EU-hosted · open models · zero data retention

Ask your documents anything. Keep them private.

Cereboid reads your contracts, reports, spreadsheets and slide decks and answers in seconds — with a source for every claim. Without sending a byte to the big US AI labs.

No credit card required.

annual_report.pdf, supplier_contract.docx, sales_2025.xlsx

Can we terminate the supplier contract early, and what would it cost us given this year's volumes?

Planned 3 steps · checked 14 passages · all claims verified

Yes — either party may terminate with 90 days' written notice 1, but early termination triggers a fee of 15% of the remaining annual minimum 2. With 2025 purchases of €1.84M against a €2.4M minimum 3, the fee would be about €84,000.
1 · supplier_contract.docx · p. 12 2 · supplier_contract.docx · p. 13 3 · sales_2025.xlsx · Q1–Q4

What you can do

Chat with PDFs, summarize them and pull out the data

One private assistant for PDFs, Word, Excel and PowerPoint — scanned pages included.

Chat with PDF

Ask questions about any PDF in plain language and get answers that cite the exact page they came from.

Summarize PDF

Turn a 200-page report into a one-page summary, or pull out just the key points, risks and deadlines.

Extract invoice data

Save the fields you need — supplier, VAT number, dates, totals — as a playbook, run it on every invoice and export a spreadsheet.

Review contracts

Find termination clauses, penalties, renewal dates and obligations across all of your contracts at once.

From pile of files to answers in three steps

1. Upload

Drop in PDFs, Office files, spreadsheets or exports. Scanned pages are read too.

2. Ask

Ask in plain language, in any language. Pick Fast, Balanced or Deep depending on the question.

3. Verify & reuse

Click any citation to see the source. Save questions into playbooks to reuse on new files.

Privacy is the product

Built for people who can't send their data to Big AI

Law firms, healthcare, finance, public sector — or anyone who would rather keep confidential documents confidential.

Zero data retention

Open-weight models run through Venice.ai's private inference. Your prompts and documents are never stored by the model provider or used for training.

Hosted in the EU

Your files, passages and knowledge graph live on our own servers in the European Union — not in a US hyperscaler.

Sandboxed parsing

Every upload is opened inside a disposable, network-less sandbox that is destroyed after reading. A malicious file can't reach the app or other users.

Encrypted per user

Files are encrypted at rest with a key unique to your account, and database rows are isolated per user by row-level security.

Delete means delete

Remove a document and every passage, embedding and graph node derived from it is erased immediately.

No big-lab APIs

No OpenAI, Anthropic or Google APIs touch your data. Models: GLM, DeepSeek, Qwen and Gemma, open-weight.

Research-grade answers

Not a chatbot bolted onto search

Cereboid combines the most advanced retrieval techniques — hybrid search, cross-encoder reranking, hypothetical document embeddings, small-to-big retrieval, PageIndex navigation, GraphRAG, multi-hop and multi-agent research, corrective and self-reflective RAG — so it finds the needle and tells you when it isn't there.

Hybrid search + reranking

Semantic vectors and exact keywords fused, then re-scored by a cross-encoder running on our own servers.

Reads like an expert

Builds a table of contents for every file and navigates it — great for long reports and contracts.

Knowledge graph

Maps people, companies, amounts and obligations across documents to answer multi-step questions.

Self-checking agents

A team of agents plans, researches, grades evidence and verifies every claim before you see it.

Simple, pay-as-you-go pricing

No subscription. Credits never expire. Start with 100 free credits.

Starter

€5

500 credits

Plus

€20

2,200 credits

Pro

€50

6,000 credits

Team

€100

13,000 credits

Fast answer ≈ 0.3 credits

Verified answer ≈ 2 credits

Deep research ≈ 8 credits

See full pricing

Questions

Who can see my documents?

+

Only you. Files are encrypted with your own key, parsed in an isolated sandbox, and never shared. Model inference is zero-retention.

Can I chat with a PDF?

+

Yes. Upload a PDF — or a Word, Excel or PowerPoint file — and ask questions in plain language. Every answer cites the document and the page it came from.

Can it summarize long PDFs?

+

Yes. Ask for an overall summary, a summary per chapter, or one focused on risks, deadlines or amounts. Cereboid navigates the document's outline, so it works on reports hundreds of pages long.

How do I extract data from invoices?

+

Create a playbook with the fields you need — supplier, VAT number, date, total — run it on every invoice and export the answers as a CSV spreadsheet.

Which files can I upload?

+

PDF (including scanned pages), Word (.docx), PowerPoint (.pptx), Excel (.xlsx, .xls), CSV, JSON, XML, HTML, Markdown and plain text, up to 50 MB each.

How does pricing work?

+

You buy credits and pay only for what you use. A quick answer costs about 0.3 credits, a thorough fact-checked answer about 2, and a deep multi-agent investigation about 8. Processing a typical 20-page document costs around 1 credit.

How accurate are the answers?

+

Answers are grounded only in your documents, every statement links to the exact passage it came from, and a verifier agent removes claims the sources don't support.

Your documents have answers. Keep them yours.

Get started free