Pharos-ai/Meridian

Meridian: A Tool-Integrated Research Chat Model

Meridian is a 12 billion parameter research-focused chat model developed by Pharos-ai, specifically engineered for integration within applications that provide live search, browsing, and source-reading capabilities. Unlike general-purpose chatbots, Meridian is designed to operate in an environment where it can actively use tools to gather and process information before formulating a response.

Key Capabilities

• Tool Calling: Meridian can emit structured tool calls (e.g., browser.search, browser.open) to external applications for real-time information retrieval.
• Source Grounding: It processes retrieved source text from tools and generates answers that are grounded in this information.
• Citation Generation: The model is trained to cite sources inline (e.g., [source:1]) within its responses, enhancing transparency and verifiability.
• Concise Answering: Meridian prioritizes concise answers, avoiding verbose or rambling outputs, especially when responding to factual or source-sensitive questions.
• App-Integrated RAG: It facilitates advanced Retrieval Augmented Generation (RAG) experiences by integrating seamlessly with an application's data retrieval and processing pipeline.

Good For

• Developing research assistant workflows that require up-to-date or verified information.
• Building live search and browsing agents that interact with web resources.
• Creating AI assistants that provide citation-style answers.
• Applications requiring tool-call formatting for external actions.
• Generating short, factual summaries from provided source documents.

Meridian is not intended as a standalone offline chatbot but thrives when connected to robust external tools for search, document retrieval, and source validation.