> ## Documentation Index
> Fetch the complete documentation index at: https://docs.cobalt.peoplereign.io/llms.txt
> Use this file to discover all available pages before exploring further.

# FAQ

> Cross-cutting answers about billing, identity, channels, integrations, and the platform.

# FAQ

Answers to the questions that span more than one feature. For questions about a
specific integration or channel, see that page's own FAQ.

## Billing & usage

<AccordionGroup>
  <Accordion title="What is a credit?">
    The unit Cobalt meters usage in. One credit covers one end-user turn — an
    employee asks, the agent works and answers. Building, configuring, and
    testing in [Preview](/get-started/create-your-first-agent) never cost
    credits; only real employee conversations on a live channel do. See
    [Billing & usage](/operate/billing-and-usage).
  </Accordion>

  <Accordion title="Is there a free trial?">
    Yes — every new tenant starts with **100 free credits** and no card
    required. See [Sign up & start free](/get-started/sign-up).
  </Accordion>

  <Accordion title="What happens when we run out of credits?">
    Agents pause and employees get a brief "temporarily unavailable" reply until
    you top up. Any top-up reactivates the tenant immediately. Turn on
    **auto-renew** to avoid pausing at all — Cobalt warns before the balance runs
    out and retries a failed renewal over several days.
  </Accordion>
</AccordionGroup>

## Identity & access

<AccordionGroup>
  <Accordion title="Do end users have to sign in to talk to an agent?">
    Not necessarily. On channels where they're already authenticated (Slack,
    Teams, an embedded widget in an app they're logged into) they're identified
    automatically — no second sign-in. A public web widget can run
    **anonymously**. Whether an agent can *act as* a user depends on how strongly
    that identity is verified — see
    [End-user identity](/identity/end-user-identity).
  </Accordion>

  <Accordion title="What's the difference between admins and end users?">
    **Admins** build and run agents in the console; their access is governed by
    [roles](/identity/roles-and-team) and, optionally, [SSO](/identity/sso).
    **End users** are the employees who talk to the agents; they're managed as
    [end-user identity](/identity/end-user-identity) and provisioned via
    [SCIM](/identity/scim). They never touch the console.
  </Accordion>

  <Accordion title="Can one person work in more than one tenant?">
    Yes — the same account can belong to several tenants and switch between them.
    Each tenant has its own agents, billing, and data. See
    [Working across tenants](/operate/working-across-tenants).
  </Accordion>
</AccordionGroup>

## Channels

<AccordionGroup>
  <Accordion title="Which channels can I put an agent on today?">
    The [web widget](/channels/web-widget) (including
    [SharePoint Online](/channels/sharepoint)), [Slack](/channels/slack), and
    [Microsoft Teams](/channels/teams). Email and SMS are coming. See the
    [channels overview](/channels/overview).
  </Accordion>

  <Accordion title="Do all channels support the same features?">
    The [web widget](/channels/web-widget) is the richest surface — inline
    citations, source cards, media, voice, forms, feedback. Slack and Teams
    render what their platforms allow (for example, images inline but not
    interactive walkthroughs); Slack and Teams bots are **direct-message only**.
    Where a channel can't render something, the agent degrades gracefully rather
    than dropping content — citations become a "Sources" list, for instance.
  </Accordion>

  <Accordion title="Does voice cost extra?">
    No — a spoken turn is metered exactly like a typed one (one credit). See
    [Voice](/channels/voice).
  </Accordion>
</AccordionGroup>

## Integrations

<AccordionGroup>
  <Accordion title="What does 'Experimental' mean on an integration?">
    It's built and usable, but not yet verified against a live vendor instance.
    Activating one asks you to acknowledge that first. Try it in Preview before
    relying on it. See the [Integration catalog](/integrations/catalog).
  </Accordion>

  <Accordion title="The system we use isn't in the catalog. What now?">
    If it speaks the Model Context Protocol, connect it with the
    [MCP Server](/integrations/mcp-server) integration. Otherwise, the catalog
    grows regularly — check the console, or ask us.
  </Accordion>

  <Accordion title="Will an agent see things a user shouldn't?">
    No. Read tools run **as the asking user**, so the agent only sees what that
    person can already see. Indexed content is permission-checked against the
    source at query time. Privileged actions run under admin-configured policy
    and are audited.
  </Accordion>
</AccordionGroup>

## Platform

<AccordionGroup>
  <Accordion title="Does Cobalt have a public API or webhooks?">
    Not today. Cobalt is administered through the console; the machine-to-machine
    surfaces a customer touches are [SCIM](/identity/scim) for user provisioning
    and the inbound webhooks the channel platforms (Slack, Teams) call. There is
    no public REST API or outbound event webhook to build against yet.
  </Accordion>

  <Accordion title="Can I connect Cobalt to Claude or another MCP host as a server?">
    Not yet — Cobalt **consumes** MCP servers as integrations, but doesn't expose
    an agent as an MCP server for external hosts to call.
  </Accordion>

  <Accordion title="How is our data kept separate from other customers'?">
    Each tenant is isolated — its agents, conversations, identities, and billing
    are its own, and credentials and secrets are encrypted at rest. For specifics
    on data handling, security, and compliance, talk to us.
  </Accordion>
</AccordionGroup>