okapipy 0.4.0

Python OpenAPI Client Generator made OK

okapipy

okapipy turns an OpenAPI 3.x document into a typed, hierarchical Python client — sync and async, sharing one tree. It lifts flat paths into Namespaces, Collections, Resources, Singletons, and Actions, then emits a project where regenerated base/ code carries the wiring and a one-shot user layer carries your customizations. Re-run the generator after a spec change and your code is left strictly alone.

The generated client is built to be navigated by hover: every class docstring opens with a short summary and lists the children you can reach from it — so client. in your IDE shows the whole tree without opening a single file.

📚 Full documentation: https://ffaraone.github.io/okapipy/

Install

okapipy needs Python 3.12+.

pip install okapipy            # or: uv add okapipy

The first NLP-dependent run downloads the spaCy en_core_web_sm model (~12 MB) into ./.spacy/. To pre-warm it (recommended in CI):

okapipy fetch-language en

Use

Sanity-check the parse:

okapipy parse openapi.yaml

Generate a full client project. okapipy reads its settings from a small manifest you commit alongside your consumer code; scaffold it with okapipy init:

okapipy init openapi.yaml \
    --package acme.commerce --client-class CommerceClient

That writes ./okapipy.yml. Then:

okapipy generate --output ./my-client

This writes a runnable Python project under ./my-client: a regenerated src/acme/commerce/base/ tree (transport, models, vendored runtime, per-node base classes) plus one-shot subclass stubs you can customize. Re-running the command refreshes base/ and leaves your edits alone.

Every project-level setting (shape, templates_dir, output) and every per-spec setting (rules, strip_prefix) lives in okapipy.yml — see the Quick start for the full schema. The CLI keeps only --manifest, --output, --check (CI dry-run that exits non-zero on any drift), and --quiet.

Customize

okapipy decides what each path segment is using POS tagging plus a small heuristic registry. When the heuristics get it wrong — or when you need to fence off pieces of the spec — you decorate the spec with x-okapipy-* extensions, or carry the same overrides in a project-local rules file (useful when you don't own the OpenAPI document). Rules-file values win on every conflict.

Point a specs[] entry at the rules file via rules: in okapipy.yml:

specs:
  - namespace: ''
    source: openapi.yaml
    rules: okapipy.rules.yaml

Then run okapipy generate as usual.

OpenAPI extensions

openapi: 3.0.0
info: { title: Commerce API, version: 1.0.0 }

# Declare top-level folders so single-noun segments aren't misclassified.
x-okapipy-ns:
  - commerce
  - commerce/internal
  - settings

paths:
  /commerce/orders:                      # Plural → Collection
    get: ...
    post: ...

  /commerce/orders/{id}/submit:          # Verb → Action (NLP catches it)
    post: ...

  /me:
    x-okapipy-kind: singleton            # Force singleton; NLP can't tell apart from a namespace
    get: ...
    patch: ...

  /staff:
    x-okapipy-kind: collection           # spaCy thinks "staff" is singular — override

  /currencies:
    x-okapipy-paginated: false           # Override pagination heuristic
    get: ...

  /internal/debug:
    x-okapipy-exclude: "*"               # Drop every method on this path

  /orders/{id}:
    x-okapipy-exclude: [DELETE]          # Or just selected methods
    get: ...
    delete: ...

Allowed x-okapipy-kind values: namespace, collection, singleton, action. Path-item hints classify the segment and propagate to other paths walking through the same prefix. Operation-level x-okapipy-kind: action is narrower — it routes a single method to a synthetic Action without changing the segment classification.

Rules file

The rules file mirrors the spec extensions and is local-only (no URLs). JSON or YAML, same shape:

x-okapipy-ns:
  - commerce
  - settings

paths:
  /staff:
    x-okapipy-kind: collection
  /me:
    x-okapipy-kind: singleton
  /orders/{id}/submit:
    post:
      x-okapipy-kind: action
  /currencies:
    x-okapipy-paginated: false
  /internal/debug:
    x-okapipy-exclude: "*"
  /orders/{id}:
    x-okapipy-exclude: [DELETE]

For more — code customization, custom strategies, template overrides, client construction options — see https://ffaraone.github.io/okapipy/.

Using the generated client

from acme.commerce import CommerceClient
import httpx

with CommerceClient(
    base_url="https://api.example.com",
    auth=httpx.BearerAuth("..."),
) as client:
    # Iterate a collection — pagination is automatic.
    for order in client.commerce.orders:
        print(order.id, order.total)

    # Filter, sort, page-size — fluent and order-independent.
    for order in (
        client.commerce.orders
        .filter(status="open")
        .order_by("-created_at")
        .page_size(50)
    ):
        ...

    # Resource lookup uses [id], not (id). Indexing is request-free;
    # .retrieve() issues the GET.
    order = client.commerce.orders["ord_42"].retrieve()

    # Sub-collections walk the tree naturally.
    line = client.commerce.orders["ord_42"].lines.create(
        body={"sku": "SKU-1", "qty": 2},
    )

    # Actions return an action object whose method is `.run()`.
    client.commerce.orders["ord_42"].submit.run()

    # Singletons (e.g. /me, /health) are direct attributes.
    me = client.me.retrieve()

    # Count without a full walk.
    total = client.commerce.orders.filter(status="open").count()

Async is the same shape, Async-prefixed:

from acme.commerce import AsyncCommerceClient

async with AsyncCommerceClient(base_url="https://api.example.com") as client:
    async for order in client.commerce.orders:
        ...
    me = await client.me.retrieve()

Contributing

Bug reports, feature requests, and pull requests are welcome via GitHub issues and pull requests.

To work on okapipy locally:

git clone https://github.com/ffaraone/okapipy.git
cd okapipy
uv sync
uv run okapipy fetch-language en              # one-time spaCy model download
uv run pytest                            # full suite + coverage
uv run mypy src/okapipy/parser           # strict type-check (parser)
uv run ruff check src tests              # lint

Before opening a PR, please ensure tests, type-check, and lint all pass. The internal design notes live under design/ — read the relevant spec and plan before changing parser, generator, or customization behavior. Coding and documentation conventions are in CLAUDE.md.

License

okapipy is released under the Apache License 2.0. You're free to use it commercially, modify it, and redistribute it; please keep the copyright notice and the license text intact.