# Snitch AI: set up and query SnitchFeed by chat (https://snitchfeed.com/docs/features/snitch-ai)





**Snitch AI** is the chat assistant built into SnitchFeed. It reads your workspace (mentions, listeners, feeds, reports, webhooks and Brand Context) and can make changes for you after you approve them. It uses the same tools as the [SnitchFeed MCP server](/docs/mcp), so anything you can ask Claude or Cursor through MCP you can also ask here.

Open it from **Snitch AI** at the top of the sidebar, or go to `app.snitchfeed.com/home`.

## Start a conversation [#start-a-conversation]

<Steps>
  <Step>
    ### Open Snitch AI [#open-snitch-ai]

    Click **Snitch AI** in the sidebar. The page greets you with a composer and a row of prompt suggestions.
  </Step>

  <Step>
    ### Ask a question or pick a prompt [#ask-a-question-or-pick-a-prompt]

    Type your request and press **Enter** to send. **Shift+Enter** adds a new line. Clicking a prompt suggestion fills the composer with that prompt so you can edit it before sending.
  </Step>

  <Step>
    ### Approve any changes [#approve-any-changes]

    When Snitch AI wants to create, update or delete something, it pauses and shows an approval card. Nothing changes until you confirm. See [Approve changes](#approve-changes).
  </Step>
</Steps>

### Prompt suggestions [#prompt-suggestions]

| Suggestion                    | What it asks                                                                     |
| ----------------------------- | -------------------------------------------------------------------------------- |
| **Find opportunities**        | Your highest Fit Score opportunities from the last 7 days that are worth a reply |
| **Tune listeners**            | Which listeners are too noisy or missing matches, with query or platform changes |
| **Summarize mentions**        | Today's new mentions by platform, listener and Fit Score                         |
| **Set up alerts**             | How to send Slack alerts only for high fit Reddit and LinkedIn opportunities     |
| **Draft a reply**             | A reply to your top unread opportunity using your response profile               |
| **Improve fit scores**        | Why recent mentions score low and what to change in Brand Context                |
| **Spot weekly trends**        | Topics or keywords trending across your listeners this week                      |
| **Track competitor mentions** | Recent competitor mentions and how their sentiment compares to yours             |
| **Suggest new listeners**     | New listeners to set up, based on your Brand Context                             |

When your workspace has two listeners or fewer, a **Track Intent** suggestion appears first. It asks Snitch AI to interview you about what you want to find and then set up a listener.

### Create a listener from anywhere [#create-a-listener-from-anywhere]

The **+ Listener** button in the top bar (on pages such as Feeds, Listeners, Opportunities and Dashboard) opens Snitch AI with a listener setup prompt already typed into the composer. Snitch AI asks clarifying questions, tests queries, and then asks you to approve the new listener. To fill in the listener form yourself instead, see [Create a listener](/docs/listeners/create).

### Thinking level [#thinking-level]

The **Thinking** menu under the composer sets how much reasoning effort Snitch AI uses: **High** or **Max**.

## What Snitch AI can do [#what-snitch-ai-can-do]

* **Answer questions about your data**: list and show mentions, summarize feeds, run analytics queries and read saved reports, listeners, queries, feeds, webhook destinations, Brand Context and credit usage.
* **Build and test listeners**: create and edit listeners and their keyword queries, test a query on Reddit, X (Twitter), LinkedIn or Hacker News before saving it, look up LinkedIn industries and tracked profiles, and audit a query for noise.
* **Search platforms on demand**: search X, LinkedIn posts, Reddit posts and comments, and Hacker News for results that are not in your feeds yet.
* **Run LinkedIn research**: fetch a person's recent posts or comments, and scrape the comments or reactions on a post. These are the same jobs as [GTM tools](/docs/features/gtm-tools).
* **Manage feeds, alerts and reports**: create and update feeds, change a feed's notifications, create and update analytics reports, and manage webhook destinations.
* **Update Brand Context**: edit the description of your business that drives [Fit Score](/docs/fit-score).

Mentions that Snitch AI shows in the chat appear as cards; click one to view the mention's details. For the full tool list and parameters, see [MCP tools](/docs/mcp/tools).

## Approve changes [#approve-changes]

Reading data never needs approval. These actions always pause for your confirmation:

| Area          | Actions that need approval                                         |
| ------------- | ------------------------------------------------------------------ |
| Listeners     | Create, update or delete a listener                                |
| Queries       | Create, update or delete a keyword query                           |
| Feeds         | Create, update or delete a feed, and update a feed's notifications |
| Reports       | Create or update an analytics report                               |
| Webhooks      | Create, update or delete a webhook destination                     |
| Brand Context | Update Brand Context                                               |

Each approval card shows a heading such as &#x2A;*Create listener?** and the values Snitch AI wants to save. Then:

* Click the confirm button (**Create**, **Update** or **Delete**), or press **Cmd+Enter**.
* Click **Cancel**, or press **Ctrl+C**, to refuse. The change is not made, and Snitch AI tells you it did not proceed.

Delete actions are shown in red because they cannot be undone. When several changes are waiting, the cards stack above the composer; hover over the stack to see each one.

## Credits [#credits]

<Callout type="info" title="Usage based">
  Snitch AI has no fixed price per message. Each reply is charged in credits
  based on how much AI model usage it took, so a short answer costs less than a
  long, multi-step task.
</Callout>

* The composer shows the credits used so far in the current conversation.
* A reply you stop before it finishes is not charged.
* Tools that fetch fresh platform data, such as searching X or scraping LinkedIn comments, charge their own credit cost on top, the same as through MCP. See [Credits and usage](/docs/credits).
* Chat charges appear on **Workspace Settings → Credits Usage** as **AI agent chat message**.
* You need credits remaining to send a message. With none left, Snitch AI replies "You have no credits left. Purchase credits or upgrade your plan." See [Plans, billing and credit top-ups](/docs/workspace/plans-and-billing).

## Limits [#limits]

| Limit                | Value                                                 |
| -------------------- | ----------------------------------------------------- |
| Message length       | 16,000 characters per message                         |
| Messages per minute  | 10 per user                                           |
| Messages per hour    | 60 per user                                           |
| Messages per day     | 200 per user                                          |
| Conversation context | The latest 20 messages are sent with each new message |

If you hit a rate limit, Snitch AI shows "Too many chat messages. Please wait and try again."

A long task can time out. Snitch AI then shows "This took too long and stopped." with a **Pick up where we left off** button that asks it to continue from where it stopped. For very long investigations, start a new conversation with a narrower question.

## Conversation history [#conversation-history]

* **Private to you.** Conversations belong to your user in the current workspace. Teammates cannot see them.
* **Find past chats.** On the Snitch AI page, open the **History** panel and use **Search conversations**. You can also hover the history icon next to **Snitch AI** in the sidebar.
* **Rename or delete.** Each conversation gets a title automatically. Rename it or delete it from the history list, or rename it from the page header.
* **Start fresh.** Use **New chat** in the page header.
* **Retention.** Conversations with no activity for longer than your plan's data retention period are deleted: 3 months on Starter, 6 months on Pro.

## Use Snitch AI tools in your own AI client [#use-snitch-ai-tools-in-your-own-ai-client]

The link under the composer, **Connect to ... for better intelligence**, opens the MCP setup page. Connecting Claude, ChatGPT or Cursor gives that client the same tools against your workspace. See [Connect the MCP server](/docs/mcp/setup).

## Related [#related]

* [SnitchFeed MCP server](/docs/mcp): the same tools from Claude, Cursor or ChatGPT.
* [Create a listener](/docs/listeners/create): set up a listener with the form instead of chat.
* [How Fit Score, intent tags and sentiment work](/docs/fit-score): what Snitch AI means by fit and intent.
* [Credits and usage](/docs/credits): full credit cost table.