> ## Documentation Index > Fetch the complete documentation index at: https://docs.shinecx.com/llms.txt > Use this file to discover all available pages before exploring further. # AI overview > What ShineCRM AI does, what it does not do, and what data it sees. ShineCRM AI is a set of features that help you run your business with less manual work. It is part of **Suds AI**, which you turn on for your organization. Once on, some features run automatically on customer activity, and others you trigger yourself. A separate surface, the MCP server, lets you connect your own Claude account and operate the CRM by chat. ## Turn it on first Suds AI is **off by default**. An owner enables it under **Settings then Suds AI** and accepts a short responsibility attestation. Until then, none of the automatic AI features below run. You can also add custom instructions there to tune how the AI works for your business. ## What it does Automatically reads every inbound text and email and suggests follow-ups, pipeline moves, and draft replies for you to approve. After a job completes or an invoice is paid, drafts a personalized review request for the customer. On-demand polish for a text or email draft before you send it, tuned to the channel. Connect your own Claude account to ShineCRM and operate your CRM in plain language. **Runs automatically** (once Suds AI is on): incoming message suggestions, the daily Suds review of dormant customers, review request drafting, and call summaries (when you have [call recording](/guides/unified-inbox) turned on). **You trigger it**: the message rewriter, and the per-contact AI summary on a contact's profile. ## How approvals work Most AI output is a **suggestion**, not an action. Suggestions appear on the Suds AI page and on the relevant contact and conversation, where you approve, edit, or dismiss them. Approved messages are sent through your normal [scheduled messages](/guides/unified-inbox) queue. The one place AI can send without approving each message is **review requests under an auto-send rule**, which you set up yourself under **Settings then Reviews**. Even then, the scheduled messages are visible and you can edit or cancel them before they go out. If you prefer to approve every review request by hand, use a hold rule instead. See [Review request drafting](/guides/ai/review-requests). ## What data the AI sees To do the work above, the AI is given the relevant customer's details and conversation history, scoped to your organization. It does **not** have access to: * Photos or file attachments * Payment card details * Any other organization's data (every request is scoped to yours) Call and voicemail **recordings are an exception worth calling out**: when you have call recording enabled, recordings are transcribed and summarized so you can read a summary instead of re-listening. See [Call recording, transcripts, and summaries](/guides/unified-inbox). ## Privacy and AI providers Your data is processed by two AI providers, each only for the length of a single request: * **Anthropic** powers the text features (suggestions, drafting, rewriting, summaries). Under Anthropic's commercial API terms, your inputs and outputs are not used to train its models. * **OpenAI** provides the speech-to-text model (Whisper) used to transcribe call and voicemail recordings, and only when you have call recording turned on. Under OpenAI's API terms, API inputs are not used to train its models. No customer data is shared with any other AI provider. You can read Anthropic's [commercial terms](https://www.anthropic.com/legal/commercial-terms) and [training policy](https://privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training) directly. ## Usage Every member with owner or admin access can see the organization's AI usage under **Settings then Usage**, measured in tokens and requests for the billing period you choose. Usage is shown as quantities, not costs.