# AI access and tokens (https://docs.akollo.com/en/ai/access-and-tokens)



## Who can use AI [#who-can-use-ai]

An admin decides who in the organisation can use the AI features: the assistant, questions about your data, AI help inside screens, and AI apps connected through MCP. The choice is made on **Settings › AI › Access & usage**.

| Option                    | Who can use AI                                                        | Seats needed |
| ------------------------- | --------------------------------------------------------------------- | ------------ |
| **Off**                   | Nobody. AI is switched off completely and no request reaches a model. | —            |
| **Admins only** (default) | Organisation admins and owners.                                       | No           |
| **Admins and managers**   | Admins, plus anyone who manages a team.                               | No           |
| **Everyone**              | Every member of the organisation.                                     | No           |
| **People with a seat**    | Only the people an admin gives a seat.                                | Yes          |

<Mermaid
  title="How Akollo decides whether a person can use AI"
  chart="`flowchart TD
A[Person opens an AI feature] --> B{Emergency stop on?}
B -->|Yes| X[No request reaches a model]
B -->|No| C{Option is Off?}
C -->|Yes| X
C -->|No| D{&#x22;Person is in the chosen group or has a seat?&#x22;}
D -->|No| N[Short message: ask an admin]
D -->|Yes| E{Monthly limit reached?}
E -->|Yes| P[Paused until next month]
E -->|No| F[&#x22;AI answers within the person's own permissions&#x22;]`"
/>

Someone outside the chosen group sees a short message that they don't have access and can ask an admin. People still see and use only what their own permissions allow. AI never widens access. During a support session, AI answers can be read, but nothing can be sent or changed.

Answers the product computes by itself, such as a count it already holds, continue when model calls are off. A weekly summary that an administrator sends is not limited by this choice. The same choice covers the assistant, the short actions on a page, the connected tools, search in ordinary language, and writing help.

### Choosing who can use AI [#choosing-who-can-use-ai]

<Steps>
  <Step>
    ### Open the settings [#open-the-settings]

    Go to **Settings › AI › Access & usage**.
  </Step>

  <Step>
    ### Pick one option [#pick-one-option]

    Select **Off**, **Admins only**, **Admins and managers**, **Everyone** or **People with a seat**.
  </Step>

  <Step>
    ### Give seats if needed [#give-seats-if-needed]

    If you picked **People with a seat**, give a seat to each person who should use AI from the list on the same page.
  </Step>
</Steps>

## Seats [#seats]

With **People with a seat** selected, give a seat to a person from the list on the same page, or remove it. Seats only count while that option is selected.

## Monthly limit per person [#monthly-limit-per-person]

You can set the most AI tokens one person may use in a calendar month. Leave the field empty for no limit. When a person reaches the limit, their AI features pause until the next month.

## AI tokens and usage [#ai-tokens-and-usage]

The page shows this month's AI tokens: how many are used, how many are left and, at the current pace, how many will be used by the end of the month and the day they would run out. Below it, usage is listed by person, by automation, by feature and by model.

<BarChart
  title="AI token use by feature in one month"
  caption="Illustrative example, not real data. It shows how a month's use against the included ten million tokens might be split by feature."
  unit=" M"
  data="[
  { label: 'Assistant', value: 4.2 },
  { label: 'Search in plain language', value: 1.1 },
  { label: 'Writing help', value: 1.6 },
  { label: 'Planning and staffing', value: 1.3 },
  { label: 'Thread summaries', value: 0.6 },
]"
/>

Managed models draw on a monthly allowance of AI tokens. The included allowance is ten million tokens each calendar month. Packs added later are used after that allowance, oldest first. Administrators are told once when most of the month is used, and once when it is used up. A heavier request then uses the ordinary model. When nothing remains, the request stops. A model the institution runs, and a connection the organisation brings, do not spend the allowance.

<Mermaid
  title="Where the tokens for a request come from"
  chart="`flowchart LR
A[AI request] --> B{&#x22;Own connection or model the institution runs?&#x22;}
B -->|Yes| Z[Allowance not spent]
B -->|No| C{&#x22;Monthly allowance left?&#x22;}
C -->|Yes| D[&#x22;Uses the ten million monthly allowance&#x22;]
C -->|No| E{&#x22;Pack tokens left?&#x22;}
E -->|Yes| F[Uses packs, oldest first]
E -->|No| G[Request stops]`"
/>

<Screenshot src="/screens/en/settings-usage.webp" alt="The Usage page with the AI input tokens meter selected, a monthly budget field and a daily usage range" caption="Admin › Usage: metered usage day by day, including the AI input tokens meter." />

<Callout type="idea" title="Tip">
  Watch the forecast on **Access & usage**. If the day the tokens would run out falls before the end of the month, request a token pack in time.
</Callout>

## Token packs [#token-packs]

If the month's tokens are not enough, request a token pack of 1, 5, 10 or 50 million tokens, with an optional note. The request is recorded, and the pack is added to your organisation's balance once it is processed.

<Steps>
  <Step>
    ### Open the token section [#open-the-token-section]

    Go to **Settings › AI › Access & usage**.
  </Step>

  <Step>
    ### Request a pack [#request-a-pack]

    Choose **Request a token pack**, pick 1, 5, 10 or 50 million tokens and add a note if you like.
  </Step>

  <Step>
    ### Wait for processing [#wait-for-processing]

    The request is recorded. Once it is processed, the pack appears in your organisation's balance and is used after the monthly allowance.
  </Step>
</Steps>

## Emergency stop [#emergency-stop]

The **Emergency stop** card on the same page stops every request of the AI features in your organisation at once, whatever option is selected above. When the reason is resolved, turn AI back on from the same place. While AI is off or stopped, no request of the AI features reaches a model. Screenshot analysis in activity tracking has its own switch in the activity settings.

<Callout type="warn" title="Warning">
  The emergency stop does not pause screenshot analysis in activity tracking. That has its own switch in the activity settings.
</Callout>

## Who can change these settings [#who-can-change-these-settings]

Organisation owners and admins. Every change is recorded with who made it and when.

## FAQ [#faq]

<Accordions type="single">
  <Accordion title="What is the default access option?">
    **Admins only**: organisation admins and owners can use AI until an admin picks another option.
  </Accordion>

  <Accordion title="Can AI show someone more than their permissions allow?">
    No. People still see and use only what their own permissions allow. AI never widens access.
  </Accordion>

  <Accordion title="How many AI tokens are included each month?">
    Ten million tokens each calendar month for managed models. Packs you add are used after that allowance, oldest first.
  </Accordion>

  <Accordion title="What happens when the tokens run out?">
    Administrators are told once when most of the month is used and once when it is used up. A heavier request then uses the ordinary model. When nothing remains, the request stops. Answers the product computes by itself continue.
  </Accordion>

  <Accordion title="Does our own model connection use the allowance?">
    No. A model the institution runs, and a connection the organisation brings, do not spend the allowance.
  </Accordion>

  <Accordion title="What happens when a person reaches their monthly limit?">
    Their AI features pause until the next calendar month. Leave the limit field empty if you do not want a limit.
  </Accordion>
</Accordions>

## Related pages [#related-pages]

* [The organisation's own connection](/en/ai/connections)
* [Running the model in the institution](/en/ai/on-prem)
* [AI assistant](/en/ai/assistant)