Humata: our review
Humata queries a corpus of internal documents in plain language, built for a small team rather than a single user. The vendor states in writing that it doesn't train its models on uploaded files, a commitment few competitors state so clearly.
What we take away from it
What counts in its favor
- States plainly that it doesn't use uploaded documents to train its AI models — rare and verifiable on its security page.
- Retention capped at 30 days for data processed by the engine, documented in a dedicated FAQ rather than buried in generic text.
- Certified SOC 2 Type II, with a program said to be informed by ISO 27001 and GDPR.
- Billed by page rather than by seat alone: the Expert tier covers three people for 500 pages a month, at $9.99.
What counts against it
- No hosting region is published: the infrastructure relies on AWS, GCP and Supabase with no geographic detail, making it impossible to guarantee European hosting in writing.
- The trust center (trust.humata.ai), which would carry the full subprocessor list and a DPA, sits behind an access form — name, email, company required before viewing it.
- The language model that actually answers the questions is named nowhere publicly, only the infrastructure providers are.
- The free tier (60 pages a month) is too short for real use: it's for testing, not working.
The most serious choice in this batch on the no-training commitment, but check the actual server location before uploading a file bound by professional secrecy: it's written nowhere.
The facts
- First paid tier
- $9.99/month — Expert
- Data hosting
- Hosting not verified
- Categories
- Documents and PDF,RAG and knowledge bases
- Data verified on
- 30 July 2026
What happens to your data
- Is your data used to train their models?
- No
- Subprocessors
- The security page names three infrastructure providers — Amazon Web Services, Google Cloud Platform and Supabase — and states they are all SOC-2 compliant. It names no language model provider, however, and the trust center that would carry the full subprocessor list and the DPA is reserved for named access requests.
- Retention
- Document data used by the engine isn't kept beyond 30 days. Dashboard data — your files, the exchange history — stays accessible until you request its deletion.
Humata in detail
Humata lets you query a corpus of documents in natural language — not a single file, but a set of contracts, reports, or notes a team consults regularly. The question you ask reads like the one you'd put to a colleague who had read everything: "which clause governs termination in our supplier contracts?"
What it actually does
You upload documents into shared folders, with per-team permissions on paid tiers. Humata indexes them and answers questions by citing the source passage, the same logic as chatting with a single PDF, but at the scale of an entire folder. The Team tier adds OCR for scanned documents and access control by department — two details that matter as soon as a corpus mixes material of different sensitivity.
Why its data commitment interests us
On the question that matters most for a client file — what happens to it once uploaded — Humata is one of the few tools in the sector to answer directly rather than dodge it. Its security page states plainly: "We do not use any of the data provided by our users to train our AI models." It adds a precise duration too: data processed by the engine is not retained beyond 30 days, which stands out against the market's usual silence on this point. It's the kind of guarantee an advisory audit looks for before recommending a tool to a law firm or an accounting firm.
Where it falls short
The training guarantee says nothing about server location. The infrastructure runs on AWS, GCP and Supabase, three global hosts, without specifying any region: there's no way to guarantee in writing that a file stays in Europe. The document that would settle the question — the trust centre, which promises a list of subprocessors and a DPA — isn't open: you must fill in a form with name, email and company before accessing it, which rules out any quick independent check. Finally, the language model that actually generates the answers isn't named anywhere: only the storage providers are.
Our recommendation
For a small team that wants a written commitment on the training use of its documents, Humata keeps a promise most competitors don't make. Before adopting it for a file covered by professional confidentiality, ask the sales team for the exact server location and the DPA — both probably exist, but aren't public.
Who it's for, and who it isn't
Suited for
- A small team of 3 to 10 people that regularly queries a corpus of contracts, reports or internal notes
- An organization that needs a written no-training commitment to reassure a client, beyond a marketing promise
Not a fit
- A firm bound by a strictly European hosting requirement: no region is published, impossible to guarantee it by contract without asking directly
- A single user with occasional use: the free tier is too short, a tool like ChatPDF costs less for this case
Frequently asked questions
- Do you need to know how to code to use Humata?
- No, the interface is a standard web dashboard: you upload files into folders, ask questions, get a sourced answer. No line of code is needed for everyday use.
- Does my client data leave my infrastructure?
- Yes: Humata is an online service hosted by third parties (AWS, GCP, Supabase). The vendor states it doesn't train its models on this content, but the exact geographic location of the servers is published nowhere.
- Does Humata train its models on my documents?
- No, the vendor states it explicitly on its security page: "We do not use any of the data provided by our users to train our AI models." It's a written commitment, not an inference.
- How much does Humata cost for a team of ten?
- The Team tier bills $49 per user per month for ten people and 5,000 pages a month, with per-folder permissions and OCR included. The Expert tier, at $9.99/month for three people and 500 pages, is enough for a smaller team.
- What's the difference with ChatPDF?
- ChatPDF targets individual use, free or nearly so, with no written commitment on model training. Humata targets a team, bills by page, and makes that commitment in writing — at the cost of a free tier too short for real use.
Alternatives to Humata
Documents and PDF
ChatPDF
Asks a PDF questions and returns the answer with the original passage.
Documents and PDF
AskYourPDF
AI chatbot for querying a PDF or a website, with a Chrome extension and ChatGPT plugin.
RAG and knowledge bases
DocsBot AI
Creates an AI chatbot trained on a site's documentation or content, to answer customer or internal questions.
Our sources
- Pricing and tiers pageaccessed on 30 July 2026
- Security page (training, retention, subprocessors, certifications)accessed on 30 July 2026
- Privacy policyaccessed on 30 July 2026
- Terms of serviceaccessed on 30 July 2026
Prices and hosting terms change often. We recheck them periodically, but the publisher's page remains the only up-to-date reference.