# Ask Rowsafe

> Ask about your databases in plain words and get answers from Rowsafe's data, with the fix as a button.

Source: https://rowsafe.sh/docs/guides/ask

**Ask Rowsafe** answers questions like "Why is app-prod slow today?" or "Could I restore billing right now, and to when?". It reads what the dashboard shows, answers briefly with numbers, and links its sources.

If a fix is the next step, the answer offers it as a button. Nothing runs until you click and confirm.

## Turn it on

An owner or admin turns it on once.

Open **Settings → Ask Rowsafe** to see what's sent and who handles it.

Tick **I understand what is sent** and click **Turn on Ask Rowsafe** (audited as `ask.enabled`).

Press <kbd>⌘</kbd> <kbd>J</kbd> (<kbd>Ctrl</kbd> <kbd>J</kbd> on Windows and Linux) or click **Ask**. On a database's pages, questions are about that database.

Then everyone can ask. **Explain** on findings, alerts and queries asks about that item. **Think deeper** uses a stronger, slower model.

## What the AI sees, and what it never sees

| Sent to the AI provider                                     | Never sent                                      |
| ----------------------------------------------------------- | ----------------------------------------------- |
| Your question                                               | The rows in your tables                         |
| Health scores, findings and alerts                          | Passwords, API keys, tokens, connection strings |
| Metrics and statistics: sizes, times, counts                | Backup contents and your encryption passphrase  |
| SQL statements with every value removed (`WHERE email = ?`) | IP addresses and file paths on your servers     |
| Names of databases, tables, indexes and columns             |                                                 |
| Backup, Proof and Mark history                              |                                                 |

The model uses read-only lookups limited to your organization. SQL values and anything that looks like a credential are removed first.

Who handles a question with the built-in AI:

1. **Rowsafe** runs the lookups and keeps the conversation 30 days after its last message.
2. **[OpenRouter](https://openrouter.ai/docs/guides/privacy/data-collection)** routes it, without storing it by default.
3. **The model's provider**: only ones that don't store or train on your data (`data_collection: deny`, [details](https://openrouter.ai/docs/guides/routing/provider-selection)).

With your own key, it's Rowsafe and your provider, under your account's settings. Answers can be wrong: check the linked data first.

## Plans and your own AI key

|                              | Free           | Pro            | Business       |
| ---------------------------- | -------------- | -------------- | -------------- |
| Built-in questions per month | none           | 500            | 3,000          |
| Your own AI provider key     | yes, unlimited | yes, unlimited | yes, unlimited |
| Your own AI assistant (MCP)  | yes            | yes            | yes            |

Questions reset on the 1st (UTC); failed ones don't count.

**Your own key** (OpenRouter, Anthropic or OpenAI): click **Use your own AI provider key**, paste it, and click **Check and save key**. It's stored encrypted and never shown again in full. Your provider bills you, with no Rowsafe limit; set a spending limit there. **More options** picks the models.

## Use your own AI assistant instead

Connect Claude, ChatGPT, Cursor, Codex or VS Code to Rowsafe's MCP server, free on every plan. The **Use your own AI** tab (plug icon) shows each setup with a prompt to copy. It uses a read-only API key today; with write access an assistant [acts as the person who connected it](https://rowsafe.sh/docs/guides/agent-autonomy). See the [MCP reference](https://rowsafe.sh/docs/reference/mcp).

## Conversations and feedback

**History** reopens conversations; the trash icon deletes one. Admins can delete all of them in **Settings → Ask Rowsafe**. They're deleted 30 days after the last message. Thumbs up or down records only the rating.

## Limits

- It knows only what Rowsafe collects: not your code, your data, or servers without the agent.
- At most 10 lookups per question (16 with Think deeper). Ask about one database for best results.
- Monitoring is at most a minute old, table statistics about 30 minutes.
