💻 Installation & Activation

Downloading LibrAIry

You can download LibrAIry for free from the official website at librairy.app. The download page offers installers for Windows (installers for macOS and Linux will be provided, depending on the demand). Simply click the download button for your operating system and follow the standard installation procedure for your platform.

On Windows, this means running the downloaded .exe installer and following the on-screen instructions. The installation is straightforward and does not require any special configuration.

First Launch — Activating a License

When you start LibrAIry for the first time, an activation window will appear. LibrAIry requires a license key to operate, but obtaining one is quick and free:

Getting a free trial key

1. In the activation window, click the 🎁 Get Free 30-Day Trial button. This opens your web browser to the LibrAIry checkout page.
2. On the checkout page, enter your email address. The trial is completely free — no credit card is required, and you will not be charged anything.
3. After completing the checkout, you will receive an email within a few seconds containing your trial license key. The key looks something like B726FE62-5A29-45D3-B763-01675C792D27.
4. Copy the license key from your email, go back to the LibrAIry activation window, paste it into the "License Key" field, and click ✓ Activate.
5. LibrAIry will verify your key online and start immediately. You now have 30 days to evaluate the software with all features enabled.

Buying a license key

If your trial has expired or you are ready to upgrade, click on 🛒 Buy License in the activation window to see the available options. You have the choice between two types of permanent licenses:

• Personal License: This removes the 50-article limit, allowing you to have an unlimited number of PDF files in your library. It includes 8,000 cloud AI credits, which connects seamlessly via LibrAIry Cloud (no complex API configuration required on your part). These credits allow you to extract metadata, analyze, and synthesize complete reviews for a massive number of articles. When your 8,000 credits are completely used, you can easily buy extra credits by following the prompts in the software.

• Professional License: This tier also allows for unlimited PDF files and gives you total flexibility over your AI backend. It includes 8,000 cloud AI credits for immediate use via LibrAIry Cloud — just like the Personal license. In addition, the Professional license unlocks two more AI options:

  • Your own Google Gemini API Key: You can generate your own key from Google for free, with no credit card required. The free tier allows up to 250 AI requests per day, which is sufficient for most researchers. If you need more, a pay-as-you-go tier is available at extremely low cost (fractions of a cent per article). Details on how to set this up, including exact costs and rate limits, are in the "Setting Up AI" section.

  • Ollama (Local or Remote AI): You can install this free, open-source AI directly on your computer, or connect to Ollama running on a more powerful machine on your network (such as a lab workstation with a dedicated GPU). This option is completely free to use and has the massive advantage of 100% privacy — no data ever leaves your computer or your local network.

(Activation for purchased licenses works exactly the same way as the trial: check your email for the purchased key, paste it into the "License Key" field, and click ✓ Activate.)

Activating on a New Computer

Each license can be activated on a limited number of computers (1 for Trial, 2 for Personal, 3 for Professional). If you need to move LibrAIry to a new computer, you should first deactivate the license on the old computer by going to 📚 LibrAIry → 🔑 License and clicking Deactivate. This frees up an activation slot. You can then activate the same key on your new computer.

🚀 Quick Start Guide

This guide walks you through the essential steps to get your first library up and running. The entire process takes about 5 minutes for a small collection of papers.

Step 1 — Create a Library

A "library" in LibrAIry is simply a folder on your computer where your articles and their metadata will be stored. To create one, go to 📂 File → 📁 New Library in the menu bar. A dialog will appear asking you to select a folder.

You can either choose an existing empty folder or create a new one. For example, you might create a folder called My_Research_Papers on your desktop or in your Documents directory. LibrAIry will then create the necessary internal structure inside that folder, including subdirectories for your PDF files and index files.

Step 2 — Import Your PDF Articles

Now that your library is created, you need to add articles to it. Go to 📂 File → 📥 Import Files. A dialog will appear offering you two options: importing individual files, or importing an entire folder of PDFs.

If you choose to import a folder, LibrAIry will scan it recursively (meaning it will also look inside subfolders) and find all PDF files. Each PDF will be copied into your library folder — your original files are never modified or moved. If LibrAIry finds files that are already in your library (based on their filename), it will skip them automatically to avoid duplicates.

After the import is complete, you will see your articles listed in the main table. At this point, the title column will show the filename of each PDF (without the .pdf extension), and the author and year columns will show "Unknown" — this is normal, because metadata has not been extracted yet.

Step 3 — Extract Metadata

This is where the magic happens. Click 📝 Extract Metadata in the menu bar. LibrAIry will process each article through its extraction pipeline, which works in several stages:

1. First, it extracts text from the PDF and searches for a DOI (Digital Object Identifier) — a unique identifier that most published articles have.
2. If a DOI is found, it looks up the article's complete bibliographic information from OpenAlex and CrossRef, two free academic databases.
3. If the DOI lookup does not return complete information, or if no DOI was found, and if you have configured an AI backend, LibrAIry will use AI to read the first pages of the PDF and extract the metadata from the text itself.

During extraction, a live log shows you what is happening for each article. When extraction is finished, a summary report tells you how many articles were successfully processed. Click 📋 Back to Library to return to the main table and see your articles with their newly extracted metadata.

If some articles are flagged as [Scanned Article], it means they are scanned PDFs (images without selectable text). LibrAIry can handle these too — see the "OCR Processing" section to learn how to extract text from scanned articles and then retrieve their metadata.

Step 4 — Explore Your Library

With metadata extracted, you can now take full advantage of LibrAIry's features:

• Click on a column header (Authors, Year, or Title) to sort your articles. Click again to reverse the sort order.
• Click 🔍 Search in the menu bar to open the advanced search dialog where you can combine multiple search criteria.
• Right-click on any article to see its full details, edit its metadata, export its BibTeX reference, open the PDF, or run OCR on scanned articles.
• Select articles using the checkboxes in the left column, then go to 🤖 AI Tools to chat with them or generate a literature review.
• Use 📂 File → 📋 Embed Metadata in PDFs to write LibrAIry's metadata into your PDF files for compatibility with Zotero, Mendeley, and other tools.
• Use 📂 File → 🗑️ Delete Duplicates to find and remove duplicate articles automatically.

📂 Library Management

Creating a Library

To create a new library, go to 📂 File → 📁 New Library and select an empty folder. LibrAIry will create the following structure inside that folder:

Opening an Existing Library

To open a library that you created previously, go to 📂 File → 📂 Open Library and select the library folder (the parent folder, not the LIB_PDF or LIB_INDEX subfolder). LibrAIry remembers your last library and opens it automatically when you start the application.

Importing Papers

Go to 📂 File → 📥 Import Files to add articles to your library. You can choose between two options:

• Import individual files — Select one or more PDF files from anywhere on your computer.
• Import a folder — Select a folder, and LibrAIry will find and import all PDF files inside it, including those in subfolders.

In both cases, LibrAIry copies the PDFs into the library's LIB_PDF folder. Your original files are left untouched in their original location. If a file with the same name already exists in the library, it is skipped to avoid duplicates.

During import, a progress counter in the status bar at the bottom of the window shows you how many files have been copied (for example, "Importing: 847 / 3000 (28%)"). This is especially useful when importing large collections.

Deleting Duplicate Articles

Over time, especially when importing articles from multiple sources or folders, you may end up with duplicate articles in your library — the same paper present more than once under slightly different filenames. LibrAIry includes a built-in tool that detects and removes these duplicates automatically.

To use it, go to 📂 File → 🗑️ Delete Duplicates. LibrAIry will scan all articles in your library and compare their titles. When it finds a group of articles with the same title (or very similar titles), it keeps the one that has the most complete metadata and removes the others.

The removed articles are not permanently deleted. Their PDF files are moved to a folder called LIB_DUPLICATES inside your library folder, so you can review them and recover any file if needed. Only the index entries are removed from your library's article list.

📝 Metadata Extraction

Metadata extraction is the process by which LibrAIry reads your PDF files and identifies their bibliographic information: title, authors, year, journal, DOI, abstract, and more. This is done automatically through a multi-step pipeline that combines free online databases with optional AI assistance.

How the Extraction Pipeline Works

When you click 📝 Extract Metadata, LibrAIry processes each article through five stages. Each stage tries to find the missing information, and the process stops as soon as all essential fields (title, authors, year, journal) have been found.

Extraction Status Tags

After extraction, each article receives a status tag that tells you how much metadata was found:

How to Run Extraction

Extract all unprocessed articles

Simply click 📝 Extract Metadata in the menu bar without selecting any articles. LibrAIry will process all articles that do not already have complete metadata.

Extract only selected articles

If you want to process only certain articles, use the checkboxes in the left column to select them, then click 📝 Extract Metadata. Only the selected articles will be processed. This is useful when you import new articles and want to extract their metadata without re-processing your entire library.

Re-extract a single article

Right-click on any article in the table and choose 🔄 Re-extract Metadata. This forces LibrAIry to re-run the entire extraction pipeline for that article, even if it already has metadata. This can be useful if you suspect that the metadata is incorrect or incomplete.

Monitoring Extraction Progress

During extraction, the main table is replaced by a live log that shows you exactly what is happening for each article. You can see which stage of the pipeline is running, whether a DOI was found, which database returned the metadata, and whether AI was used as a fallback.

When extraction is finished, a summary report is displayed showing the number of articles in each status category, how many times AI fallback was used, and the estimated cost if you are using a paid AI service.

Click 📋 Back to Library or the ✕ button in the top-right corner of the log panel to return to the main table.

Stopping Extraction

If you need to stop extraction before it finishes (for example, if it is taking too long or if you notice a problem), click the 🛑 Stop Extraction button that replaces the "Extract Metadata" button during processing. LibrAIry will finish processing the current article and then stop. All articles that were already processed will keep their newly extracted metadata.

What Metadata Extraction Actually Extracts

The name “metadata extraction” is somewhat misleading, because LibrAIry extracts far more than just bibliographic metadata. During the extraction process, LibrAIry also captures the full text content of each article — the abstract, conclusion, and body text. This extracted text is stored in your library’s index and serves three important purposes:

• Keyword search: once the text is extracted, you can search for any word or phrase across all your articles using the multi-criteria search. This is much more powerful than searching only titles or abstracts.
• AI Chat and Synthesis: when you ask questions about your articles or generate a literature review, the AI uses the extracted text. The more text that has been extracted, the more detailed and accurate the AI’s analysis will be.
• Embed Metadata in PDFs: after extraction, you can write the bibliographic metadata (title, authors, DOI, abstract, journal, etc.) directly into your PDF files’ internal properties, using industry-standard formats (XMP Dublin Core, PRISM). This makes your PDFs compatible with reference managers like Zotero, Mendeley, and EndNote, and allows you to see article information in Windows File Explorer properties.

Why Extraction Can Be Slow — and Why It Is Worth It

Because LibrAIry extracts both metadata and the full article text, and because it queries multiple online databases (OpenAlex and CrossRef — both free, open-access scholarly catalogs containing hundreds of millions of article records) for each article, extraction takes longer than a simple file import — typically 2 to 5 seconds per article. For a library of 500 articles, this means the initial extraction can take 20 to 40 minutes.

However, extraction is done once and for all. Once an article has been processed, its metadata and text are stored permanently in your library’s index file. You never need to re-extract the same article unless you want to update its metadata.

LibrAIry’s multi-source extraction pipeline (DOI detection → OpenAlex → CrossRef → AI fallback) achieves a very high success rate. In our testing, over 85% of articles from standard academic publishers are fully identified (title, authors, year, journal, and DOI) without any manual intervention. The remaining articles are typically conference papers, technical reports, or documents without a DOI, for which the AI fallback can often extract partial metadata from the first page.

🔍 OCR Processing (Scanned Articles)

Some PDF files are "scanned" documents — they contain images of pages rather than actual text. This is common with older papers, photocopied book chapters, or documents that were scanned using a physical scanner. When you open such a file in a PDF viewer, it looks normal, but if you try to select or copy the text, nothing happens — the text is "trapped" inside an image.

LibrAIry automatically detects these scanned articles during metadata extraction. When it encounters a PDF with no extractable text (or very little text), it flags the article with a [Scanned Article] tag. At that point, the normal metadata extraction pipeline cannot work, because there is no text to search for a DOI or to send to the AI.

To solve this problem, LibrAIry includes a built-in OCR (Optical Character Recognition) engine powered by Tesseract, one of the most widely used open-source OCR tools. OCR works by analyzing the image of each page in the PDF, recognizing the letters and words in the image, and converting them back into selectable, searchable text. Once OCR has been performed on a scanned article, LibrAIry can then use that text to search for the article's DOI and metadata, exactly as it would for a normal text-based PDF.

How Scanned Articles Are Detected

LibrAIry uses an intelligent word-count analysis to determine whether a PDF is scanned or text-based. For each page of the PDF, it counts the number of actual text words that can be extracted from the native text layer (as opposed to text embedded in images). A page that is a pure scan will typically yield zero or very few words, while a normal text-based page will contain hundreds of words.

The detection uses two criteria: if the average number of words per page is below 25, or if more than 65% of the pages contain fewer than 20 words, the PDF is classified as a scanned document. This approach is reliable for most types of scanned articles, including those that have a few text elements overlaid on the scan (such as headers or page numbers added by the scanning software).

This detection happens automatically during the regular metadata extraction process (when you click 📝 Extract Metadata). You do not need to do anything special — LibrAIry will detect scanned articles on its own and flag them with the [Scanned Article] tag.

The OCR Workflow: Three Steps

Processing scanned articles in LibrAIry is a three-step workflow. Each step is a separate action, which gives you full control over the process.

Practical Example

Imagine you have imported 200 articles, and after running 📝 Extract Metadata, 15 of them are flagged as [Scanned Article]. Here is what you would do:

1. Click the checkbox in the header row to select all articles, or use the search to find scanned articles, then select them manually.
2. Right-click and choose 🔍 OCR (Scanned Articles). LibrAIry will only process the 15 scanned articles (non-scanned articles are automatically skipped). Wait for the OCR to complete — this may take a few minutes depending on the number of pages.
3. Once OCR is done, right-click again and choose 📝 Extract from OCR'd Articles. LibrAIry will attempt to find the DOI and metadata for each of the 15 articles using their OCR text.

After these steps, most of your scanned articles should have at least partial metadata. For any that remain unidentified, you can always enter the metadata manually using the editor (right-click → ✏️ Edit Metadata).

📋 Embed Metadata in PDFs

When LibrAIry extracts metadata for your articles (title, authors, year, journal, DOI, abstract, keywords), this information is stored in LibrAIry's own index file (Index.json). However, the PDF files themselves may not contain this information in their internal properties. This is a problem if you want to use your PDFs with other tools — for example, when you import them into Zotero, Mendeley, or EndNote, or when you browse your files in Windows Explorer and want to see the article's title and author in the file properties.

The Embed Metadata in PDFs feature solves this by writing LibrAIry's metadata directly into the PDF files' internal properties. After running this tool, your PDF files will carry their bibliographic information with them wherever they go, regardless of whether LibrAIry is installed or not.

What Gets Written

LibrAIry writes metadata into the PDF files using three industry-standard formats, which ensures maximum compatibility with a wide range of tools and operating systems:

How to Use It

Go to 📂 File → 📋 Embed Metadata in PDFs. A dialog appears showing a summary of the fields that will be written (Title, Authors, Abstract, Keywords, Full Reference + DOI) and the number of articles that will be processed.

If you have selected specific articles using the checkboxes in the main table, only those articles will be processed. If no articles are selected, LibrAIry will process all articles in your library.

Before running, you can choose one of three writing modes:

After choosing a mode, click 📋 Embed Metadata. A progress indicator shows how many files have been processed, and a detailed log shows which fields were written for each file. Articles that have no metadata to write (for example, articles whose metadata is still unknown) are skipped automatically.

Why This Is Useful

Embedding metadata in your PDF files has several practical benefits:

• Interoperability with other tools. When you import your PDFs into Zotero, Mendeley, or EndNote, these tools can read the embedded metadata and automatically create bibliographic records without requiring you to enter the information manually or run a separate lookup.
• Better file browsing. On Windows, you can see the article's title and author in the "Details" column of File Explorer, making it easy to identify files without opening them.
• Portability. If you share a PDF file with a colleague, the bibliographic information travels with the file. The recipient does not need LibrAIry to see the metadata — any PDF viewer or file manager can display it.
• DOI-based lookup. By embedding the DOI in the PRISM metadata, tools like EndNote can use it to automatically retrieve the full bibliographic record from CrossRef, ensuring that your references are complete and accurate.

🔍 Search & Filtering

Sorting the Table

You can sort your articles by clicking on any of the three sortable column headers in the table:

• Authors — sorts alphabetically by the first author's last name.
• Year — sorts chronologically.
• Title — sorts alphabetically by title.

Clicking the same column header a second time reverses the sort order (from ascending to descending, or vice versa). Articles that have missing data for the sorted field are always placed at the bottom of the list.

Advanced Search Dialog

For more precise searches, click 🔍 Search in the menu bar to open the multi-criteria search dialog. This dialog allows you to combine up to three search conditions using logical operators.

Each search row consists of three parts:

• Operator (for the second and third rows) — determines how this condition relates to the previous one. Choose AND (both conditions must be true), OR (either condition can be true), or NOT (this condition must not be true).
• Field — choose which field to search in: Title, Author, Year, Journal, Key Words, Abstract, Conclusion, DOI, or All (searches all fields at once).
• Value — the text you are searching for. The search is case-insensitive, which means that searching for "smith" will also match "Smith" and "SMITH".

Search Examples

• Find all articles by a specific author published after 2020: set Field to Author, Value to Smith; then AND, Field to Year, Value to 202 (this matches 2020, 2021, 2022, etc.).
• Find articles from either Nature or Science: set Field to Journal, Value to Nature; then OR, Field to Journal, Value to Science.
• Find articles about climate change that are not review articles: set Field to Title, Value to climate; then NOT, Field to Title, Value to review.

Clearing a Search

To return to the full list of articles after a search, click the Clear Search button in the table header (next to the "Title" column). This restores the complete article list.

Selecting Articles

Use the checkboxes in the leftmost column of the table to select articles for batch operations such as extraction, AI Chat, or AI Synthesis. The checkbox in the header row selects or deselects all currently visible articles. The Unselect button in the header clears all selections.

📄 Managing Papers

Right-Click Context Menu

Right-clicking on any article in the table opens a context menu with the following options:

Editing Metadata Manually

When you right-click an article and choose ✏️ Edit Metadata, an editor opens with all available fields: Title, Authors, Year, Volume, Pages, Journal, DOI, URL, Keywords, and Abstract. You can modify any of these fields and save your changes. The changes are applied immediately to both the JSON index and the article's individual BibTeX file.

Manual editing is particularly useful in the following situations:

• Correcting an author name that was extracted with a typo.
• Adding a publication year that the extraction pipeline could not find.
• Adding metadata for a scanned article where automatic extraction and OCR did not produce usable results.
• Completing the abstract for an article where only the title and authors were found.

BibTeX Export

Exporting a single article

Right-click on the article and choose 📚 Export BibTeX. You can either save the reference as a .bib file or copy it to your clipboard to paste it directly into your LaTeX document or another reference manager.

Exporting your entire library

LibrAIry automatically maintains a master BibTeX file called Index.bib in your library's LIB_INDEX folder. This file contains all the references in your library and is updated automatically every time you extract metadata or edit an article. You can use this file directly as the bibliography source for a LaTeX project.

🧠 Setting Up AI

LibrAIry's AI features (Chat, Synthesis, and AI metadata fallback) require an AI backend to function. This section explains each option in detail and walks you through the setup process step by step. If you are not familiar with AI services, API keys, or running local models, this guide will cover everything you need to know.

Understanding Your Options

LibrAIry offers three ways to connect to an AI service. Here is a quick comparison to help you decide:

── Cloud Proxy ──

Option 1: LibrAIry Cloud (Simplest)

Trial, Personal, and Professional Licenses

This is the easiest option and requires absolutely no configuration on your part. If you have a Trial, Personal, or Professional license, your license includes a certain amount of cloud AI credits. When you use AI features, the cost is deducted from these credits. You do not need to create any account or obtain any API key.

This option (LibrAIry Cloud) is automatically selected in 📂 File → 🌐 Network & AI Settings as your AI backend. If your license is activated, everything will work immediately.

Your remaining credits are displayed in the status bar at the bottom of the main window (for example, "5,200 / 8,000 Credits"). You can click on this indicator to refresh it at any time. If you are using a Professional license with your own API key, the status bar will display your Token usage instead.

You can view the available AI models (Large Language Models, or LLMs) by clicking the Check button. A list of available models will appear in the dropdown menu, and the model recommended by LibrAIry is automatically selected for you.

── Gemini ──

Option 2: Google Gemini (Your Own API Key)

Professional License only

Select Own Google Key in 📂 File → 🌐 Network & AI Settings.

Google Gemini is a powerful AI model developed by Google. LibrAIry can connect to it directly using an API key that you obtain for free from Google. This option gives you full control over your AI usage and is extremely affordable.

How to Get a Google Gemini API Key

Follow these steps carefully. The entire process takes about 5 minutes and does not require any programming knowledge:

1. Open your web browser and go to aistudio.google.com. This is Google’s AI development platform, also called “Google AI Studio”.
2. Sign in with your Google account (the same account you use for Gmail or Google Drive). If you do not have a Google account, you will need to create one first.
3. Once signed in, look for a button or link that says “Get API key” — it is usually in the top-left area of the page or in the navigation menu.
4. Click “Create API key”. Google will generate a long string of characters that looks something like AIzaSyB1a2c3d4e5f6g7h8i9j0kLmNoPqRsTuVw. This is your API key.
5. Copy this key carefully (select it all and press Ctrl+C). You will need to paste it into LibrAIry.

Entering Your API Key in LibrAIry

1. In LibrAIry, go to 📂 File → 🌐 Network & AI Settings.
2. In the “AI Backend” section, select Own Google Key.
3. Paste your API key into the “Google API Key” field.
4. Choose an AI model. The recommended model is typically the latest “Flash” version available in the dropdown (such as gemini-2.5-flash), which offers the best balance of speed and very low cost. 💡 Pro Tip: Keep your models updated! Google frequently releases newer, smarter, and often cheaper models. Click the refresh button occasionally to fetch the latest versions.
5. Click Save. LibrAIry will test the connection and confirm that your key works.

Free Tier (No Credit Card) — What You Get and What the Limits Are

When you create an API key on Google AI Studio, you are automatically on the free tier. This requires no credit card, no billing account, and no payment of any kind. The free tier never expires — you can use it indefinitely.

However, the free tier has rate limits that restrict how many requests you can make per minute and per day. Here are the current limits (as of early 2026; Google may adjust these at any time):

What this means in practice for LibrAIry:

• Metadata extraction (one AI call per article, only when the DOI/database lookup fails): up to ~250 articles per day with Gemini 2.5 Flash. In practice, most articles are identified by their DOI without needing AI, so you will rarely reach this limit.
• AI Chat: each message counts as one request — up to ~250 messages per day, more than enough for any research session.
• AI Synthesis: each synthesis counts as one request — up to ~250 per day.
• Speed: at most one request every 6 seconds (10 RPM). Batch extraction is slightly slower than with a paid tier, but still perfectly usable.

For most individual researchers, the free tier is sufficient. You only need to consider the paid tier if you regularly process very large batches (500+ articles) or need the guarantee that your data will not be used by Google.

Paid Tier (Credit Card Required) — Higher Limits, More Privacy

If the free tier’s rate limits are too restrictive, you can enable billing. There are two situations:

A. New Google Cloud users — $300 free credits for 90 days. If you have never used Google Cloud before, Google offers $300 of free credits valid for 90 days when you create a billing account. You must provide a credit card for identity verification, but you will not be charged unless you manually upgrade to a full paid account after the trial. When the 90 days pass or the $300 are spent, your account is simply paused — no automatic charges. To set this up, visit console.cloud.google.com/billing.

B. Pay-as-you-go. After the $300 trial, or if you already have a Google Cloud account, you pay only for what you use. The costs are extremely low for academic work:

Exact Cost Estimates for LibrAIry Operations

All costs below use Gemini 2.5 Flash (recommended), at March 2026 pricing of $0.15 per million input tokens and $0.60 per million output tokens. One token ≈ 4 characters of English text.

To put this in perspective: 1,000 AI Credits allow you to perform over 1,000 metadata extractions, or 150 full-text chat conversations, or 50 multi-article syntheses.

How to Monitor Your Spending

LibrAIry displays an estimated cost in the status bar and in the Network & AI Settings dialog. This is calculated from Google’s published token prices and is generally accurate, but it is an estimate — LibrAIry cannot know your exact Google billing status (free tier, $300 trial, or pay-as-you-go).

To see your actual billing, visit Google Cloud Console → Billing. This is the only authoritative source for what Google will charge you. We recommend setting up budget alerts in the Google Cloud Console (for example, $5/month) so you receive an email if your spending exceeds a threshold you define.

── Ollama ──

Option 3: Ollama (Local or Remote AI, Completely Free and Private)

Professional License only

What is Ollama?

Ollama is a free, open-source application that lets you run Large Language Models (LLMs) directly on your own computer or on another machine on your network. An LLM is the same type of artificial intelligence that powers tools like ChatGPT, Google Gemini, or Claude — it understands and generates natural language, making it capable of analyzing scientific texts, answering questions about your papers, and synthesizing information across multiple articles.

When you use Ollama with LibrAIry, the AI runs on your machine or on a server you control. No data is ever sent to any external cloud service. This makes it the ideal option if you work with confidential documents, unpublished research, or sensitive data. It is also completely free — there are no usage limits, no API costs, and no subscription required. The only requirement is that the machine running Ollama has enough resources to run the models (see below).

The main trade-off compared to cloud-based AI (LibrAIry Cloud or Google Gemini) is that local models are generally slower and may produce slightly less sophisticated results, since the models that can run on a personal computer are smaller than those running on powerful cloud servers. However, for most academic tasks — metadata extraction, paper summarization, comparative analysis — local models perform very well. If your own computer is not powerful enough, you can also connect LibrAIry to Ollama running on a more powerful machine on your network (see “Using a Remote Ollama Server” below).

What are LLMs and how do they differ ?

Large Language Models come in different sizes, measured in billions of parameters (abbreviated as "B"). The number of parameters roughly corresponds to how much knowledge and reasoning capability the model has:

Larger models produce better results but require more disk space, more memory (RAM), and more processing power. If your computer has a dedicated NVIDIA GPU (graphics card), models that fit entirely in the GPU's memory (VRAM) will run significantly faster. If a model is too large for your GPU, Ollama will automatically split the workload between your GPU and your system RAM, which is slower but still works.

You can browse all available models at ollama.com/library. New models are released regularly by companies like Google (Gemma), Meta (Llama), Microsoft (Phi), Alibaba (Qwen), Mistral AI, and others. LibrAIry's model selector shows the most popular and well-tested options, but you can install any model available in the Ollama library.

System Requirements

Ollama works on Windows 10/11, macOS, and Linux. The minimum and recommended specifications are:

If your system has less than 4 GB of RAM or fewer than 2 CPU cores, local AI will not work reliably. In that case, use LibrAIry Cloud or Google Gemini instead.

Installing Ollama from LibrAIry (Recommended)

LibrAIry includes a built-in setup assistant that handles the entire Ollama installation process for you. You do not need to use the command line or visit any website — everything is done from within LibrAIry's interface.

1. Open the Network & AI Settings. In LibrAIry, go to 📂 File → 🌐 Network & AI Settings. In the "AI Backend" section at the top, select 🖥️ Ollama. Two sub-options appear: 🖥️ This PC (to run Ollama locally) and 🌐 Remote Server (to connect to Ollama running on another machine). Select 🖥️ This PC to install Ollama on this computer.
2. Download and install Ollama. If Ollama is not yet installed on your computer, the settings panel will display a "Set up Local AI" screen showing your hardware specifications (CPU, RAM, GPU). Click the "⬇️ Download & Install Ollama" button. LibrAIry will automatically:
   • Download the Ollama installer (~100 MB) — a progress timer shows the elapsed time.
   • Install Ollama silently in the background (no separate installer window appears).
   • Start the Ollama service automatically.
   The entire process typically takes 1–3 minutes depending on your internet connection. A progress bar with a timer keeps you informed of each phase (downloading, installing, starting).
3. Choose an AI model to download. Once Ollama is installed and running, LibrAIry displays a model selection screen. All available models are listed in a dropdown menu, organized by size category (Ultra-light, Small, Medium, Large, Very Large). Each model shows:
   • A compatibility indicator: ✅ Runs well on your system, 🟡 Should work (may be slower), 🟠 Will be slow, or 🔴 Too heavy for your hardware.
   • The disk space required (e.g., "3.3 GB").
   • A ⭐ RECOMMENDED label on the model that LibrAIry considers the best choice for your specific hardware.
   Select the model you want (the recommended one is pre-selected) and click "⬇️ Download". A progress bar shows the download progress, speed (in MB/s), and estimated time remaining. Model downloads can take anywhere from 30 seconds (for a 1B model on a fast connection) to 15+ minutes (for a large model on a slower connection).
4. You are ready. Once the model is downloaded, the settings panel switches to the normal view, showing your installed model(s) and a "Test" button to verify everything works. Click 💾 Save & Close.

Installing Ollama Manually (Alternative)

If you prefer to install Ollama yourself (for example, on a Linux server or if the automatic installation does not work), you can do so:

1. Visit ollama.com/download and download the installer for your operating system.
2. Run the installer and follow the on-screen instructions.
3. Open a terminal (on Windows: press the Windows key, type cmd, and press Enter) and type: ollama pull gemma3:4b This downloads the Gemma 3 4B model. Replace gemma3:4b with any model name from ollama.com/library.
4. Once downloaded, go to LibrAIry → 📂 File → 🌐 Network & AI Settings → select Ollama → 🖥️ This PC. Your installed model(s) will appear in the dropdown.

Using a Remote Ollama Server

If your own computer is not powerful enough to run AI models efficiently (for example, a laptop without a dedicated GPU), but you have access to a more powerful machine on your network — such as a lab workstation, a colleague’s desktop with a high-end GPU, or a dedicated server — you can run Ollama on that remote machine and connect LibrAIry to it over the network.

This gives you the best of both worlds: the speed and quality of a powerful GPU-equipped machine, with no cloud service involved — all data stays within your local network.

Setting up the remote machine

1. Install Ollama on the remote machine. Visit ollama.com/download and follow the instructions for the remote machine’s operating system (Windows, macOS, or Linux).
2. Pull one or more models on the remote machine. Open a terminal and run, for example: ollama pull gemma3:12b Since the remote machine is presumably more powerful, you can install larger, higher-quality models than what your own laptop could handle.
3. Configure Ollama to accept remote connections. By default, Ollama only listens on 127.0.0.1 (localhost), which means it refuses connections from other computers. You need to set the OLLAMA_HOST environment variable to 0.0.0.0 so that Ollama listens on all network interfaces:
   • Windows: Open System Settings → Environment Variables → add a new system variable: name = OLLAMA_HOST, value = 0.0.0.0. Then restart the Ollama service.
   • Linux: Edit the Ollama systemd service file: sudo systemctl edit ollama Add the line Environment="OLLAMA_HOST=0.0.0.0" under [Service], save, then restart: sudo systemctl restart ollama
   • macOS: Run in a terminal: launchctl setenv OLLAMA_HOST "0.0.0.0" Then restart the Ollama application.
4. Note the remote machine’s IP address. On Windows, open a command prompt and type ipconfig. On Linux or macOS, type hostname -I or ifconfig. You need the local network IP (typically something like 192.168.1.xxx or 10.0.0.xxx).
5. Verify that the remote machine is reachable. From your own computer, open a web browser and go to http://<remote-IP>:11434. You should see a message like “Ollama is running”. If you get a connection error, check your firewall settings — port 11434 must be open on the remote machine.

Connecting LibrAIry to the remote server

1. In LibrAIry, go to 📂 File → 🌐 Network & AI Settings.
2. Select 🖥️ Ollama as the AI backend.
3. Select 🌐 Remote Server (instead of “This PC”).
4. Enter the remote machine’s URL in the Server URL field — for example, http://192.168.1.42:11434.
5. Click 🔗 Connect. LibrAIry will contact the remote Ollama server and, if successful, display the list of models available on that machine. Select the model you want to use.
6. You can click 🧪 Test to send a quick test prompt and verify the connection speed.
7. Click 💾 Save & Close. From now on, all AI operations (Chat, Synthesis, metadata extraction) will be processed by the remote Ollama server.

Managing Models

You can install multiple models and switch between them at any time. In the Network & AI Settings panel, the Ollama section (when Ollama is running) shows:

• Installed models — displayed as tags with a ✕ button to delete each one. Deleting a model frees the disk space it was using. You can always re-download it later.
• "Add a model" section — a categorized dropdown (identical to the initial selection screen) lets you download additional models with the same ✅🟡🟠🔴 compatibility indicators. You can also type any model name manually to install models not in the preset list.
• "Test" button — sends a simple test message to the selected model and displays the response time, letting you verify the model is working and giving you a sense of its speed on your hardware.

Recommended Models for Academic Work (March 2026)

The following models are well-suited for use with LibrAIry. The "best choice" depends on your hardware — LibrAIry automatically detects your system configuration and highlights the most appropriate option.

Cloud Mode for CPU-Only Systems

If your computer does not have a dedicated GPU (or has a GPU with less than 4 GB of VRAM), running AI models locally with Ollama will be very slow — not just for metadata extraction, but for all AI operations including Chat and Synthesis. On a typical laptop without a GPU:

• Metadata extraction frequently fails (the model cannot produce valid JSON reliably on CPU).
• AI Chat responses can take several minutes instead of seconds.
• AI Synthesis can take 10–30 minutes for a single request, and may appear to freeze.

To solve this, LibrAIry offers a Cloud Mode that redirects all AI operations to LibrAIry Cloud. This uses your AI credits but provides fast, reliable results on any hardware.

💬 AI Chat

AI Chat allows you to have a conversation about your articles in natural language. You type a question, and the AI reads the relevant articles from your library and generates an answer based on their actual content. The AI does not search the internet and does not make up information — it works exclusively with the papers you provide.

Opening Chat

Go to 🤖 AI Tools → 💬 AI Chat. A new Chat window opens. If you have already selected articles using the checkboxes in the main table, those articles are automatically pre-loaded as context for the conversation. If no articles are selected, the Chat window will ask you to specify which articles you want to discuss.

Referencing Articles in Your Messages

You can refer to specific articles in your messages using three different methods:

Analysis Modes

In the Chat window, you can choose between two analysis modes that control how much of each article the AI reads:

The "Abstract only" mode is recommended for most questions. It is fast, inexpensive, and works well for questions about methodology, main findings, and general comparisons. The "Full text" mode is useful when you need detailed information that is only found in the body of the article (for example, specific data points, detailed statistical results, or information from the discussion section).

Example Questions

Here are some examples of questions you can ask in AI Chat:

• "What are the main findings across all selected articles?"
• "Which papers use machine learning methods, and which use traditional statistical methods?"
• "Summarize the key contributions of #3, #7, and #12."
• "Do any of these papers contradict each other? If so, on what points?"
• "What limitations are commonly mentioned across these studies?"
• "Compare the sample sizes and methodologies of Smith (2023) and Johnson (2024)."

Token Estimation

Before sending a message, Chat displays an estimate of the number of tokens that will be used and the approximate cost (if you are using a paid AI service). This helps you make informed decisions about which analysis mode to use and how many articles to include. Remember that using fewer articles or switching to abstract-only mode will reduce both the cost and the response time.

🔬 AI Synthesis

AI Synthesis is a tool that generates a structured literature review from a group of selected articles. Unlike AI Chat, which is an interactive conversation, Synthesis produces a complete document in a single operation. The output is a Word document (.docx) that you can edit, format, and integrate into your own writing.

Opening Synthesis

Go to 🤖 AI Tools → 🔬 AI Synthesis. If you have already selected articles using checkboxes in the main table, Synthesis will use those articles directly. If no articles are selected, a selection dialog will appear where you can choose articles.

Selecting Articles

There are three ways to select articles for Synthesis:

• Checkboxes (recommended) — Select articles in the main table before opening Synthesis. This is the fastest and most intuitive method.
• #N references — In the selection dialog, type #3, #5, #12 to pick specific articles by their row number.
• Author (Year) references — Type Smith (2024), Lee (2023) to pick articles by their citation.

You can also click the All button in the selection dialog to include all currently displayed articles (useful after a search that has narrowed down the list to a specific topic).

Configuration

Before starting synthesis, you can adjust two parameters:

• Method — Choose between "Abstract only" (faster, uses only abstracts) and "Full text" (more comprehensive, reads the complete articles). Abstract-only mode is sufficient for most purposes and is significantly faster.
• Number of sections — Choose how many thematic sections you want in the output document (between 2 and 8). The AI will identify the main themes across your articles and organize the review into that many sections.

What the Output Looks Like

Synthesis produces a professional Word document (.docx) that typically includes:

• A title and introduction summarizing the scope of the review.
• Thematic sections, each covering a specific topic identified across your articles. The sections include cross-paper comparisons and synthesis — not just summaries of individual papers.
• References to specific articles by author and year throughout the text.
• A conclusion section highlighting key findings and research gaps.

The document is formatted and ready for editing. You can open it in Microsoft Word, Google Docs, or LibreOffice Writer and modify it as needed.

📖 Glossary

This glossary defines technical terms used throughout LibrAIry and this manual. If you encounter an unfamiliar word, check here first.

Documents & Bibliography

Artificial Intelligence

Technical Terms

⚙️ Settings

Network & AI Settings

Access these settings through 📂 File → 🌐 Network & AI Settings. This is where you configure which AI service LibrAIry uses for Chat, Synthesis, and AI metadata fallback. The three options (LibrAIry Cloud, Google Gemini, and Ollama) are described in detail in the “Setting Up AI” section of this manual. When using Ollama, you can choose between running it on this PC or connecting to a remote server.

Interface Colors

Access through 📂 File → Preferences → 🎨 Interface Colors. This opens a dedicated settings window where you can customize the colors of the header, menu bar, search area, and article table. You can modify each color individually using a color picker or by entering a hex color code. After saving your changes, restart LibrAIry for them to take effect.

Fonts & AI Display

Access through 📂 File → Preferences → 📏 Fonts & AI Display. A slider lets you increase or decrease the global font size. This can be useful if you find the default text too small or too large for your screen. This window also lets you choose the AI Response Colors palette, which controls how headings, bold text, and other formatting appear in AI Chat and Synthesis responses. Four palettes are available: Tokyo Night (cool violet and blue), Warm Academic (earthy amber tones), Ocean Breeze (teal and marine), and High Contrast (vivid and saturated for maximum readability).

Theme

Click the 🌙 Dark / ☀️ Light button in the top-right corner of the menu bar to switch between dark and light themes. The dark theme (default) is easier on the eyes in low-light environments, while the light theme may be preferable in bright conditions.

🔑 License & Pricing

LibrAIry is a paid application with a free trial period. All licenses are one-time purchases — there is no subscription and no recurring fee. Here is what each license tier offers:

Which License Should You Choose?

Trial is for evaluating LibrAIry. It gives you 30 days with full features but limits you to 50 articles. This is enough to test all the features and decide if LibrAIry suits your workflow.

Personal is the best choice for most individual users. It removes the article limit, includes 8,000 cloud AI credits (enough for thousands of metadata extractions and hundreds of chat questions), and can be activated on two computers (for example, your desktop and your laptop). All future updates are included.

Professional is designed for advanced users who want maximum flexibility. It includes the same 8,000 cloud AI credits as the Personal license, but also unlocks two additional AI backends: your own Google Gemini API key (for unlimited cloud AI at very low cost) and Ollama (for completely free, private AI — either locally or on a remote server you control). This gives you complete control over your AI usage and data privacy, with the cloud credits as a convenient fallback. It can be activated on three computers.

Managing Your License

You can view and manage your license at any time by going to 📚 LibrAIry → 🔑 License. This dialog shows your current license type, status, machine ID, and (for trial licenses) the number of days remaining. You can also activate a new license key or deactivate your current license if you need to move it to a different computer.

AI Credits & Tokens

If you have a Trial, Personal, or Professional license using LibrAIry Cloud, your license includes a certain amount of AI credits. These credits are used when you access AI features through LibrAIry Cloud (the built-in cloud AI proxy). Each AI operation (metadata extraction, chat question, synthesis generation) has a cost that is deducted from your credits.

Your remaining credits are displayed in the status bar at the bottom of the main window. The indicator shows both the remaining amount and the total (for example, "5,200 / 8,000 Credits"). You can click on it to refresh the display. For Professional licenses using a custom API key, the status bar displays your Token usage instead.

If your credits run out, you can purchase additional credit packs (LibrAIry → License). Professional license holders can also switch to using their own Google Gemini API key or Ollama for unlimited AI usage at no additional cost. The AI backend can be changed at any time in the Network & AI Settings.

🔧 Troubleshooting

Metadata Extraction Issues

OCR Issues

AI Issues

General Issues

❓ Frequently Asked Questions