Design Philosophy

Our design philosophy is based on open-source principles and governed by the desire to keep up with the rapid pace of LLM development. We adopt an agile style of developing modular components, aiming for quick iteration on proof-of-concept implementations. The rapid introduction of new models and functionalities (e.g., tool binding) necessitates a workflow that can quickly adapt to these changes.

This often involves the need of frequent refactorings of individual modules of the codebase, making modularity and separation of concerns a vital concern. Further, in order to facilitate collaboration and onboarding of new contributors, the codebase needs to be easily understandable and well-documented.

Feature Selectiveness

For the average research initiative, even when factoring in community contributions, it is not feasible to compete with the resources of large companies, which are numerous in the current LLM ecosystem. As a result, it often is not sensible to work on solutions that are subject to high company interests. For instance, early BioChatter developments included a workflow for automating podcast-like summaries of scientific papers; we discontinued this effort in light of later developments by Google to provide this same functionality inside their NotebookLM platform. We aim to continue to be mindful of this issue when deciding on which features to implement, which requires constant monitoring of the broader LLM ecosystem.