Agent not working properly

[newway-anshul]

Agent is going in loop

king...INFO: 172.18.0.1:41494 - "GET /health HTTP/1.1" 200 OK backend | INFO: 172.18.0.1:41508 - "GET /latest_answer HTTP/1.1" 200 OK backend | INFO: 172.18.0.1:41486 - "GET /screenshots/updated_screen.png?timestamp=1751272732289 HTTP/1.1" 200 OK I will search for the current share price of TCS on a reliable financial website. backend | backend | json backend | { backend | "plan": [ backend | { backend | "agent": "Web", backend | "need": [], backend | "task": "Search for the current share price of Tata Consultancy Services (TCS) on a reliable financial website such as NSE India, BSE India, or Yahoo Finance." backend | } backend | ] backend | } backend | backend | Failed to make a plan. This can happen with (too) small LLM. Clarify your request and insist on it making a plan within json. backend | Failed to make plan. Retrying... Task: Search for the current share price of Tata Consultancy Services (TCS). backend | backend | json

and we are unable to stop it from front end.

[Fosowl]

What model are you using ?

[newway-anshul]

What model are you using ?

gpt-4.1-mini

[Fosowl]

I would advice against using model from openai as they are really bad at following instructions. "Failed to make plan. Retrying... " happen when the model didn't follow our prompting correctly. You might consider using Deepseek r1/Qwen/Llama3.1/Magistral (with deepseek being a first choice as prompt were optimized for it), you could use togetherAI if you don't have sufficient hardware.