Every view. A real diagram for each.
Not marketing boxes — these diagrams are drawn from the actual deployed system: the control plane, the bot-node fleet, the data stores, the connector runtime, and the security controls that are really running. Jump to any view.
System architecture
The layered runtime and how components own their boundaries.
View diagram ->Production & deployment
The real container topology, data stores, tunnel, and secrets.
View diagram ->Runtime & system routing
How a request travels from the edge to a bot and back.
View diagram ->Integration architecture
Connector runtime, the token broker, and external systems.
View diagram ->Workflow flows
The ticket lifecycle and the build / ticket / video pipelines.
View diagram ->Security & scanning
What is enforced, and what the Security Center scans.
View diagram ->Business processes
Trading, operations, and root-cause flows on their own pages.
View diagrams ->Components and how work flows through them.
Work flows down from a surface to the controller, fans out to bot nodes, reaches outside systems only through brokered connectors, and lands in isolated data. Results flow back up.
The same diagram appears on the Platform page alongside the eight-layer breakdown.
The real running topology.
This is the actual deployment: a Cloudflare tunnel for ingress, one control-plane container, a fleet of bot-node containers each consuming its own Redis stream, five data stores, plus Vault for secrets and Headscale for remote nodes.
Drawn from the live container set: one oshal-local-api control plane and ~26 bot-node containers, each on its own Redis stream. Today this runs on Docker Compose; the same images target Kubernetes.
How one request travels the system.
From the edge, through the auth gate, onto the mesh, to a bot that runs the model, and back — with cost captured on the way out.
Each bot consumes only its own stream, so lanes scale independently. Token and dollar cost is captured once, centrally, on the way out.
How a bot safely reaches an outside system.
A bot never holds a raw credential. A tool call passes a per-agent auth gate, runs on the shared connector runtime, and pulls a scoped token from the broker before it touches an external API.
Live signed connectors (commerce, travel, media, finance) are a curated subset; the marketplace catalog is the broader, importable set. More on Platform → Connectors.
The ticket lifecycle.
Every unit of work is a ticket on a queue. Build, ticket, and video pipelines are first-class shapes of this same flow.
The build / ticket / video pipelines are on Platform → Workflows.
Two layers: what's enforced, and what's watched.
Preventive controls stop a bad action before it happens; the Security Center is the detective layer that scans the platform itself and triages findings into tickets.
Tenant RLS and least-privilege DB roles are shipped and enforced. The Security Center and fail-closed execution are built and rolling out; some findings are open work, tracked as tickets. See Platform → Security.
The application-level flows.
Each flagship application has its own process-flow diagram on its page.
Trading desk flow
Signal → Risk → Execute → Ledger → Recap, every fill tied to its signal.
View on Trading ->Operations loop & RCA
Monitor → Gather → Diagnose → Heal → Report, plus the prep → box-in → orchestrate engine.
View on Operations ->Build / ticket / video
The three first-class workflow shapes and the workflow generator.
View on Platform ->Want a walk-through?
Happy to go through any layer in depth — the real deployment, the routing, the isolation, or the security model.