Tend AI Assistant Privacy Review: Safe for Freelance Client Work?

Short answer: Tend does not sell your data or train on what it reads in your connected apps, but it passes your messages through a chain of more than ten outside vendors, routes them to AI models you never pick, and names no company you could sign a contract with. Our verdict is AVOID FOR CLIENT WORK. Policy reviewed October 2026, one day after Tend launched on Product Hunt as a personal AI agent you text over iMessage or SMS. Connect Gmail, your calendar, QuickBooks or HubSpot and it drafts replies in your voice, chases leads and places phone calls for you. For a solo consultant that sounds like a free assistant. It also means one text thread can reach every client inbox, invoice and deal you own. We scored it using our review methodology, based only on Tend’s own published documents.

What Tend does with your data

Privacy at a glance

Tend is an agent, not a chatbot. Your phone number is your account, and you log in with a six-digit code texted to it. From there, Tend keeps your conversations, the files and photos you send, a running memory of facts about you, an optional vault of logins and addresses, a log of the actions it takes, and transcripts of any phone calls it makes for you (per Tend’s privacy policy, retrieved 2026-10-21).

The policy, last updated October 3, 2026, makes several promises worth crediting:

• Data pulled from connected apps is used only to carry out your requests. It is not sold, not used for ads and not used to train AI models.
• Tend asks for a yes before anything that changes an account: sending email, posting, editing a CRM record.
• It says it does not keep a full copy of your mailbox, CRM or books, only an action log and selected memory notes.
• The website runs no advertising or analytics cookies.

The sharing section is where the picture changes. Tend lists its vendors by role. Fly.io hosts accounts, conversations, memory and files in Secaucus, New Jersey. OpenRouter forwards your requests to model providers, with Anthropic, xAI, DeepSeek, OpenAI and Mistral given as examples. Composio and Nango hold the sign-in tokens for your connected apps. Kernel runs the browser Tend uses and stores any website logins you give it. Exa, Parallel, SerpApi and Firecrawl handle searches. TypeSafe receives new events from your apps to decide what deserves your attention. ElevenLabs handles outgoing calls. Redline, a quality vendor, receives conversations and agent steps so the team can catch mistakes. The text-message carrier itself is not named.

Three gaps stand out. The policy never says which model provider handles a given request, or what retention and training terms those providers apply. The terms of service name no company, no registered address and no country of incorporation. And the only contact besides a shared support inbox is a personal Gmail address listed for the developer.

What this means for solo freelancers

Tend’s pitch is breadth: one assistant across email, calendar, books and CRM. For client work, that breadth is the risk. Based on the policy as written, here is how it can go wrong. The pattern will look familiar if you read our readywhen privacy review, another assistant that lives in your messaging app.

Scenario 1: a client email reaches vendors you never approved. You connect the Gmail account where clients send briefs. Tend says it is notified when new mail arrives, and those events go to TypeSafe for triage. Ask Tend to summarize a client thread and the content goes through OpenRouter to whichever model it picks. If your contract with that client lists approved processors, or simply bans sharing with third parties, you now have at least four you cannot name in advance.

Scenario 2: an EU client asks for your sub-processor list. If you handle personal data for a client under GDPR, the tools you route it through are part of your processing chain. That normally means a named counterparty, a location and a data processing agreement. Tend offers none of the three. Everything is processed in the United States, and some vendors operate elsewhere. Based on the policy as written, you would be listing an unnamed service with no transfer safeguards described.

Scenario 3: someone takes over your phone number. Tend’s login is an SMS code, and its confirmation step (“say yes before I send”) also happens over text. If a SIM-swap attacker ports your number, they inherit an agent already connected to your Gmail, QuickBooks and HubSpot, plus any saved vault logins. The safeguard and the attack surface share the same channel.

Scenario 4: client financials in a quality-review queue. Ask Tend to pull last month’s unpaid invoices from QuickBooks and the steps it takes go to Redline for quality checks. The policy does not say how long Redline keeps them or who reviews them.

How to use it safely

If you want to try Tend for personal errands, keep client data structurally out of reach:

• Connect a separate Google account. Create a personal-only Gmail and calendar with no client threads. Never connect the inbox clients write to.
• Skip QuickBooks, HubSpot, Stripe and Slack. These hold client names, amounts and deal history. Tend’s value there is real, but so is the exposure.
• Leave the vault empty. Keep logins in a password manager you control and fill them yourself. Do not hand Kernel’s browser your client portal credentials.
• Lock your phone number first. Ask your carrier for a port-out PIN or SIM lock, and protect the Google account you connect with a hardware security key instead of SMS codes.
• Avoid sending files and photos. Contracts, IDs and screenshots you text Tend are stored with your account on Fly.io until you delete them.
• Audit the activity log weekly. It keeps the latest 500 actions. Check what Tend read, not just what it changed.
• Disconnect properly. Remove each app from its page under heytend.me/integrations, then revoke access at myaccount.google.com/permissions as a second check.
• Delete memory and request account deletion. Clear memory items in the app, then email support from the same phone number to delete the account. Tend commits to reply within 30 days.
• Turn off training where you can. Tend has no toggle, but the accounts it touches may. Our guide to opting out of AI training for freelance work lists the switches.

Privacy-friendlier alternatives

Tend combines an AI assistant, a credential vault and account access. Swap each piece for a tool with a named owner and clearer data terms.

1. Lumo by Proton. Lumo is the AI assistant from Proton, a Swiss company that publishes who it is and where it operates. Chats are protected with zero-access encryption, and Proton says it does not train on them.

• What it gives you that Tend doesn’t: an identifiable vendor, a known jurisdiction and chat history Proton itself cannot read.
• Pricing: free tier with limits; Lumo Plus runs about $10 to $13 per month depending on billing.
• Best for: consultants who want help drafting replies or summarizing a client document they paste in deliberately. Our Lumo privacy review covers the details.

2. Bitwarden. Instead of saving logins in an agent’s vault, keep them in Bitwarden, an open-source password manager with end-to-end encryption and published security audits.

• What it gives you that Tend doesn’t: credentials stay encrypted on your devices and are filled by you, never typed by a cloud browser you do not control.
• Pricing: free plan covers unlimited passwords; Premium costs well under $20 a year.
• Best for: any freelancer juggling client portals, hosting dashboards and shared logins.

3. A hardware security key (YubiKey 5 NFC). If you connect any account to an AI agent, protect it with phishing-resistant login. A YubiKey 5 NFC works with Google, Microsoft, QuickBooks Online and most password managers.

• What it gives you that Tend doesn’t: login security that does not depend on SMS codes and cannot be stolen through a SIM swap.
• Pricing: roughly $50 to $60 per key; buy two so one can be your backup.
• Best for: anyone whose phone number currently guards their business email.

If you still need an agent that acts across your inbox, wait for a vendor that publishes its legal entity, a DPA and a fixed list of model providers. Several larger suites offer this on business plans, at higher cost.

The verdict

Frequently asked questions

Does Tend train AI models on my data?

Tend says data from your connected apps is not used to train AI models, and it states it does not sell personal information or use it for advertising. What the policy does not cover is the downstream side. Your messages go through OpenRouter to providers such as Anthropic, xAI, DeepSeek, OpenAI and Mistral, and Tend does not say which retention or training terms apply at each of those providers.

Is Tend GDPR-friendly for EU client data?

Based on the policy as written, it carries real risk for EU client data. Tend acknowledges GDPR rights and says it honors access, correction and deletion requests for everyone. But it names no legal entity, offers no data processing agreement, processes everything in the United States and describes no transfer safeguards. If you act as a processor for an EU client, you would struggle to document Tend in your records.

Can I use Tend for HIPAA or health data?

The policy does not mention HIPAA, business associate agreements or health data handling, and Tend’s terms say its answers are not medical advice. Without a signed BAA, routing protected health information through a consumer AI agent and its vendor chain leaves you exposed. Health coaches, therapists and medical billers should keep patient details away from Tend entirely and use a tool that offers a BAA.

Which AI models see my Tend messages?

Tend sends requests through OpenRouter, a routing service, which forwards them to model providers. The policy names Anthropic, xAI, DeepSeek, OpenAI and Mistral as examples, but does not promise a fixed list or tell you which model handled a given message. Those providers receive your text and whatever Tend found for you, which can include email content or CRM records.

Is it safe to connect Tend to my work Gmail?

Not for an inbox that holds client correspondence. Tend says it is notified of new mail so it can flag what matters, and those events go to a triage vendor. Content you ask about then travels to an AI model. Access tokens sit with Composio rather than in Tend’s own database, which helps, but the exposure is still broad. Use a separate personal account instead.

Sources

• Tend privacy policy, last updated 2026-10-03, retrieved 2026-10-21: https://www.heytend.me/privacy
• Tend terms of service, effective 2026-10-03, retrieved 2026-10-21: https://www.heytend.me/terms
• Tend security page, last updated 2026-10-03, retrieved 2026-10-21: https://www.heytend.me/security
• Tend contact page, retrieved 2026-10-21: https://www.heytend.me/contact
• Tend homepage, retrieved 2026-10-21: https://www.heytend.me
• Tend on Product Hunt, launched 2026-10-20, retrieved 2026-10-21: https://www.producthunt.com/products/tend-5
• Proton Lumo product page, retrieved 2026-10-21: https://proton.me/lumo
• Bitwarden homepage, retrieved 2026-10-21: https://bitwarden.com

Reviewed by Jérémy, founder of AidTaskPro and GreenBudgetHub. Based in central France. Privacy posture sourced from public policies and vendor documentation as of October 21, 2026.