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

# Approvals in chat

> Why Connie stops to ask before some actions in a chat, what each button on the approval card does, and how to change what Connie asks about.

Before Connie does something that costs credits, changes or deletes data, or leaves Connie (like sending an email), it pauses the chat and shows an **approval card**. Nothing happens until you answer. This keeps you in control of anything risky, expensive or visible to other people.

**Where:** in the chat thread, under Connie's latest message. The workspace rules that decide what needs approval live in **Tools** → open a tool → **Permission** · `https://connie.ai/<workspace-id>/tools`

## When Connie asks first

Unless your workspace has set a rule for that tool (see [below](#change-what-connie-asks-about)), Connie asks before:

* **Using a built-in tool that costs credits**, for example enrichment or Lead Finder. Once you allow a paid tool in a chat, Connie doesn't ask again for that tool in the same chat.
* **Deleting** whole workflows, lists or records.
* **Anything that leaves Connie:** sending an email or message, posting, or a change in another app that Connie can't confirm contacts nobody.
* **Running on more than 1,000 records** at once.
* **Unverified actions** that Connie can't check in advance.

Connie runs these **without asking**: reading data, free changes inside your workspace, writing values into your Records, and everyday changes in a connected app that don't contact anyone (for example updating a record in your CRM, adding a note, a task or a spreadsheet row). Actions in connected apps are never held just because of their price.

## Answer an approval card

<Steps>
  <Step title="Read the card">
    The card names the tool, says what it will do, and shows labels such as **Costs credits**, **Changes data**, **Updates records**, **Sends to someone outside Connie** or **Large run**. The credit badge shows the price, for example "1 credit" or "up to 60 credits". Click **Show breakdown** to see how the cost adds up.
  </Step>

  <Step title="Check what will change">
    For record updates, the card previews the changes ("Currently: …") and may offer **Show all N …**. Message drafts can be edited on the card; click **Save draft** to keep your edits.
  </Step>

  <Step title="Choose a button">
    The buttons depend on the kind of action (see the table below). To say no, click **Reject**. You'll see "You rejected this. Nothing was changed."
  </Step>
</Steps>

<Note>
  You can't send a new message while a card is waiting. The chat says "This chat is waiting for you to approve or reject the pending action." Answer the card first. If Connie is still writing its reply, the card says "Connie is replying. Decide when the reply finishes."
</Note>

### What the buttons do

| Card | Button | What happens |
| - | - | - |
| Paid tool | **Allow** | Runs it now, and allows this tool for the rest of this chat |
| Paid tool | ⌄ (**More options**) → **Allow in this chat** | Same as Allow |
| Paid tool | ⌄ → **Always allow** | Runs it now and sets the tool to **Always allow** for the whole workspace. Owners and admins only |
| Paid tool | ⌄ → **Disable this tool** | Turns the tool off for the whole workspace. Nothing runs ("You turned this tool off for the workspace. Nothing was run."). Owners and admins only |
| Delete | **Approve & delete** | Approves this one action only |
| Send | **Approve & send** | Approves this one action only |
| Anything else | **Approve** | Approves this one action only |
| Any | **Reject** | Cancels it. Nothing changes |
| Large run | **Run a sample first** | Runs on a few records first ("Run N of the M first, then decide on the rest") so you can check the result before paying for the rest |

When Connie needs several approvals at once, they're grouped: "N actions need your approval", with an **Estimated total**. Tick the ones you want and click **Approve selected (x of y)**, or use **Approve all** / **Reject all**.

### Other cards you may see

* **Questions:** "One quick question so I can get this right — answer below and I'll continue." Answer and click **Next** / **Submit**, or **Skip**.
* **Plans:** "Build this plan?" with **Build it** or **Change something**.
* **Enrichment quotes:** a **Quote** / **Estimate** table with "Nothing charged until you confirm". The cost of Connie's own reply is separate ("Copilot message costs are separate.").
* **Account needed:** "No \[app] account is connected. Connect one, then ask again." Click **Connect \[app]** to [connect the account](/tools/connect-accounts), then ask again.

## Change what Connie asks about

Workspace owners and admins can set a rule per tool. A rule you set always wins over the defaults above, and **Disabled** wins over everything.

<Steps>
  <Step title="Open the tool">
    Left sidebar → **Tools** → open the tool.
  </Step>

  <Step title="Choose a Permission">
    * **Always allow**: "Connie AI chat can run it without asking."
    * **Ask before running**: "Connie shows an Allow / Deny card before it runs."
    * **Disabled**: "Connie AI chat and MCP cannot run it."
  </Step>

  <Step title="Confirm">
    For a tool that writes, sends, deletes or costs credits, choosing **Always allow** asks "Let Connie run this without asking?". Click **Always allow** to confirm.
  </Step>
</Steps>

You can also do this straight from a card, with **Always allow** or **Disable this tool** in the ⌄ menu. Full details: [Tool permissions](/tools/permissions).

## Good to know

* **Who can approve:** anyone who can chat in the chat (owners, admins and members). Only **owners and admins** can choose **Always allow** or **Disable this tool**, or change a tool's Permission. Others see "Only workspace owners and admins can change this."
* **No deadline in a live chat.** A card in a chat you're using waits until you answer.
* **Scheduled tasks and follow-ups:** when a [scheduled task](/automations/scheduled-tasks) or a post-call follow-up needs approval, nobody is watching the chat, so Connie notifies you and the card has a deadline ("expires in …"). The task shows **Needs approval** until you answer. If time runs out: "This request expired before you answered. Nothing was changed." Click **Ask again** to retry.
* **Undo for record updates:** record-update cards may say "Undo: available for N days." Undo restores the record values only: "Workflow runs, messages and other effects the update triggered are not reversed."
* **Credits:** approving a paid tool charges its price on top of the cost of Connie's reply. If the tool fails, its credits are refunded. See [Credits](/billing/credits).
* **Agents and workflows** have their own approval steps. See [Approvals in automations](/automations/approvals).
* **Requests from Claude, ChatGPT and other AI apps** connected to Connie through MCP are approved on a separate page with **Approve** / **Reject**, and expire after 15 minutes. See [MCP overview](/mcp/overview).

## Troubleshooting

<AccordionGroup>
  <Accordion title="This chat is waiting for you to approve or reject the pending action.">
    **Why:** an approval card is open. **Fix:** scroll up to the card and choose **Approve** (or **Allow**) or **Reject**. Then send your next message.
  </Accordion>

  <Accordion title="I don't see Always allow or Disable this tool on the card">
    **Why:** only workspace owners and admins can change workspace rules. **Fix:** use **Allow** (for this chat) or ask an owner or admin to set the tool's Permission in **Tools**.
  </Accordion>

  <Accordion title="Connie keeps asking me to approve the same tool">
    **Why:** the tool is set to **Ask before running** (every call then gets its own **Approve** card), or the action is a send or delete, which is approved one action at a time. **Fix:** for a paid tool that asks only because of its price, click **Allow** to allow it for the rest of the chat. Otherwise an owner or admin can set the tool to **Always allow** in **Tools** → tool → **Permission**.
  </Accordion>

  <Accordion title="Connie ran something without asking">
    **Why:** it was free (reading data, a free change inside your workspace, or writing values into Records), or the tool is set to **Always allow**. **Fix:** an owner or admin can set the tool to **Ask before running** or **Disabled** in **Tools** → tool → **Permission**.
  </Accordion>

  <Accordion title="Connie says it can't use a tool">
    **Why:** the tool is set to **Disabled** for the workspace, or it needs an account you haven't connected. **Fix:** ask an owner or admin to change the tool's Permission, or [connect the account](/tools/connect-accounts).
  </Accordion>

  <Accordion title="This request expired before you answered. Nothing was changed.">
    **Why:** the approval came from a scheduled task or follow-up and its deadline passed. **Fix:** click **Ask again**.
  </Accordion>

  <Accordion title="My scheduled task shows Needs approval">
    **Why:** a run is waiting for your approval. **Fix:** open the chat (or your notification) and answer the card.
  </Accordion>
</AccordionGroup>

## Related

* [Tool permissions](/tools/permissions) — Always allow / Ask before running / Disabled for each tool
* [Chat](/chat/overview) — using Connie in a chat
* [Approvals in automations](/automations/approvals) — approvals for agents and workflows
* [Credits](/billing/credits) — what costs credits
* [Notifications](/notifications/overview) — where approval requests show up


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.