Nissist: An Incident Mitigation Copilot based on Troubleshooting Guides

This repository contains the offical implementations of our paper Nissist: An Incident Mitigation Copilot based on Troubleshooting Guides.

We are honored to be accepted by ECAI 2024 Demo Track.

Nissist Setup

1. Create virtual environment

   Conda option:

   conda create -n tsgcopilot python=3.10
   conda activate tsgcopilot
   git clone **
   pip install -r requirements.txt

   Docker option:

   docker build -t tsgcopilot .
   docker run -p 2000:2000 tsgcopilot

2. Initialize the parameters for OpenAI and Kusto.

   You need to configure AOAI information before running the code.

   • copy config_example.yaml and rename to config.yaml and fill in your AOAI information and kusto information. Note that we offer two kusto login, if set KUSTO_LOGIN as AAD, you don't need to specify KUSTO_SECRET, KUSTO_AAD_APP_ID, and KUSTO_AUTHORITY_ID. If you donn't need KUSTO, just ignore it.
   • Note that the AOAI model needs to be new enough to support JSON format output, required to be version after 2023-10, refer to this json mode

3. Prepare Troubleshooting Guides

   We give an example of the TSG we use in example_data/TSGs. You need to convert your documents into the same structure, and please refer to tsg_reformulation folder for more details.

Run Nissist

You can run Nissist on Terminal or Interactive UI.

Run on Terminal

python TSG_Copilot.py

Run On Interactive UI

The interactive UI is implemented on ChainLit package.

• Run the Nissist

  First, you should start your TSG_Copilot API service, that means firstly run python main.py. And the service will listen on this port locally (http://127.0.0.1:2000/api/tsg_copilot).

  • Example request:

  {
      "query": "I got a incident that ...?",
      "conversation_id": "10000"
  }

  Note: use the same conversation_id if you would like to continue the conversation.

• Run the Web UI

  After that, run the following script in terminal chainlit run app.py, it will automatically open the webpage (http://localhost:8000/), and you can input your message to get the response of Nissist.

  • The function send_message inside app.py will call the API you have started.

  • As your want to start a new round of conversation, you can type 'exit' or just start 'New Chat'.

Intergrate Nissist into Taskweaver for Automation

1. Start up Nissist as the backend service, refer to Run On Interactive UI.

2. Clone and install the latest version of TaskWeaver.

3. Copy the ext_role/tsg_copilot directory to the TaskWeaver as the taskweaver/ext_role/tsg_copilot directory.

4. Copy all files in the project/plugins directory to the TaskWeaver project/plugins directory.

   Here we provide an example of the Kusto plugins to automate the execution of Kusto Query. You are required to set the Kusto configurations in the pull_kusto_data.yaml file.

5. Copy all files in the project/planner_examples directory to the TaskWeaver project/planner_examples directory.

6. Follow the instructions in the TaskWeaver QuickStart to start the TaskWeaver console chat.

Note that we should add the below new lines in the configuration file taskweaver_config.json, besides the default LLM configurations.

{
"session.roles": [
    "planner",
    "code_interpreter",
    "tsg_copilot"
],
"execution_service.kernel_mode": "local"
}

Nissit Demo Video

A demo is also provided.

https://github.com/AL-377/Nissist-Demo/blob/master/.asset/Nisst-demo.mp4

Citation

If you find this paper or code useful, please cite us:

@article{an2024nissist,
  title={Nissist: An incident mitigation copilot based on troubleshooting guides},
  author={An, Kaikai and Yang, Fangkai and Li, Liqun and Ren, Zhixing and Huang, Hao and Wang, Lu and Zhao, Pu and Kang, Yu and Ding, Hua and Lin, Qingwei and others},
  journal={arXiv preprint arXiv:2402.17531},
  year={2024}
}