --> trace

Deep Dives

Each page answers one specific architectural question about how traffic flows and where the trust boundaries lie:

Page	Question
Request Path	How a client call reaches the model
Network Boundaries	Who may talk to whom inside the cluster
Sandbox Isolation	Where generated code executes
MCP Governance	How a tool call is isolated and authenticated
Run Trace	What one run looks like in Langfuse
North-South Exposure	What is published to the internet versus ClusterIP
Envoy Graph Routing	How Envoy routes incoming calls to the correct agent deployment
Drop-In Agent Contract	How Zelkor sandboxes your agent (Intercept, Wrap, MCP)
Tenant Isolation	How one tenant identity is applied on a run
Agent Datastores	Stateful infrastructure (Postgres, Qdrant) and BYO options