pawsona 0.1.0a2

Train personalities, not commands.

Pawsona

Train personalities, not commands.

Pawsona is a terminal-based pet personality and training simulator inspired by reinforcement learning, memory, adaptation, generalization, and overfitting.

The first public version focuses on dogs. The behavior model is intentionally simple: it is a small scoring engine for playable training loops, not a biologically realistic simulator.

Current Status

This repository is in the early CLI phase. Pawsona can load built-in pet profiles, inspect their model state, create new YAML profiles, generate simple behavior-model actions, and persist trained skills and memory between CLI runs.

Try It

After release, the main install command will be:

pip install pawsona

Primary CLI:

pawsona --help

Short alias:

paw --help

If paw conflicts with another command on your machine, use pawsona.

Environment

Create and activate the conda environment:

conda create -n paw python=3.11 pip
conda activate paw

Install the project components from requirements.txt:

python -m pip install -r requirements.txt
pawsona --help

Inspect A Pet

pawsona inspect hermes
pawsona inspect hera

Inspect only the raw base YAML profile:

pawsona inspect hermes --base

Create A Pet

pawsona create

Trait questions use a 1 to 5 scale and are normalized into 0.0 to 1.0 YAML values.

Generate An Action

pawsona act hermes
pawsona act hera

Train In The Terminal

pawsona play hermes
pawsona play hera

Training sessions run for 10 rounds by default. Feedback options are reward, praise, ignore, and correct.

Training updates are saved to saves/<pet>.json. Base YAML profiles are not modified.

Run a temporary training session without writing a save file:

pawsona play hermes --no-save

Project Shape

pawsona/
  cli.py
  behavior.py
  pet.py
  simple_yaml.py
  state.py
  training.py
pets/
  hermes.yaml
  hera.yaml
  template.yaml
  community/
saves/
docs/