Customer Support GPT

Try it out:

Founders and customer support representatives often get an influx of repetitive emails that could be easily automated using GPT-3. The challenge is that GPT-3 knowledge is general and doesn't have company-specific information.

Enter Customer Support GPT, it uses the company's internal knowledge such as frequently asked questions, articles, etc. to give company-specific answers. If it doesn't know the answer, it'll flag to be answered by a human.

How it works

To get company-specific responses, Customer Support GPT does the following:

  1. Parse and create an index of company data.
  2. When an incoming customer support query is presented, the index is searched for the most relevant result(s). If none are present, flag it to a human reviewer.
  3. This result(s) are passed into GPT-3 along with the incoming query, to return a personalized, relevant response.

Future work

  • Integrate with Zendesk, Crisp, and other providers to pre-generate answers and use in real life.
  • Improve the GPT-3 prompt to return results that are closer to how customer support agents communicate


  • Abdellatif - Ex-twitter engineer and founder of
  • Ahmed - 17-year-old hacker, creator of

Questions and Contact

We're happy that you've reached this far, if you'd like to use the tool or have any questions please don't hesitate to reach out


Built With

  • gpt-3
  • gptindex
  • openai
  • python
Share this project: