Scenario-based interaction with LLMs


I’d like to share my new project for interacting with LLM-based API, i.e. ChatGPT, with using MHBP Scenarios. Scenario is DAG-based definition of steps for prompting LLM-based API, collecting responses, making transformation of such results, creating sub steps.

Current capabilities:

  • Store definitions of APIs’ end-points, New API end-points can be easily registered
  • Agnostic to LLM - can work with any prompt-based API
  • Create Scenario with simple UI
  • Scenario is supporting DAG, Directed acyclic graph, execution
  • Aggregating results of prompts in final txt document
  • Mixing different LLM API in one Scenario
  • Application is in form for executable located at github, so no needs to publish your keys anywhere
  • Support multi-tenant-based access - many companies, many users
  • Deployment can be as desktop application, as SaaS, support different relational databases

Short video of using Scenario with real interaction with ChatGPT API:


  • Add Acceptance criteria to each Step which will be a separate call for LLM’s API for confirmation that Acceptance criteria was met.
  • Add a bulk changing of API used in Steps of Scenario

I will appreciate any feedback about MHBP Scenario

Free to use for very wide variety of cases - personal, research, non-profit, other cases.

Wiki - Home · sergmain/metaheuristic Wiki · GitHub

Sergio Lissner