[Improvement]Scenario-based LLM defaults and a curated model list

[earayu]

To improve user experience and optimize performance, we should provide different default LLM configurations and a curated list of options for various scenarios. The selectable list should be kept simple (e.g., less than 10 models).

Proposed Scenarios:

1. For Agentic tasks:

   • Needs: Models with strong reasoning, large context windows, and robust tool-use.
   • Recommended Defaults: gemini-2.5-pro, claude-4-sonnet.

2. For Graph Indexing:

   • Needs: Fast, cost-effective models with strong instruction-following.
   • Recommended Defaults: gemini-2.5-flash, gpt-4o-mini.

Implementation:

The configuration should be intuitive and easy to manage for users in both SaaS and local self-hosted environments.