Preparing documents, running a model and checking its answers are three different jobs in an AI project. Today's GitHub Picks covers a tool for each: Docling, Ollama and Promptfoo. Recent updates expand document-processing options, increase Ollama's web-search limit and fix errors in AI testing.
Docling turns documents into usable input
A PDF can contain several columns, scanned text and tables. Simply copying its text can lose the order or relationships that an AI application needs. Docling, a project started at IBM Research Zurich, parses documents while identifying features such as page layout, reading order and table structure. It can export the result as Markdown, HTML or JSON for another application to use.
The October 3 release, version 2.133.0, adds a choice of model size for RapidOCR, an engine that reads text from images. It also fixes several extraction problems, including preserving extracted text when a batch in its code-and-formula processing fails. These are practical changes for teams turning reports or manuals into searchable material.
Docling's installation guide in the repository requires Python 3.10 or later and supports Windows, macOS and Linux. Local processing is available. Its code has an MIT license, but the models it uses can have separate licenses. Teams still need to check whether the extracted text and tables preserve the meaning of their own documents.
Explore Docling on GitHub · Documentation and examples
Ollama makes models available to an application
Ollama lets developers run supported models and connect them to software through an API, the interface an application uses to send a request and receive an answer. It provides downloads for Windows, macOS and Linux, along with Python and JavaScript libraries. This makes it a useful starting point for a team that wants to try an open model in a prototype.
Its September 29 release, version 0.35.1, lets models using web search make up to ten searches per response, up from three. It also changes how supported tasks are reported to applications. Specialized decision models are now identified separately, so an application can avoid presenting them as general-purpose chatbots.
Ollama supports both local operation and cloud features. Its FAQ provides a local-only setting that disables cloud models and web search. Local operation depends on the computer's memory and the selected model; the Ollama software's MIT license does not replace that model's own terms.
Explore Ollama on GitHub · Local operation and hardware guidance
Promptfoo checks answers across repeatable tests
One convincing answer does not tell a product team how an AI application behaves on the next hundred requests. Promptfoo lets developers define test inputs and checks, compare prompts or models, and inspect the results side by side. It also supports red teaming: deliberately challenging an application to look for unsafe or unwanted behavior.
The project's September 18 release, version 0.123.1, includes fixes to how some tests handle failures. One fix addresses cases where a grader, the component judging an answer, could fail while the test incorrectly reported a pass. That distinction matters when a team uses automated results to decide whether a change is ready to ship.
Promptfoo is MIT-licensed software, but evaluations can call paid model APIs. Its quick-start instructions require credentials for most providers. A useful test set also needs examples and expected behavior drawn from the intended product: a passing result covers those tests, not every situation users may encounter.
Explore Promptfoo on GitHub · Getting-started guide
Start with the problem your prototype actually has: difficult source documents, model access or inconsistent answers. These projects address different parts of that work; choosing all three is not a prerequisite for getting started.
GitHub Picks is GTN's editorial selection based on project documentation and release notes. This issue is not a popularity ranking or a hands-on performance review.
Sources
- Docling: version-specific project guide
- Docling 2.133.0 release notes
- Docling code license
- Ollama: version-specific project guide
- Ollama 0.35.1 release notes
- Ollama code license
- Ollama FAQ: local operation, cloud features and memory
- Promptfoo: version-specific project guide
- Promptfoo 0.123.1 release notes
- Promptfoo code license
Photo: Nenad Stojković / Shixart1985, Wikimedia Commons, CC BY 2.0. Illustrative workplace photograph; resized for display.