Page-aware retrieval
Text is extracted on device first, and OCR fills low-text pages. A retrieval model indexes bounded passages while retaining the document name, page, and OCR provenance used by citations.
Verified browser-local RAG
Ask across PDFs without uploading a single page.
Ask questions across up to ten PDFs while document text, retrieval, and answer generation remain on this device. The workspace returns grounded responses with links to the source document and page, including follow-up questions and an honest not-found result. Answers can still omit or misinterpret details, so important conclusions should be checked against the cited pages.
No document upload · no external AI APIUp to ten PDFs · starter actions + follow-upsVerify important answers against cited pages
Local AI download
112.6 MiB · one time
Local document search and answers, cached only after consent.
Recommended device
Desktop Chromium · 4 GB RAM
8 GB is preferred for longer collections. Mobile is a limited tier.
Private collection
0/10 PDFs · stored in this tab
PDFs, text, embeddings, questions, context, and answers stay local.
Explicit local AI setup
Verified retrieval and answer models
Setup downloads only pinned public assets. Three executable model files must match their SHA-256 hashes before local use.
- Retrieval files
- Pinned and integrity-checked
- Answer files
- Pinned and integrity-checked
- Runtime
- Verified local answer model
- License
- Apache-2.0 · English-first
Local documents
Add up to ten PDFs
Native page text is extracted first. Low-text pages reuse the completed OCR engine before local indexing.
Browser-local conversation
Ask, follow up, and inspect evidence
Your local conversation starts here.
Add PDFs, set up the verified local AI, then use a starter action or ask your own question. Citations open to the exact retrieved document and page excerpt.
Checking the local model cache.
Grounded, general, or absent
Document answers cite retrieved passages. Weak evidence returns a clear not-present result, while an optional general explanation is separately labeled and never attributed to the PDFs.
Honest device limits
Desktop Chromium with at least 4 GB RAM is recommended, and 8 GB is preferred for longer collections. Mobile, private windows, and low-storage devices are limited tiers.
One-time local AI setup
Know exactly what downloads and where it stays.
1. Review the bundle
The 112.6 MiB executable bundle contains local retrieval and answer models under Apache-2.0.
2. Consent and verify
No model request starts on page load. Setup pins immutable revisions and checks all three executable SHA-256 hashes.
3. Chat locally
PDF bytes, extracted and OCR text, embeddings, questions, chat history, retrieved context, and generated answers stay in this browser.
4. Reuse or remove
Verified assets can run offline from browser cache. Remove the models separately from clearing documents and chat.
Continue your workflow
Related tools
These available PdfEditorOnlineFree tools are connected to chat with pdf by the tool catalog.