Skip to main content

Whole-repository Quality-Diversity optimization for real git codebases.

Project description

Loreley

Whole-repository Quality‑Diversity optimization for real git codebases.

Loreley is a distributed system that evolves entire git repositories (the unit of search is a git commit). It continuously samples base commits, asks external planning/coding agents to implement repo-wide changes, evaluates the result with your evaluator, and stores metrics plus a MAP‑Elites archive in Postgres for later sampling and reuse.

Why use it

  • Whole-repo evolution: cross-module refactors and “production-style” changes are first-class.
  • QD-native (MAP-Elites): keeps multiple high-performing but different solutions instead of a single champion line.
  • Learned behaviour space: behaviour descriptors come from repo-state code embeddings (cached by git blob SHA), not hand-crafted heuristics.
  • Production loop: scheduler + Redis/Dramatiq workers + Postgres, with preflight checks, logs, and reproducible git history.

Quick start (local)

Requirements: Python 3.11+, uv, Git (worktrees), PostgreSQL, Redis, and an OpenAI-compatible API for embeddings (OPENAI_API_KEY). You also need:

  • Planning/coding backend: default is the codex CLI on PATH (override via WORKER_PLANNING_BACKEND / WORKER_CODING_BACKEND).
  • Evaluator plugin: WORKER_EVALUATOR_PLUGIN=module:callable that runs unattended and returns structured metrics.
git clone <YOUR_FORK_OR_ORIGIN_URL> loreley
cd loreley
uv sync
docker compose up -d postgres redis

cp env.example .env
# Minimal required vars:
# - OPENAI_API_KEY
# - MAPELITES_EXPERIMENT_ROOT_COMMIT=<git commit hash>
# - SCHEDULER_REPO_ROOT=/abs/path/to/your/target-git-checkout
# - WORKER_REPO_REMOTE_URL=<git remote URL with push access>
# - WORKER_EVOLUTION_GLOBAL_GOAL="..."
# - WORKER_EVALUATOR_PLUGIN=module:callable
# - (optional) WORKER_EVALUATOR_PYTHON_PATHS=["/abs/path/to/plugin_dir"]
# - (optional) SCHEDULER_STARTUP_APPROVE=true  # skip interactive startup approval

uv run loreley doctor --role all
uv run loreley scheduler
uv run loreley worker

Optional UI (read-only)

uv sync --extra ui
uv run loreley ui

Documentation

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

loreley-0.6.1a0.tar.gz (2.0 MB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

loreley-0.6.1a0-py3-none-any.whl (194.3 kB view details)

Uploaded Python 3

File details

Details for the file loreley-0.6.1a0.tar.gz.

File metadata

  • Download URL: loreley-0.6.1a0.tar.gz
  • Upload date:
  • Size: 2.0 MB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: uv/0.9.15 {"installer":{"name":"uv","version":"0.9.15","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"macOS","version":null,"id":null,"libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}

File hashes

Hashes for loreley-0.6.1a0.tar.gz
Algorithm Hash digest
SHA256 3db1cda2253565feb446fb9c19e80ad15066b0ea5c394780e0649ba78eefbd31
MD5 add3799b976c83fe38f03f6514293c8f
BLAKE2b-256 de0b10f6c296db7f0299c3bfc55689fce1ab1f049e40ce7c4381c893ff3ca2c7

See more details on using hashes here.

File details

Details for the file loreley-0.6.1a0-py3-none-any.whl.

File metadata

  • Download URL: loreley-0.6.1a0-py3-none-any.whl
  • Upload date:
  • Size: 194.3 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: uv/0.9.15 {"installer":{"name":"uv","version":"0.9.15","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"macOS","version":null,"id":null,"libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":null}

File hashes

Hashes for loreley-0.6.1a0-py3-none-any.whl
Algorithm Hash digest
SHA256 1da12235884d14c350d42485065ae67d0ab25fa1c177054eec374c0da713d12b
MD5 1ed31834d412cec127dc0e4962610e44
BLAKE2b-256 64f32e6255ed3266311d9e1a9f1532c84576ef44c07d13ae438292ed7e4d8089

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page