> ## Documentation Index
> Fetch the complete documentation index at: https://docs.getvector.me/llms.txt
> Use this file to discover all available pages before exploring further.
# AI usage and cost
> See what your AI work has cost, who or what spent it, what a single call cost, and how spend is kept separate per workspace.
Vector runs on your own [AI key](/getting-started/connect-a-provider), so every research run, comparison and monitoring check is billed to you by your provider, not by Vector. That makes one question practical: what did this actually cost?
Open **Settings** and go to the **AI Usage** tab. **Token usage & cost** is the whole record.
## See what you've spent
The tiles across the top add up the whole log:
* **Total cost**, and **AI cost** — the part spent on AI rather than fetching pages
* **AI calls**, input and output tokens, and prompt-cache tokens written and read
* **Web searches** and page **Fetches**
* **Web scrapers** — how many paid fetches the [web scrapers](/ai/web-scrapers) made, and what they cost
## Find out who spent it
**Who spent it** splits the last 30 days of spend in two:
* **Automatic (scheduled)** — runs nobody started by hand: the [monitoring](/monitoring/set-up-monitoring) schedule, and the checks it catches up on after downtime.
* **You & your team** — everything started by a person.
Each side shows its cost, how many calls it made, and the features that spent the most. If the automatic side is higher than you'd like, check the competitors you watch least closely less often.
## See what a single call cost
Below the totals, the log has a line for each of the most recent 100 AI calls: **When**, the **Feature** that made it — competitor search, product description, a pricing scan — the **Model** that ran it, tokens in and out, cache reads and writes, web searches, and its **cost**.
Fetches by a web scraper appear in the same log, so fetching a page and reasoning about it read side by side.
## Check whether a cheaper model saves money
Because every line names its model and feature, the log is how you measure [routing](/ai/model-per-function): move a heavy function to a cheaper model, and that feature's lines get cheaper from its next run.
## How a call's cost is worked out
A call's cost is its real cost, priced at the rate of the model and provider that actually ran it: input tokens, output tokens, prompt-cache writes and reads, plus the provider's fee when the call used web search. Token and cache prices come from Vector's **AI models** dataset, so the figures track what providers actually charge.
The log is a record after the fact: Vector doesn't cap spending or alert you when it grows.
## Keep spend separate per company
Each [workspace](/account-and-team/workspaces) has its own key and its own record. Work in one company's workspace is never billed to another's key.
Guests can't spend your budget at all: AI actions are blocked for them before any provider is called. See [Members and roles](/account-and-team/members-and-roles).
## Clear the log
**Clear usage log** empties the record, after asking you to confirm. It's a log of what happened, not a bill: clearing it changes nothing about what your provider charges you.
## Where to go next
The main lever on what you spend.
The other cost recorded here.