FlyPython recommends current Python projects only after a maintainer reviews the project's source, maintenance state, license, documentation, release history, and practical user value. This directory is intentionally not seeded with unverified or AI-generated recommendations.
Each reviewed project is one YAML file in this directory (<id>.yml). The
table below is generated from those files by tools/render_readmes.py; edit
the YAML, never the table. radar.json (repository root) is the
machine-readable export for consumers who pin a repository commit.
ai_familiarity records whether mainstream model training data covers the
project and its current API — see docs/CURATION_POLICY.md for the grading
rules.
| Project | Category | Status | AI familiarity | Why it matters | When not to use / Risk | Reviewed |
|---|---|---|---|---|---|---|
| fastapi fastapi/fastapi · MIT |
Web & APIs | stable |
AI: high | Production-standard ASGI framework with automatic OpenAPI docs, Pydantic validation, and dependency injection. | For non-HTTP services, or teams standardized on Django's bundled ORM/admin stack.<br>Risk: Ensure background tasks handle errors properly and use async endpoints responsibly; blocking calls inside async routes degrade the whole service. | 2026-09-02 |
| instructor 567-labs/instructor · MIT |
AI Tools | stable |
AI: medium | Production standard for extracting structured JSON from LLMs using Pydantic models with retry validation. | When your provider already enforces structured outputs natively and you need nothing beyond it.<br>Risk: Requires API keys for the target LLM providers; retry loops add latency and token cost. | 2026-09-02 |
| marimo marimo-team/marimo · Apache-2.0 |
Interactive Notebooks | rising |
AI: low | Reactive, pure-Python notebook stored as standard executable .py files with deterministic state execution. | When your workflow depends on Jupyter-only extensions, or kernel-state debugging is central to your process.<br>Risk: Requires a modern browser environment and replaces the Jupyter workflow rather than extending it. | 2026-09-02 |
| polars pola-rs/polars · MIT |
Data & Pipelines | stable |
AI: medium | High-performance DataFrame library built in Rust on Apache Arrow with lazy query optimization. | When your pipeline leans on the pandas ecosystem (accessors, narrow libraries) or you need index-heavy semantics.<br>Risk: API differs from pandas and memory layout is columnar; budget migration time rather than assuming drop-in parity. | 2026-09-02 |
| pydantic-ai pydantic/pydantic-ai · MIT |
AI Agents | rising |
AI: low | Model-agnostic agent framework prioritizing type-safe structured outputs, dependency injection, and testability. | When you need a stable long-lived API surface today, or heavy multi-agent orchestration features.<br>Risk: Rapidly evolving API surface; pin minor versions and re-run your evals on every upgrade. | 2026-09-02 |
| ruff astral-sh/ruff · MIT |
Code Quality | stable |
AI: high | 10-100x faster linter and formatter that unifies Flake8, Black, isort, and pyupgrade rules in a single configuration. | When a project depends on plugin ecosystems (e.g. Flake8 plugins) that have no Ruff equivalent yet.<br>Risk: Drop-in Black compatibility; rare syntax-parsing differences surface on unusual code bases. | 2026-09-02 |
| uv astral-sh/uv · MIT OR Apache-2.0 |
Tooling & Packaging | stable |
AI: medium | Extremely fast Rust-based package and project manager that replaces pip, pip-tools, venv, and pyenv with lockfile determinism. | When you must pin an existing pip-tools or Poetry workflow, or in air-gapped environments without wheel mirrors.<br>Risk: Actively maintained by Astral; relies on prebuilt binary wheels, so supply-chain review applies on upgrade. | 2026-09-02 |
radar.json— deterministic export of every reviewed project (pinned by website consumers together withcatalog.json).candidates.json— output oftools/radar_scan.py; raw discovery candidates with no descriptions and no status until a human reviews them.
Use the project proposal form
to suggest a project. An accepted record will use one of these lifecycle states:
new, rising, stable, major-update, experimental, or archived. "New"
describes a recent reviewed discovery, not an unverified quality claim.