MatrAIx Persona Testing

Create and test an AI persona to mimic you

A local Mac app for chatting with a personalised agent, filling gaps in its persona and measuring how closely its answers match yours.

View project on GitHub macOS - ChatGPT sign-in - no separate API key

What it does

Build a replica of your personality as a collection of dimensions. Measure how a ChatGPT agent with this persona matches your answers on a set of surveys, and how its answers improve the more dimensions you fill in.

Details

On first start up, the app asks you to fill in a list of personality dimensions from your ChatGPT history and memory.

Then there are three functionalities:

  1. Chat directly with the personalised agent based on your local persona YAML.
  2. Update persona by answering questions on the missing personality dimensions.
  3. Validate how similar the agent's answers are to yours on a set of questions.
    • You answer the questions once, then they are used as the benchmark.
    • It is recommended to run Validation each time you add a batch of new personality dimensions.
    • The similarity of the agent's answers to your benchmark is summarised and visualised.

Project notes

Local and lightweight

The app runs on localhost, while persona files, survey answers and validation results stay in ignored local paths. Model-backed onboarding, chat and validation send approved inputs to OpenAI through Codex.

Independent adaptation

This is an independent testing adaptation of the original MatrAIx-Persona-8B project.

Normal use needs no separate OpenAI API key or local model download. The app, installation steps and technical documentation remain in the existing GitHub repository.

Screenshots

MatrAIx chat showing a question and a reply from Digital Alfred
Chat. Have a direct conversation with the agent defined by your local persona.
MatrAIx Update persona screen asking how directly the user communicates
Update persona. Answer focused questions that fill unresolved personality dimensions and save them to the active persona.
MatrAIx Validation results showing benchmark match and agent consistency charts
Validation. Compare your benchmark answers with repeated agent runs, then track answer similarity and consistency.