Stored on your Mac
Your PDFs, embeddings and search index stay on your Mac. CleverGhost finds evidence in your documents to ground each answer.
Version 10.0.0 · Coming soon
Cloud-scale AI.
Private by design.
Your PDFs, meet Apple Private Cloud Compute. CleverGhost 10 brings large language model power to your research, with no separate LLM to download and no API key to manage.
Your library stays on your Mac. Choose PCC when you want Apple to privately process your question and supplied document excerpts.
PCC requires macOS 27+, an eligible Apple Intelligence Mac and internet.
Daily usage limits apply. Version 10 is not yet available.
A little ghost. A powerful new chapter.
No separate LLM downloadNo API keySource-grounded answers
Coming in 10.0.0 / Private RAG + Apple PCC
Your library stays yours.
Your AI can go further.
Ask about a paper, connect ideas across your research, or follow the evidence through your library. CleverGhost brings relevant document text to Apple's larger cloud model, with privacy built into the choice.
More room to reason
Choose Apple PCC for a larger language model and more context than Apple's on-device model. Apple processes your question and the document text supplied for that request.
Skip the LLM download
No separate cloud model to install. No AI API key. On a supported Mac with Apple Intelligence enabled, select Apple PCC for your next conversation.
Cloud when you choose it. Local when you prefer it. Apple On-device stays the default. Local MLX models remain available after download.
How your data is handledAvailable in 9.0.2 · Continuing in 10.0
Multilingual search.
Still on your Mac.
EmbeddingGemma 2 is available now in CleverGhost 9.0.2 and continues in version 10. Google's multilingual search model finds passages in your PDF library by meaning and runs locally on Apple Silicon, after a separate download of approximately 1.52 GB.
English or multilingual search
Keep Apple's built-in English search model, or choose Google's multilingual EmbeddingGemma 2 in AI Settings → Library Search. Download the Google model once; document embeddings are then generated on your Mac.
Switch and reuse saved work
Apple and Google keep separate document indexes. A first switch prepares any missing embeddings in the background; switching back reuses saved embeddings for unchanged documents.
Choose how answers are written
Your search model is separate from your answer-writing AI engine. Search downloads use the AI Engine folder by default, with an option to choose a separate folder.
| Search model | Language support |
|---|---|
| Apple — built in | English sentence model used by CleverGhost |
| EmbeddingGemma 2 — local | 100+ languages, according to Google. CleverGhost retrieval checks covered English, French, and Turkish. |
EmbeddingGemma 2 can find passages across languages—for example, a French question about an English PDF. Search quality varies by language and document. Google's language support and model details ↗
Requires macOS 26+ and Apple Silicon. EmbeddingGemma 2 is a local search model; Gemini remains an optional cloud answer provider.
01 / Why our RAG goes further
Better evidence.
Answers you can trace.
A useful answer starts with finding the right evidence. CleverGhost's private RAG combines semantic and exact-term search, prioritizes relevant documents, and brings source passages into the answer. Choose on-device AI to keep document processing on your Mac.
Find ideas and exact details
Semantic search finds related ideas even when the wording differs. Exact-term search helps surface names and technical terms. Together, they give your question a broader route to relevant evidence.
Check the evidence behind an answer
Relevant passages from your PDFs give the AI concrete context. Document and page references help you verify a claim and read the original passage.
Go deeper, with control
Ask about one PDF, a library section, or several papers, then follow up in context. Larger local-AI searches work through relevance-ranked batches, with control over how far to continue.
Research papers. Everyday documents. Try asking:
- Passport
When does my passport expire?
- Research
Find me papers related to amorphous materials.
- Bills
How much do I need to pay on this bill?
02 / Inside CleverGhost
Your research, in one place.
A preview of the refreshed interface in CleverGhost 10.0.0, coming soon.
Organize papers and personal PDFs into a searchable library.
CLICK TO ENLARGE ↗Ask questions across documents and follow the source references.
CLICK TO ENLARGE ↗Ask about one paper, follow up in context, and save useful conversations.
CLICK TO ENLARGE ↗Explore connections between papers, topics, and ideas.
CLICK TO ENLARGE ↗Review highlighted edits for grammar, tone, and clarity before submission.
CLICK TO ENLARGE ↗Choose Apple On-device, optional Apple PCC, a downloaded local model, or Gemini.
CLICK TO ENLARGE ↗Choose Deep Dive, Quick Commute, or Peer Review Debate for your research podcast.
CLICK TO ENLARGE ↗Your PDF and research workflows, together in one Mac app.
CLICK TO ENLARGE ↗03 / Also included
Smaller PDFs. Papers you can hear. References ready to use.
PDF compression, podcast creation, and reference extraction are built into CleverGhost. Grammar Studio adds PDF-to-TeX extraction and a final polish before submission.
Our best PDF compressor
Included in CleverGhost. Make PDFs easier to share with light, balanced, or maximum compression, processed on your Mac.
Compression & more
Choose the balance between file size and image quality, with presets for web, email, and print. Merge, split, repair, watermark, protect, and extract pages in the same workspace.
Create research podcasts
Turn research papers into two-host audio conversations. Listen to the ideas in your library while you walk, commute, or take a break.
Meet Research Podcast
Create a podcast episode from a paper and choose the hosts' voices for your listening session.
See Research Podcast ↗ · Episode formats ↗
Extract PDF references
Bring references from your PDFs into BibGhost. Check citation metadata, organize your sources, and export a bibliography.
References & BibTeX
Prepare a bibliography for LaTeX or your reference manager. Online reference verification uses external scholarly services when you choose it.
Grammar Studio: PDF to TeX, then polish your draft
Extract PDF pages into editable text with LaTeX equations, or paste your draft. Correct grammar, spelling, punctuation, and sentence structure before submitting a paper, thesis, report, or cover letter.
How to apply corrections before submission
- Bring in your text. Select a PDF page and choose Extract Page, or paste text into the editor. PDF extraction uses the local Qwen3-VL model after setup.
- Fix and review. Choose Fix grammar for the whole text and review the highlighted changes. For a selected passage, preview the suggested replacement, then Accept, Try another, or Dismiss. Use Undo to revert an applied correction.
- Update your manuscript. Choose Copy and paste the reviewed text and LaTeX equations into your manuscript or source file. You can also refine clarity, academic tone, and US or UK spelling.
04 / Privacy, by choice
Your research.
Your Mac.
Your PDF library, embeddings, and search index are stored on your Mac. Apple On-device is the default AI engine. With Apple On-device or a downloaded local MLX model, document analysis and AI conversations run on your Mac. CleverGhost does not collect your documents or app telemetry.
Cloud AI requires an explicit engine choice. From version 10.0.0, selecting Apple Private Cloud Compute (PCC) sends your question and supplied document text to Apple's PCC service. No separate AI API key is needed. Selecting PCC keeps background library organization on your Mac, and failed requests do not silently switch engines. Selecting Gemini sends the question and content supplied for that request to Google's service using your own API key. Online citation verification separately contacts scholarly services.
- Apple On-device ON DEVICE
- Apple's built-in on-device model on supported Macs. This is the default AI engine.
- Local MLX models ON DEVICE
- Download a model once, then run AI locally. Choose Gemma, Mistral, or Qwen to suit your Mac's memory.
- Apple Private Cloud Compute OPTIONAL CLOUD
- From CleverGhost 10.0.0. Explicitly choose PCC to process your questions and supplied document text with Apple's service. Requires macOS 27 or later, an eligible Apple Intelligence Mac, internet access, and available daily quota.
- Gemini OPTIONAL CLOUD
- Bring your own API key when you choose cloud processing.
What is coming in CleverGhost 10?
Version 10.0.0 will add Apple Private Cloud Compute as an optional AI engine for PDF and library conversations. Apple's cloud model gives you more context and reasoning capacity than the on-device model, with no separate LLM download or API key.
PCC requires macOS 27 or later, an eligible Mac with Apple Intelligence enabled, internet access and available daily quota. Apple Intelligence itself may require system model downloads. Your library stays on your Mac; PCC processes your question and supplied document text when you choose it. Version 10 is coming soon and is not yet on the App Store.
Why does Apple Library Search say “English”?
CleverGhost currently uses Apple's built-in English sentence model to find passages by meaning. The English label describes that search model; it does not mean Apple Intelligence or the whole app is English-only. Matching names and exact words is a separate part of library search.
For multilingual and cross-language semantic search, choose EmbeddingGemma 2. Google reports support for 100+ languages; our local retrieval checks covered English, French, and Turkish. Your selected answer-writing AI engine determines which languages it can use in replies.
What is the optional EmbeddingGemma 2 search model?
An optional Google model for multilingual library search, available in CleverGhost 9.0.2 and continuing in version 10. It runs on your Apple Silicon Mac after an approximately 1.52 GB download. Choose it in AI Settings → Library Search. It finds relevant passages; your separately selected AI engine writes the answer.
Apple search remains available. Each search model keeps its own index, so unchanged documents can reuse saved embeddings when you switch back. Search model downloads follow your AI Engine storage folder by default, or you can choose a separate folder.
What is private RAG?
RAG stands for retrieval-augmented generation. CleverGhost searches your PDF library for relevant evidence and supplies it to the AI engine you choose. Your PDF library, embeddings, and search index are stored on your Mac. With Apple On-device or local MLX, the answer is generated on your Mac. With PCC (from version 10.0.0) or Gemini, your question and the document text supplied for the request are sent to the selected cloud service.
Does CleverGhost upload my documents?
Local AI processes document text on your Mac. When you explicitly choose Apple Private Cloud Compute (from version 10.0.0) or Google Gemini, the question and document content supplied for that request are sent to the selected service. Your library and search index remain stored on your Mac. Online reference verification separately sends citation queries to scholarly services. Cloud AI, reference verification, model downloads, and App Store purchases require an internet connection.
Which Mac and local model should I use?
CleverGhost requires macOS 26 or later. Downloaded MLX models require Apple Silicon. Apple Intelligence requires a supported Mac and enabled Apple Intelligence.
| Model | Mac RAM |
|---|---|
| Gemma 4 E2B (OptiQ) | 8 GB+ |
| Gemma 4 E4B (OptiQ) | 16 GB+ |
| Mistral Nemo 12B | 16 GB+ |
| Ministral 3 8B | 16 GB+ |
| Qwen 2.5 Coder 7B | 16 GB+ |
| Qwen 3.6 35B-A3B (experimental) | 64 GB+ |
These are starting recommendations, not memory guarantees. Long document context needs extra RAM. Model availability depends on your app version.
Is this only for academic papers?
No. Use it for research papers, textbooks, contracts, invoices, and other personal PDFs. Ask about one document or search across the library. Academic tools such as BibGhost and Research Atlas are available when you need them.
A PDF is stuck on “Indexing…” — what should I do?
Select that document in Smart Library and choose Re-analyze to rebuild its search index. Your original PDF is not changed. For a single stuck document, re-analyze it before considering a database reset.
What does “Reset Database Engine” remove?
A reset removes extracted metadata and the search index, not your original PDF files. Export your BibGhost library to a .bib file first, then re-import PDFs to rebuild the library. Use Maintenance & Recovery only for persistent database problems.
Start with your own library
A quieter way to research.
1-month Pro trial for eligible users.
$12.99/year or $39.99 lifetime.
Subscriptions and purchases are managed by Apple. See the App Store for local pricing and terms.