Your usage
The Your LLM usage (via Harriet proxy) card on My AI shows what your AI use costs — totals attributed to you across all your devices.
What the card shows
Three numbers, always for you personally:
| Figure | Meaning |
|---|---|
| Estimated cost | The USD cost of your LLM calls, based on the real upstream provider and model each call ran on. |
| Input tokens | Tokens sent to models in your requests: your prompts plus the context the AI reads. |
| Output tokens | Tokens the models generated in response. |
What counts, and what doesn't
Only calls that go through the Harriet LLM proxy are counted — that's why the card title says "via Harriet proxy". The proxy applies when Harriet Desktop or Claude Desktop (no Claude account) is enabled on your devices. If you also use Claude Desktop signed in with your own Claude account, that traffic goes straight to Anthropic and doesn't appear here.
The Model and LLM proxy card above the usage card, and the Proxy column in Your devices, tell you which of your devices route through the proxy.
Why usage is visible
Because your devices are linked to your account, every proxied call can be attributed to a person. That's what lets your organization see where AI spend actually goes (by person, team, and device) instead of one opaque bill. Your card is the personal slice of that: the same transparency the organization gets, applied to you.
Who else can see it
As the card says, "Organization admins see the same in LLM Usage reports." Admins with usage access see per-person, per-team, and per-device breakdowns in LLM usage analytics, and can set budgets on that spend. What's recorded is cost and token counts (the usage figures on this card), not a transcript of your conversations.
If My AI shows a plain Usage card instead of your totals, your account doesn't have the permission that unlocks detailed figures: "Detailed LLM usage totals are available to users with the provisioner audit permission. Ask an admin if you need a usage report."