agentq - advanced reasoning and learning for autonomous AI agents

agentq utilises various kinds of agentic architectures to complete a task on the web reliably. it has

1. a planner <> navigator multi-agent architecutre
2. a solo planner-actor agent
3. an actor <> critic multi-agent architecture
4. actor <> critic architecture + monte carlo tree search based reinforcement learning + dpo finetuning

this repo also contains an oss implementation of the research paper agent q - thus the name.

setup

1. we recommend installing poetry before proceeding with the next steps. you can install poetry using these instructions

2. install dependencies

poetry install

3. start chrome in dev mode - in a seaparate terminal, use the command to start a chrome instance and do necesssary logins to job websites like linkedin/ wellfound, etc.

for mac, use command -

sudo /Applications/Google\ Chrome.app/Contents/MacOS/Google\ Chrome --remote-debugging-port=9222

for linux -

google-chrome --remote-debugging-port=9222

for windows -

"C:\Program Files\Google\Chrome\Application\chrome.exe" --remote-debugging-port=9222

4. set up env - add openai and langfuse keys to .env file. you can refer .env.example. currently adding langfuse is required. If you do not want tracing - then you can do the following changes

   • directly import open ai client via import openai rather than from langfuse.openai import openai in the ./agentq/core/agent/base.py file.
   • you would also have to comment out the @obseve decorator and the below piece of code from the run function in the same file

   langfuse_context.update_current_trace(
               name=self.agnet_name,
               session_id=session_id
         )

5. run the agent

python -u -m agentq

run evals

 python -m test.tests_processor --orchestrator_type fsm

generate dpo pairs for RL

python -m agentq.core.mcts.browser_mcts

citations

a bunch of amazing work in the space has inspired this.

@misc{putta2024agentqadvancedreasoning,
title={Agent Q: Advanced Reasoning and Learning for Autonomous AI Agents},
author={Pranav Putta and Edmund Mills and Naman Garg and Sumeet Motwani and Chelsea Finn and Divyansh Garg and Rafael Rafailov},
year={2024},
eprint={2408.07199},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2408.07199},
}

@inproceedings{yao2022webshop,
  bibtex_show = {true},
  title = {WebShop: Towards Scalable Real-World Web Interaction with Grounded Language Agents},
  author = {Yao, Shunyu and Chen, Howard and Yang, John and Narasimhan, Karthik},
  booktitle = {ArXiv},
  year = {preprint},
  html = {https://arxiv.org/abs/2207.01206},
  tag = {NLP}
}

@article{he2024webvoyager,
title={WebVoyager: Building an End-to-End Web Agent with Large Multimodal Models},
author={He, Hongliang and Yao, Wenlin and Ma, Kaixin and Yu, Wenhao and Dai, Yong and Zhang, Hongming and Lan, Zhenzhong and Yu, Dong},
journal={arXiv preprint arXiv:2401.13919},
year={2024}
}

@misc{abuelsaad2024-agente,
title={Agent-E: From Autonomous Web Navigation to Foundational Design Principles in Agentic Systems},
author={Tamer Abuelsaad and Deepak Akkil and Prasenjit Dey and Ashish Jagmohan and Aditya Vempaty and Ravi Kokku},
year={2024},
eprint={2407.13032},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2407.13032},
}