Support Local LLMs / Bring Your Own Model (Ollama, LM Studio, custom endpoints) for Biblio AI

[thec0dewriter]

Feature Description

With the release of Biblio AI in bibliometrix / biblioshiny, AI-assisted interpretation of bibliometric results is now possible. However, the current implementation relies solely on the Google Gemini API.

I would like to propose expanding Biblio AI to support Bring Your Own Model (BYOM)—specifically allowing integration with local LLM runtimes like Ollama, LM Studio, LocalAI, or custom OpenAI-compatible local/self-hosted endpoints.

---

Why is this needed? (Use Cases & Benefits)

1. Data Privacy & Academic Compliance:

   • Many researchers and academic institutions work with unpublished manuscript drafts, proprietary literature reviews, or datasets subject to strict data governance (e.g., GDPR, institutional IRB guidelines). Sending raw bibliometric texts or summaries to cloud APIs (like Google Gemini) can violate these policies.
   • Local LLMs guarantee 100% data privacy since all prompts and outputs remain on the user's local machine.

2. Cost & Rate Limits:

   • API fees and quota/rate limits can restrict large-scale or high-frequency bibliometric analyses. Local models (e.g., Llama 3, Mistral, Qwen via Ollama) run with zero API costs and no rate limit throttling.

3. Flexibility & Customization:

   • Allows researchers to choose smaller/faster models for quick summaries or specialized open-source models fine-tuned for academic literature synthesis.

---

Proposed Solution / Technical Implementation

Add a configuration section in the Biblio AI settings panel (in both bibliometrix R functions and biblioshiny UI) with the following options:

• Provider Dropdown: Gemini (Default) | Ollama | OpenAI API / Compatible Endpoint | Custom
• Base URL: (e.g., http://localhost:11434 for Ollama or http://localhost:1234/v1 for LM Studio)
• Model Name: (e.g., llama3:8b, mistral, qwen2.5)
• API Key / Token: (Optional for local models, mandatory for remote providers)

R Ecosystem Implementation Ideas:

• If Biblio AI uses R packages for LLM integration under the hood, libraries like ellmer or httr2 provide native support for standard OpenAI-compatible endpoints, Ollama, and multiple cloud backends with minimal code refactoring.

---

Additional Context

• Target Audience: Non-coder researchers using biblioshiny as well as R power-users running bibliometrix programmatically.
• Open Source Compatibility: Adding local model support aligns strongly with bibliometrix's mission as an open-source research ecosystem.

Thank you for all your hard work on bibliometrix and biblioshiny!

[Amit-max-ui]

Hi @thec0dewriter , I would like to pick up this development. I will share the detail development plan for this. Let me know if it can be assigned to me??

[thec0dewriter]

Hello @Amit-max-ui
I'm not the maintainer of the project i just submitted as improvement idea.
Probably you can just make a fork and come up with a solution and submit a pull request