> ## Documentation Index
> Fetch the complete documentation index at: https://help.slashy.com/llms.txt
> Use this file to discover all available pages before exploring further.
# AI Privacy & Data Protection
> How Slashy protects your data when using AI: no training on customer data, provider retention policies, prompt injection protection, and Read Mode.
Slashy uses AI for importance labels, drafts, the agent, automations, and search. Here's exactly how your data interacts with AI systems and what protections are in place.
## No Training on Your Data
Your emails, calendar events, contacts, and agent conversations are **never** used to train any AI model -- not ours, not our providers'. This is a contractual guarantee in Slashy's [Data Processing Agreement](https://slashy.com/dpa).
## AI Providers and Retention
| Provider | Used For | Retention | Training |
| ---------------------- | ---------------------------- | --------------------------- | -------------- |
| **Anthropic (Claude)** | Agent, drafts, automations | 30-day logs, then deleted | No |
| **OpenAI (GPT)** | Certain features, embeddings | 30-day logs, then deleted | No (opted out) |
| **Google (Gemini)** | Certain features | No retention beyond request | No |
| **Groq** | Fast inference | No retention | No |
All providers are accessed via enterprise/API tiers with stricter data handling than consumer products.
## What Data Is Sent
Slashy sends the **minimum** data necessary per request:
* **AI Draft Reply:** The thread you're replying to + relevant memories
* **Importance Labels:** Subject, sender, content summary
* **Agent Sidebar:** Your question + relevant email content + memories
* **Automations:** Trigger condition + matched email content
Slashy does not send your entire inbox to AI providers. Only specific content relevant to the feature you're using is transmitted.
## Prompt Injection Protection
Multiple layers protect against malicious email content trying to manipulate the AI:
* **Input sanitization** -- known injection patterns detected and neutralized
* **Content boundaries** -- email text treated as untrusted data, not instructions
* **Writing style matching** -- agent defaults to your established patterns if manipulation is detected
* **Relationship context** -- requests from strangers get more scrutiny
## Agent Read Mode
Toggle in **Settings > Account**:
* **Read Mode ON:** Agent can read, search, and answer questions but cannot send emails, create events, or modify your inbox
* **Read Mode OFF (default):** Full capabilities -- the agent always asks confirmation before significant actions
## Your Controls
* **Choose AI model** -- Haiku (fast), Sonnet (balanced), Opus (most capable)
* **Manage memories** -- View, edit, delete in Settings > Memories
* **Disable AI features** -- Use Slashy as a fast email client without AI
* **Delete your data** -- Email [privacy@slashy.com](mailto:privacy@slashy.com) for full account deletion
## Common AI Privacy Questions
No. The agent always drafts emails for your review. It cannot send, delete, or modify emails without your explicit confirmation. Toggle on **Read Mode** in Settings for an extra layer of protection.
No. Each account is completely isolated. If you connect multiple Gmail accounts, data from one account is never used to generate responses for another.
When you use your own API keys, requests go directly to the AI provider under your own account and usage terms. Slashy does not proxy or log the content of BYOK requests.
No. Memories are private to your account. They are used only to personalize your AI experience and are never shared with other users, AI providers for training, or third parties.
***
## Related Articles
Certifications, encryption, and compliance
What Slashy stores and how to control it
Full list of infrastructure providers
Control your agent's writing style and autonomy