Skip to main content

​
Build & Development

# Install dependencies
poetry install

# Run API locally
poetry run uvicorn server.main:app --reload

# Run with Docker (recommended)
docker compose up -d

# Run tests
docker compose run --rm test        # Docker (preferred)
poetry run pytest tests/unit -v    # Local unit tests only

​
Coding Conventions

  • Python 3.12, PEP 8, 4-space indentation
  • Type hints on all function signatures
  • Naming: snake_case (modules/functions), PascalCase (classes)

​
Evaluation Framework

Persona includes a comprehensive evaluation framework for testing long-term memory systems against academic benchmarks.

​
Supported Benchmarks

BenchmarkWhat it TestsReference
LongMemEvalTemporal logic, multi-session aggregationICLR 2025
PersonaMemFactual precision, personalizationCOLM 2025

​
Running Evaluations

# Install dependencies
poetry install

# Download datasets
poetry run python evals/scripts/download_personamem.py

# Quick test (15 questions)
poetry run python -m evals.cli run \
  --benchmark longmemeval \
  --samples 5 \
  --seed 42

# Full evaluation (340 questions)
poetry run python -m evals.cli run \
  --config evals/configs/full_eval.yaml \
  --golden-set

​
Analyzing Results

# Summary report
poetry run python -m evals.cli analyze run_20241221_143052 --summary

# Filter by question type
poetry run python -m evals.cli analyze run_20241221_143052 --type multi-session

​
License

MIT License. See LICENSE for details.