Loading...
← Other blogs

AI Tools

Amp usage: check the balance without mixing units

The CLI, user settings and workspace settings report different pools. Read the unit before the number.

• 2 min read

Amp usage is easiest to read if you stop looking for one number. The account reports agent spending in dollars, Orb time in hours and workspace credits as their own pool. Each answers a different question, and a single total would answer none of them correctly. UsageAtlas documents how it handles those pools in its Amp guide.

Check the balance first

Amp's pricing documentation names three places to look: the amp usage command, user settings, and workspace settings. Start there, because the provider's own report is the one that matters for whether work can continue.

Question Where to look
How much agent budget is left? amp usage or user settings
How many Orb hours remain? The same surfaces, as a separate unit
What does the workspace hold? Workspace settings
What did one thread cost? That thread's details

Why the units matter

Dollars and hours are not convertible without an assumption. If you spend an hour on Orb, the dollar cost depends on what ran. A tracker that shows one combined figure has picked an exchange rate on your behalf and hidden it.

The UsageAtlas integration makes the opposite choice: it keeps Amp's reported pools separate and shows each as its own counter.

Adding a local view

Connecting an access token gives a desktop view of the same reported figures.

  1. Create a token in Amp's settings.
  2. In UsageAtlas, open Settings, then Sources, find Amp and expand Connect.
  3. Paste the token, choose Save and check, then open Limits.

The app can also read AMP_API_KEY from the environment, and needs a restart after that value changes. Credentials are stored with the operating system's encryption and excluded from cloud saves.

What this view cannot do

It reports allowances and balances. It does not import per-request tokens, models, projects or cost history, so it will not tell you which thread burned the budget. For that, use Amp's thread details.

Holding a capacity view and an activity view separately is normal. The mistake is expecting one to do both.

The unknown-versus-zero rule

A capacity meter appears only when the provider reports enough detail to draw one. When it does not, the figure is unknown.

That rule deserves repeating because the failure it prevents is subtle. A team that reads unknown as zero stops work early. A team that reads it as unlimited overspends. The only safe display is a blank with a reason.

A weekly check that takes a minute

Open Limits, note the agent balance and the Orb hours, then compare with last week. If the dollar balance fell faster than the work justified, look at thread details rather than at the total. One long agent run explains most surprises.

FAQ

Does UsageAtlas need my Amp password?

No. It uses an access token created in Amp's settings, or an environment variable if you prefer.

Why is there no capacity meter for a pool?

Because the provider did not report enough information. A missing limit is unknown, not zero.

Can I see per-request tokens for Amp?

Not through this integration. It reads account allowances and balances rather than request-level history.

Originally published on claudegpt.si