Skip to content

Usage and Server health

What your workspaces are spending on their agents, where the money actually goes, and the live health of the DevBench server.

Where
UsageIn the app (sign-in required), opens in a new tab.
Reading time
2 min read
Who
Owner

How it works

Usage is what your workspaces spend, collected from each workspace every six hours so the figure survives the box being deleted. You can look at 30 days, 90 days or 12 months. Both agents are counted: Claude Code from its transcripts, Codex from its session records, each at its provider's list prices.

The most useful thing on the page is where the money goes. Cache reads, the context an agent re-reads on every turn, are usually most of the bill. Long sessions and large contexts are what move that number, not how much the agent writes. Spend is also broken down by model, by workspace and by day, and by what the agents were doing, which is what tells you whether the money went into work or into orientation.

The page is honest about its own accuracy. Rows written before a counting bug was fixed are labelled estimated, not counted, scaled down by a measured factor, and shown alongside what they read as stored, because they cannot be recounted: Claude keeps its records inside the workspace, and those boxes are gone. Spend collected since the fix is counted.

Server health (it was called Monitor) is the live health of the DevBench server itself, host resources, Docker, and app activity, for when the question is "is the platform healthy" rather than "what did it cost". It is in the organization's menu, beside API keys, because it is about the installation rather than the work.

Scheduled-run minutes (they were called CI minutes) are the minutes of the boxes scheduled agent runs used, measured against the plan. Lathe's CI is not metered.

Steps

Following along? Tick each number off as you go.

  1. Open Usage from Insights in the top bar (or the search: Ctrl Ctrl, ⌘ ⌘ on a Mac).

  2. Pick a window: 30 days, 90 days or 12 months.

  3. Read Where the money goes first.

    If cache reads dominate, the lever is session length and context size.

  4. Look at By workspace to find the workspaces that cost the most.

  5. Check What the agents were doing to see whether the spend went into work or orientation.

  6. Open Server health from the organization's menu when you want the server's own health rather than the bill.

    Refresh re-reads it.

Worth knowing

  • Figures marked estimated are a central estimate, not an exact number, the per-box correction factor ranged widely.

Read next

Running the organizationOrganization settingsThe defaults every workspace inherits: what its container is built with, the board's columns, the git flow, the agent and accounts boxes run on, and the standing instructions every agent reads.Running the organizationTeam, roles and permissionsWho is in the organization, what each role reaches, who may spend money in the cloud, and how new people join.
Report a problem