What Observability and Agent Tooling Does CoreWeave Documentation Cover?

CoreWeave's documentation covers two related but distinct tooling areas: Observability for monitoring infrastructure and workloads (Grafana dashboards, logs, metrics, telemetry, alerts), and agent tooling split across Mission Control (an agent plus MCP server for investigating and changing your environment), Agent Lens (agent observability and improvement), and ARIA (an AI research agent for analyzing experiments and building reports). If you need to watch cluster health, use Observability and Mission Control. If you need to build or improve AI agents and models, look at Forge, Agent Lens, and ARIA.

Observability: monitoring fleet hardware and workloads

The Observability section is described as covering "Grafana dashboards, logs, metrics, telemetry, and alerts for monitoring fleet hardware and workloads." In practice, this is the layer you reach for when you want to answer questions like is a node degraded, is a workload saturating its allocation, or did something change in the last hour?

The documented components break down as:

A concrete use case: you notice inference latency rising on a deployment. You open the relevant Grafana dashboard, confirm the metric trend, cross-check logs for errors, and set or adjust an alert so the next occurrence pages you instead of being discovered after the fact.

Mission Control: investigating and making supported changes

Mission Control is documented as "Mission Control Agent and the Mission Control MCP server for investigating your environment and making supported changes." Two things matter here:

• Mission Control Agent — the agent interface for investigating your environment.
• Mission Control MCP server — an MCP (Model Context Protocol) server, which lets MCP-compatible clients interact with Mission Control programmatically.

The phrase "making supported changes" is the key constraint: Mission Control is not described as an unrestricted control plane. Treat it as a way to investigate and apply changes that the documentation explicitly supports, and verify scope before relying on it for anything destructive.

Agent and model tooling under Forge

Forge is the section for developing and evaluating: "Track experiments, evaluate and improve agents and LLM applications, prototype in notebooks, and manage models across their lifecycle." Three named tools are relevant to agent observability and improvement.

Agent Lens

Documented as "Intelligent agent observability and improvement." This is the piece to look at when your question is specifically about how an agent is behaving — as opposed to how the underlying infrastructure is behaving. Observability tells you whether the GPU fleet is healthy; Agent Lens is aimed at the agent layer itself.

ARIA

ARIA is described as an "AI research agent that analyzes experiments, explains results, and builds visualizations and reports," and is marked Preview. The Preview label is worth noting: expect the interface and behavior to be less settled than generally available tooling, and check the docs for current limitations before building a workflow around it.

A fitting scenario: you've run a set of training or evaluation experiments and want a summary of what changed and why. ARIA is the documented tool for analyzing those experiments, explaining results, and producing visualizations and reports.

Weights & Biases alongside Forge

Forge also documents Weights & Biases: "W&B Models for experiment tracking, hyperparameter sweeps, artifacts, reports, and automations, with W&B Weave alongside for LLM and agent evaluation." If your need is experiment tracking and sweeps, W&B Models is the documented path; if your need is LLM and agent evaluation specifically, that's W&B Weave.

Choosing between them

Where to start

The documentation index groups these under two top-level areas: Infrastructure (Platform, Mission Control, Infra Control, Observability, Workloads, Access Control) and Forge (develop and evaluate). If your goal is monitoring and troubleshooting, start in Infrastructure → Observability, then read Mission Control if you need agent-assisted investigation. If your goal is building or improving agents and models, start in Forge and read Agent Lens, ARIA, and the Weights & Biases entries together, since they cover overlapping parts of the experiment-and-evaluation lifecycle.