AI Usage & Cost

EM Kit tracks how much AI you use and turns it into an estimated cost, so you can keep an eye on spend and cap it with a hard monthly budget. Everything lives under Settings → AI usage & cost and is scoped to the current workspace.

Because EM Copilot runs through a local CLI — Claude Code or Codex CLI — that authenticates with your own account, the dollar figures are estimates derived from public per-token list prices. If you are on a flat subscription, treat them as a usage signal rather than an exact invoice, and adjust the rates to match your plan.

How usage is tracked

Every AI action in EM Kit records its token usage when it completes — there is nothing to switch on. This covers:

• Copilot chat — conversations in the assistant drawer
• Workspace insights — generated manager insights
• Member briefs — the “prep me for a 1:1” synthesis
• Meeting summaries — transcribed meeting summaries
• Automations — scheduled and manual automation runs

Tracking is forward-only: usage is recorded from the moment you update to this version, so the overview starts empty and fills in as you work.

Note

Claude Code reports token counts (including cached input), so its calls are fully priced. Codex CLI does not report token usage, so those calls are counted but excluded from the cost estimate. When this happens, the overview shows a “some usage isn’t priced” notice.

The overview

Use the range selector in the top-right to switch between This month, Last 30 days, and All time. The page then shows:

• Estimated cost for the selected range, plus total calls, input tokens (with how many were cached), and output tokens.
• A cost-by-day chart for the selected range.
• A breakdown table you can view by feature, by model, or by provider. Sort any column to find what is driving spend.

Estimated rates

Cost is calculated from token usage and an editable table of per-model rates, in US dollars per million tokens (MTok). Each model family has three rates: standard input, cached input (cheaper reads of repeated context), and output.

The built-in defaults are:

Model family	Input ($/MTok)	Cached input ($/MTok)	Output ($/MTok)
Claude Opus	15.00	1.50	75.00
Claude Sonnet	3.00	0.30	15.00
Claude Haiku	0.80	0.08	4.00
OpenAI GPT-5	1.25	0.125	10.00

To change a rate:

1. Open Settings → AI usage & cost and scroll to Estimated rates.
2. Edit any input, cached-input, or output value.
3. Click Save rates.

Cost is always derived from the current rates, so editing them recalculates your entire history — there are no figures frozen at the time of each call. Click Reset to defaults to discard your overrides and return to the built-in prices.

Monthly budget (hard limit)

Set a budget to put a hard ceiling on AI spend for the workspace.

1. Open Settings → AI usage & cost and find Monthly budget.
2. Enter a dollar amount under Hard limit and click Save budget.

Once a budget is set, EM Kit shows a progress bar of estimated month-to-date spend against the limit, and warns you when you pass 80%.

Caution

When estimated spend for the current calendar month reaches the limit, EM Kit blocks new AI calls — Copilot chat, insights, briefs, meeting summaries, and automations all stop until you act. In-progress requests are unaffected; only new ones are blocked.

To resume, either raise the limit or click Remove limit. The budget resets automatically at the start of each calendar month.

Troubleshooting

”Monthly AI budget reached”

You have hit your hard limit for the current month. Open Settings → AI usage & cost and either raise the Hard limit or click Remove limit. The limit resets on the first of the next month.

The cost looks too high or too low

The estimate is only as accurate as the rates. Open Estimated rates and adjust them to match your actual plan, or Reset to defaults. Remember that cached input is billed at a fraction of standard input, which is why heavy Copilot use can stay cheaper than the raw token counts suggest.

Some usage shows no cost

Calls on an unrecognized model — or made through a provider that doesn’t report token counts, such as Codex CLI — are counted but cannot be priced, so they appear with a $0.00 estimate and trigger the “some usage isn’t priced” notice.