---
title: Architecture
description: The compute-fabric view at
---

# Architecture



## Purpose

ARCHITECTURE shows which GPU providers and workers are up and which node is leading routing right now. It is the read-only fabric view; actions on pods live on [Compute](/docs/console/compute).

Deep link: `ops.useargus.co/#/architecture` (nav group COMPUTE).

## What it shows

Grouped under two headings, "Providers & workers" and "Router / leader":

* PROVIDERS table: provider, healthy, starting, capacity (cams), load, fps, and dollars per hour burn.
* WORKERS table (the current leader is named in the heading and starred in its row): id, provider (with a spot marker), region, GPU model and count, cams (current/capacity), fps (current/capacity), queue depth, GPU utilization percent, status, dollars per hour, and last heartbeat.
* ASSIGNMENTS table: camera, tenant, the worker it routes to, reason, previous worker, and when it was assigned.
* A burn block summarizing provider cost.

Data source: `GET /architecture`.

## Actions

None. This view is read-only.

## Matching CLI

```bash
argusctl compute list
argusctl compute served
argusctl compute strain
```

See [argusctl compute](/docs/cli/compute).

## Notes

* Polled on the 15 second timer with in-place repaint; stale data is kept on a failed refresh.
* Empty states are explicit: "no providers registered", "no workers reporting", "no cameras assigned".
* In production today the assignments table reads empty until the camera-assignment loop is deployed and cameras are actually assigned to a worker; this is a deployment gap, not a fault in the view.


---

For a semantic overview of all documentation, see [/sitemap.md](/sitemap.md)

For an index of all available documentation, see [/llms.txt](/llms.txt)

For agent-facing discovery, including API and MCP surfaces, see [/agents.md](/agents.md)