Pull requests

• fix(evaluation): score judge verdicts by exact match instead of first letter

  #271·Linxiushen opened Sep 21, 2026·

  ·

• fix(inference): use the last <tool_call> block in react_agent (fixes #148)

  #270·Linxiushen opened Sep 21, 2026·

  ·
  1

• fix(inference): confine parse_file to file_root_path (absolute paths and ".." escape it)

  #269·Linxiushen opened Sep 21, 2026·

  ·

• docs: clarify WebWatcher local image and search URL setup

  #266·Hanabi9248 opened Sep 10, 2026·

  ·

• Add Exa as a search backend

  #263·kesku opened Jul 30, 2026·

  ·

• fix(ParallelMuse): repair stale call sites that crash compressed_reasoning_aggregation

  #262·Osamaali313 opened Jul 11, 2026·

  ·
  1

• fix: guard against IndexError/AttributeError on empty LLM choices (19 sites)

  #259·qizwiz opened May 17, 2026·

  ·

• feat: add Exa AI-powered search tool

  #255·tgonzalezc5 opened May 4, 2026·

  ·

• fix: correct pip install order so VLSearchImage is registered (fixes #234)

  #253·octo-patch opened Apr 24, 2026·

  ·
  1

  1

• fix: skip temperature param for OpenAI reasoning/newer models in summary tool

  #252·octo-patch opened Apr 19, 2026·

  ·
  1

• feat: add MiniMax as LLM provider with M3 as default model

  #247·octo-patch opened Mar 15, 2026·

  ·
  6

• Docu save cite

  #198·mqi6 opened Oct 29, 2025·

  ·

• refactor: optimize file handling in call method

  #144·NIDNASSER-Abdelmajid opened Sep 21, 2025·

  ·
  4

• Refactor react_agent.py to improve security and maintainability

  #142·Mirza-Samad-Ahmed-Baig opened Sep 20, 2025·

  ·
  3

• Manage the vLLM lifecycle

  #116·tobrun opened Sep 18, 2025·

  ·
  1

  2