Requirements:
- AnythingLLM set with OpenRouter as LLM provider and the Google Gemini 2.0 Flash Lite model
- a workspace called "test_text", embedding the file secret.pdf into the model
- a workspace called "judge"
- a workspace called "generator"
- an API key generated from the AnythingLLM settings and saved into the system environment variables as ANYTHINGLLM_API_KEY
- branch test-rag, script test_rag_llmstudio.py, uncommeting the line
agent_anythingLLM.hack(attack_config=attack_config_anythingLLM)
and commenting the other line
Issue: even if the api_key is specified, there is a 403 error due to a wrong api key being used. However, the api_key is correct. This holds even if the judge is a local model (with a LMStudio for instance, that does not require an API key). This error is obtained as soon as the prompts are being generated.
Requirements:
agent_anythingLLM.hack(attack_config=attack_config_anythingLLM)and commenting the other line
Issue: even if the api_key is specified, there is a 403 error due to a wrong api key being used. However, the api_key is correct. This holds even if the judge is a local model (with a LMStudio for instance, that does not require an API key). This error is obtained as soon as the prompts are being generated.