fixed the changes to joint names and added env variables. - #26
fixed the changes to joint names and added env variables.#26christymanthara wants to merge 4 commits into
Conversation
KhushiSharma22
left a comment
KhushiSharma22
left a comment
There was a problem hiding this comment.
Thanks for this! Nice addition; routing the planner through OpenRouter via the OpenAI Agents SDK opens up a lot of model flexibility, and the fallback-to-Anthropic design keeps existing setups working.
Everything looks great; there's just one quick change that might block the run in case of default runs. I can fix it for you in case you want that, dropping a comment here. Approving in principle :)
| CYBERWAVE_API_KEY = os.environ.get("CYBERWAVE_API_KEY") | ||
| ANTHROPIC_API_KEY = os.environ.get("ANTHROPIC_API_KEY") | ||
| OPENROUTER_API_KEY = os.environ.get("OPENROUTER_API_KEY") | ||
| OPENROUTER_MODEL = os.environ.get("OPENROUTER_MODEL", "google/gemini-2.5-flash-lite-preview-07-15:free") |
There was a problem hiding this comment.
The default OPENROUTER_MODEL here (google/gemini-2.5-flash-lite-preview-07-15:free) disagrees with the default in planner.py (openai/gpt-5.4-mini, in _openrouter_model_name). This file only uses the value for the banner / self-check display, while planner.py uses its own default for the actual call.
Net effect: if a user sets OPENROUTER_API_KEY but leaves OPENROUTER_MODEL unset, the banner reports Gemini while the planner actually runs gpt-5.4-mini. Can we pull the default into one shared constant (or have planner.py import OPENROUTER_MODEL from here) so the displayed model always matches what runs?
There was a problem hiding this comment.
fixed with a common planner config. good catch. thanks
…d env.example with new env variables
2 participants
This PR adds Openrouter as a model provider and openai agents sdk framework for the agent in addition to already existing Anthropic. This is limited to the nl_arm_controller example in the repo