Quello AI Privacy Review: Safe for Freelance Client Files?

Short answer: Quello promises not to train models on your files, but the hosted service names no legal entity, no governing law, no storage region and no DPA, and its sub-processor list leaves out vendors its own public code relies on. Our verdict is AVOID FOR PAID CLIENT WORK on the hosted version. Policy reviewed October 2026, two days after Quello launched on Product Hunt as one place to ask questions across Google Drive, Notion, GitHub and your uploaded PDFs. The pitch is tempting for a solo consultant drowning in scattered client docs: ask one question, get an answer with citations to the exact passage. The problem is not what the policy says. It is what the policy leaves out, and who you would be trusting with every connected client folder. We scored it against how we evaluate AI tools, using only the vendor’s own documents and code.

What Quello does with your data

Privacy at a glance

Privacy dimension	Quello’s answer
Trains on your data?	No, per policy, for files, chats and Google data
Training opt-out	Not needed; no-training is the stated default
Data retention	Until you delete; backups purged within 30 days
Third-party sharing	Vercel, Render, MongoDB, Qdrant, Google, Groq listed
Storage region	Not stated anywhere we checked
Enterprise / team tier	Team workspaces exist; no DPA or SLA published
Legal entity	Not named; governing law left blank

Quello is a retrieval workspace. You connect sources (Google Drive, Notion, GitHub) or upload PDF, Word, text and Markdown files up to 20 MB. The service splits each document into chunks, turns them into vector embeddings and stores them in a per-workspace collection on Qdrant Cloud. When you ask a question, it pulls the closest passages and has a language model write an answer that cites them.

The privacy policy, dated October 12, 2026, makes four clear promises (per Quello’s privacy policy, retrieved 2026-10-20):

• Your data is never sold, rented or traded.
• Your profile, imported Drive files, uploads and chat questions are not used to train or fine-tune any general-purpose model, whether Gemini, Groq-hosted models or OpenAI.
• Model calls run through enterprise APIs that the vendor says carry zero-retention commitments for training.
• Deleting your account wipes profile, files, chats, connectors and embeddings, with encrypted backups aging out within 30 days.

On Google data specifically, the policy is careful. It uses the narrow drive.file scope, so Quello only sees files you pick in the Google Picker, not your whole Drive. Staff may not read Google data except with your explicit consent during support, for security investigations, when the law requires it, or in anonymized aggregate metrics.

Three things are missing. First, most of the policy is written around Google. Notion and GitHub connectors, which the homepage advertises, get no equivalent description of scopes or human-access rules. Second, the sub-processor list names Vercel, Render, MongoDB Atlas, Qdrant Cloud, Google Cloud and Groq, but the public GitHub README lists Cloudflare R2 for document storage and Vapi for the voice feature (per the Quello README, retrieved 2026-10-20). Neither appears in the policy. Third, the terms of service never name the company behind Quello, give no postal address and say the contract is governed by “applicable laws” without naming a country.

What this means for solo freelancers

Quello’s value grows with every source you connect, and so does the exposure. If you plug in the Drive folder holding three client contracts and the Notion workspace with your project notes, a single account now indexes all of it. Here is where that goes wrong, based on the policy as written.

Scenario 1: an EU client asks who processes their data. Under GDPR you act as a processor for your client, and any tool you route their files through becomes a sub-processor. You need a named counterparty, a location and ideally a data processing agreement. Quello gives you an email address. If you have not read what a data processing agreement means for solo consultants, this is the gap that matters: you cannot list an unnamed entity in your own records of processing.

Scenario 2: the sub-processor list does not match the code. Suppose a client NDA says their documents may only be stored with disclosed providers. Quello’s policy lists six. Its README describes a seventh (Cloudflare R2) holding the original files and an eighth (Vapi) handling voice. The hosted service may differ from the open-source code, but you cannot confirm that from the outside. That uncertainty is the risk.

Scenario 3: you connect ChatGPT or Claude through MCP. Quello’s MCP server lets an outside assistant search your workspace and fetch full documents. Once a passage from a client file lands in a ChatGPT or Claude conversation, Quello’s no-training promise no longer applies. That chat app’s own retention and training settings do. If your ChatGPT account still has model-improvement switched on, the client excerpt you just pulled in is governed by OpenAI’s rules, not Quello’s.

Scenario 4: the free Gemini tier question. The README tells self-hosters to grab a key from Google AI Studio. Google treats free and paid Gemini API usage differently (see our Google AI Studio privacy review). Quello’s policy says it uses enterprise APIs on the hosted service, which would be the paid side. That is a vendor statement you cannot audit.

None of this means the developer is careless. The policy reads as sincere and more specific than many startup notices. It is simply a one-person open-source project, a few months old, wearing enterprise language without enterprise paperwork.

How to use it safely

If you still want to try Quello, keep it away from anything a client would recognize as theirs, and set it up like this:

• Pick files, never folders, on Google Drive. In the Google Picker, select individual non-client documents. The drive.file scope means Quello sees only what you pick, so pick little.
• Skip the Notion and GitHub connectors for client work. The policy does not describe what those connectors can read, so assume full read access to whatever you authorize. If you must test them, create a separate Notion workspace and a throwaway GitHub org holding only your own material.
• Redact before you upload. Replace client names, emails, invoice amounts and contract parties with placeholders like CLIENT-A before a PDF goes in.
• Leave voice off. Voice routes through Vapi, which the policy does not list. Type your questions instead.
• Treat MCP keys like passwords. If you connect ChatGPT or Claude, use OAuth over a personal key, revoke the connection on the MCP Server page when you finish a task, and turn off training in the chat app first. Our guide to opting out of AI training for freelance work lists the toggles.
• Disconnect and delete after testing. Use Workspace Settings to disconnect each source, then delete the account. Revoke Quello at myaccount.google.com/permissions as a second check.
• Or self-host it. The code is MIT-licensed. Running your own copy puts storage under your control, though you still choose which model API sees your text, so pay for a tier with no-training terms.

Privacy-friendlier alternatives

Quello sits in the “ask questions across my documents” category. These options cover the same job with clearer ownership of your data.

1. Lumo by Proton. Lumo is Proton’s AI assistant, run by a named Swiss company with a published privacy policy. Chat history uses zero-access encryption, so Proton cannot read it, and conversations are not used for training.

• What it gives you that Quello doesn’t: an identifiable vendor, a known jurisdiction and encryption Proton itself cannot open.
• Pricing: free plan with limits; Lumo Plus is $12.99 per month, or $9.99 per month billed yearly.
• Best for: consultants who want to ask questions about one or a few uploaded client files without building a synced knowledge base. Read our Lumo privacy review for the details.

2. AnythingLLM Desktop. AnythingLLM runs a document-chat workspace entirely on your computer, with no account and no API key required when you use a local model.

• What it gives you that Quello doesn’t: client files and embeddings never leave your laptop, so there is no sub-processor list to worry about.
• Pricing: free desktop app; you need a reasonably recent machine to run local models well.
• Best for: developers, writers and analysts who keep client material on one machine and are comfortable picking a local model.

3. Self-hosted Quello behind Tailscale. If you like Quello’s cited answers, run the MIT-licensed code on a small server you control and reach it through Tailscale instead of exposing it to the internet.

• What it gives you that the hosted version doesn’t: you decide where files are stored, who has access and which model API (and tier) processes your text.
• Pricing: Tailscale Personal is free; expect a few dollars a month for a small VPS plus paid model API usage.
• Best for: technical freelancers who want Quello’s features and can maintain a server.

The verdict

Frequently asked questions

Does Quello train AI models on my documents?

Based on the policy as written, no. Quello states that Google profile data, imported Drive files, uploaded documents and your chat questions are not used to train or fine-tune any general model, and that model calls go through enterprise APIs with zero-retention terms for training. That promise covers Quello itself. Text you pull into ChatGPT, Claude or Perplexity through the MCP connector falls under those apps’ own settings.

Is Quello GDPR-friendly for EU client files?

The policy as written leaves EU freelancers short of what they need. It does not name a data controller, a postal address, a storage region, a legal basis for processing or any mechanism for international transfers, and no data processing agreement is offered. You can still delete your data and revoke access easily. But if a client asks you to document every sub-processor, Quello gives you too little to write down. This is not legal advice.

Can I use Quello for HIPAA or health data?

Nothing we reviewed mentions HIPAA, protected health information or a business associate agreement. Based on that, routing patient records, therapy notes or health claims documents through Quello carries a clear compliance risk for coaches, therapists or health-adjacent consultants. Keep clinical material in a tool whose vendor will sign a BAA, and use Quello, if at all, only for non-health research and your own business notes.

Is quello.dev related to Quello allergy drops?

No. This review covers Quello at quello.dev, the AI knowledge workspace launched on Product Hunt in October 2026. A separate consumer health brand at goquello.com sells allergy drops and publishes its own unrelated privacy policy, which currently ranks first for some Quello privacy searches. When checking terms or deleting data, make sure you are on quello.dev and that the Google consent screen names the right app.

What happens to my files if I connect Quello to ChatGPT or Claude?

Quello’s MCP server lets the connected assistant search your workspace and fetch full documents, read-only. Once a passage enters that chat, a copy now sits in the assistant’s conversation history under its own vendor’s retention and training rules. Revoke the connection on Quello’s MCP Server page when you are done, switch off training in the assistant first, and avoid fetching whole client contracts into a general chat window.

Sources

• Quello privacy policy, last updated 2026-10-12: https://quello.dev/privacy (retrieved 2026-10-20)
• Quello terms of service, last updated 2026-10-12: https://quello.dev/terms (retrieved 2026-10-20)
• Quello homepage and FAQ: https://quello.dev (retrieved 2026-10-20)
• Quello source code and README (MIT license): https://github.com/faizjamadar18/Quello (retrieved 2026-10-20)
• Quello on Product Hunt, launch of 2026-10-18: https://www.producthunt.com/products/quello (retrieved 2026-10-20)
• Lumo by Proton pricing: https://proton.me/lumo/pricing (retrieved 2026-10-20)
• AnythingLLM Desktop: https://anythingllm.com (retrieved 2026-10-20)
• Tailscale pricing: https://tailscale.com/pricing (retrieved 2026-10-20)

Reviewed by Jérémy, founder of AidTaskPro and GreenBudgetHub. Based in central France. Privacy posture sourced from public policies and vendor documentation as of 2026-10-20.