pydantic-studio 0.5.2

Interactive editor for Pydantic models — produces config.yaml/.toml/.json

pydantic-studio

Interactive editor for Pydantic models. Generate and edit config.yaml / config.toml / config.json against a strongly-typed schema, with three frontends sharing a single form-state model: sequential console prompts, a Textual TUI, and a React-backed local web app.

---

At a glance

Console prompt flow (default) — a Yeoman-style wizard for plain terminals, SSH sessions, and quick config generation. It asks one question at a time, shows the current/default value in brackets, keeps the value when you press Enter, and re-prompts invalid answers:

$ pydantic-studio edit myapp.config:AppSettings config.yaml
Editing AppSettings
name [prod]: staging
port [8080]: 9090
debug [false]: y
level (debug/info/warn) [info]: warn
saved to config.yaml

Textual TUI — a real form, not a modal editor: the focused field IS the editable field (type to edit, Tab/Enter to flow on), with inline descriptions and constraints, a required-field counter that Ctrl+N jumps through, and clickable Save/Cancel:

Containers are first-class: breadcrumb drill-in, a clickable [ + add item ] row, per-row ✕ delete, Ctrl+↑↓ reorder:

Browser UI (--frontend web) — the same session contract in a local web form: live YAML preview with your comments, nested models collapsed to one-line summaries, a required-missing counter in the header:

Validation errors anchor to their fields — the banner entries are click-to-jump links, offending rows ring red, and collapsed sections holding an error expand themselves:

Every shot above is generated from the shipped example (uv run python scripts/readme_screenshots.py, driving examples/02_server_config.py) — run it yourself to reproduce.

---

Why?

Hand-editing config files is error-prone. Pydantic schemas already encode the contract — types, constraints, defaults, descriptions. pydantic-studio turns that schema into an editor, with format round-trip that preserves your hand-written comments.

Status

v0.4.0 — Alpha. All 9 implementation phases plus the task-oriented TUI overhaul are merged on master. Production code paths are exercised by 1835 tests: 1753 default unit/integration/TUI/web smoke tests plus 82 explicit Playwright browser e2e tests. The editing session now has an explicit submit/cancel contract (run_app returns EditOutcome), and loading is symmetric with saving (existing values run through field/model validators — see docs/superpowers/specs/2026-06-11-task-oriented-editing-design.md). CI runs the default suite on Python 3.11, 3.12, 3.13, and 3.14; the browser, frontend bundle, wheel and sdist install smoke gates, and metadata checks for both wheel and sdist run on Python 3.13. Tag-triggered PyPI publishing uses GitHub OIDC Trusted Publishing with the pypi environment, not a repository API-token secret. The same release artifact is also published to piesource with independent failure handling and a final aggregate result; see docs/site/release.md for the required publisher setup. Version changes are recorded in CHANGELOG.md; vulnerability reporting guidance is in SECURITY.md; contributor setup and validation guidance is in CONTRIBUTING.md.

Install

pip install pydantic-studio
# or
uv add pydantic-studio

For EmailStr support, install with the email extra:

pip install 'pydantic-studio[email]'

Quick start

Programmatic

from pydantic import BaseModel, Field, HttpUrl, SecretStr
from pydantic_studio import build_form_tree, save_config


class AppSettings(BaseModel):
    name: str = Field(default="prod", description="Service identifier")
    port: int = Field(default=8080, ge=1, le=65535, description="Listening port")
    api_url: HttpUrl = Field(default=HttpUrl("https://api.example.com"))
    api_key: SecretStr = Field(default=SecretStr("change-me"))


tree = build_form_tree(AppSettings)
tree.set_value("port", 9090)
save_config(tree, "config.yaml")

# Service identifier
name: prod
# Listening port
port: 9090
api_url: https://api.example.com
api_key: change-me

CLI

# Stub a fresh config from defaults
pydantic-studio fill myapp.config:AppSettings --out config.yaml

# Validate without launching anything
pydantic-studio check myapp.config:AppSettings config.yaml

# Print the validated model
pydantic-studio run myapp.config:AppSettings config.yaml

# Ask one console prompt per field, then save
pydantic-studio edit myapp.config:AppSettings config.yaml

# Omit the file to start from defaults and save to AppSettings.yaml
pydantic-studio edit myapp.config:AppSettings

# Or open the Textual TUI
pydantic-studio edit --frontend tui myapp.config:AppSettings config.yaml

# Or the React-backed browser UI
pydantic-studio edit --frontend web myapp.config:AppSettings config.yaml

Format is auto-detected from extension (.yaml / .yml / .toml / .json).

Console Mode

edit defaults to the console renderer. It is intentionally not a full-screen UI: it behaves like a Yeoman-style prompt wizard, asking one question at a time in the current terminal.

Each prompt includes the field path and current/default value. Press Enter to keep that value, or type a replacement. Choice fields show the allowed labels inline (level (debug/info/warn) [info]:), booleans accept common yes/no forms, and invalid input is rejected without mutating the tree:

port [8080]: abc
cannot parse 'abc' as int
port [8080]: 9090

Containers stay sequential as well: lists and mappings first ask for a count, then prompt through each item or entry. When all prompts are complete, the tree is validated through the normal save path and written to the configured target; run_console_app(...) returns EditOutcome("submitted") on that successful completion.

Textual TUI

The TUI is a form: the focused field is the editable field — type to edit (focus selects the text, like Tab in a browser), no edit mode to enter or leave. Mouse is on by default (click rows, toggles, the Save/Cancel buttons; wheel scrolls).

Key	Action
type	Edit the focused field (persistent input, masked for secrets)
Tab / Shift+Tab / ↑ ↓	Commit the pending value, move to the next/previous field (invalid values block the move and show why)
Enter	Commit + advance; on containers: open; on large enums: chooser
Space / ← →	Toggle bools / cycle choices and union variants
Ctrl+S	Submit: flush + validate, write save_path if configured, exit submitted. On failure the cursor jumps to the first offending field
Ctrl+C	Cancel: clean tree exits cancelled; dirty tree asks (S save & exit / D discard / Esc keep editing)
Ctrl+N	Jump to the next missing-required field
Ctrl+F	Filter fields by name substring (group screens)
Esc	Layered: revert field → clear filter → pop child screen → cancel session
F2 / Del / [ + add item ] / ✕	Rename mapping key / delete / add items on container screens
Ctrl+Z / Ctrl+Y	Undo / redo

run_app(tree, save_path=None, readonly_paths=()) returns an EditOutcome; persist the tree only when outcome.submitted is true. A one-line HelpBar describes the focused field (type, constraints, required-ness, FieldInfo.description) and counts missing required fields. readonly_paths marks caller-owned fields: the row is labeled (read-only) and edits are rejected with a visible message. Indexed children may use either dotted or bracket syntax (tags.0 and tags[0] are equivalent).

Browser UI

--frontend web boots a local FastAPI app on a random free port and opens your browser (a React form sharing the TUI's session contract: run_html_app(...) -> EditOutcome, readonly_paths supported). Submit errors anchor to their fields (click-to-jump, red ring, auto-scroll); nested models collapse to one-line summaries (untouched optional models say not set); the header counts and cycles missing required fields; the YAML preview updates live. Closing the tab triggers a 30-second heartbeat timeout (configurable via run_html_app(..., heartbeat_timeout_seconds=...)) and the server shuts down with a cancelled outcome.

Embedding

The Web renderer is ASGI-first and can be mounted into a larger host app:

from fastapi import FastAPI
from pydantic_studio import build_form_tree, mount_html_app

app = FastAPI()
server = mount_html_app(app, "/studio", tree=build_form_tree(AppSettings))

Use StudioScreen(EditSession(...)) when embedding inside an existing Textual app; use StudioApp / run_app when pydantic-studio owns the terminal session.

Type coverage

Family	Types
Primitives	str, int, float, bool, Decimal
Choices	Enum, Literal[...]
Containers	list[T], set[T], tuple[T, ...], tuple[T1, T2, ...], dict[K, V]
Unions	T | U, Optional[T]
Temporal	datetime, date, time, timedelta
Network	IPv4Address, IPv6Address, IPv4Network, IPv6Network, AnyUrl/HttpUrl/FileUrl, EmailStr
Special	pathlib.Path, uuid.UUID, SecretStr, SecretBytes, re.Pattern, bytes
Constraints	Pydantic v2 Annotated constraints (ge/le/min_length/pattern/etc.) — auto-wired

Add custom types via register_builder(MyBuilder()).

File-format support

Format	Read	Write	User-comment round-trip
YAML	ruamel.yaml	ruamel.yaml	✓
TOML	tomllib (stdlib)	tomlkit	description comments only
JSON	stdlib json	model_dump_json(indent=2)	n/a (JSON has no comments)

from pydantic_studio import load_config, save_config

tree = load_config("config.toml", AppSettings)   # picks parser by extension
tree.set_value("port", 9090)
save_config(tree, "config.toml")                 # picks writer by extension

Format-specific helpers — load_yaml / save_yaml, load_toml / save_toml, load_json / save_json, save_draft_yaml (skips validation for mid-edit drafts) — are also exported.

Public API surface

from pydantic_studio import (
    # Tree construction
    build_form_tree, FormTree, FormNode,
    # Root model variants
    VariantSpec, VariantRegistry, build_variant_form_tree,
    # 24 node types
    StringNode, IntNode, FloatNode, BoolNode, DecimalNode,
    DatetimeNode, DateNode, TimeNode, TimedeltaNode,
    IpAddressNode, IpNetworkNode, UrlNode, EmailNode,
    PathNode, UuidNode, SecretNode, PatternNode, BytesNode,
    EnumNode, LiteralNode, SequenceNode, MappingNode, UnionNode, GroupNode,
    # I/O
    load_config, save_config,
    load_yaml, save_yaml, save_draft_yaml,
    load_toml, save_toml,
    load_json, save_json,
    # Drafts
    save_draft, load_draft, delete_draft, find_draft, draft_newer_than,
    # Renderers
    run_console_app,              # sequential console prompts -> EditOutcome
    EditSession, SubmitResult,    # shared renderer session lifecycle
    StudioApp, StudioScreen, run_app,  # Textual TUI
    EditOutcome,                  # session result: submitted | cancelled
    StudioServer, mount_html_app, run_html_app,  # HTML / React SPA / ASGI
    # Registry
    Registry, NodeBuilder, register_builder,
    default_registry, reset_default_registry,
    # Validation + exceptions
    ValidationResult,
    PydanticStudioError, NoBuilderError,
    CancelledByUser, ValidationFailedError,
)

Root model variants

When one editor entry point can produce several Pydantic model classes, keep that selection in pydantic-studio instead of coupling your app to a project-specific config factory:

from pydantic import BaseModel
from pydantic_studio import VariantRegistry, VariantSpec, build_variant_form_tree


class EmailNotifier(BaseModel):
    address: str = "[e-mail]"


class SlackNotifier(BaseModel):
    channel: str = "#ops"


variants = VariantRegistry(
    [
        VariantSpec(id="email", model=EmailNotifier, label="Email"),
        VariantSpec(id="slack", model=SlackNotifier, label="Slack"),
    ]
)

tree = build_variant_form_tree(
    variants,
    selected_id="email",
    discriminator="class_name",
    persistence="inline_discriminator",
)

The selection is renderer-native: console asks variant (email/slack) [email]:, TUI shows a root Variant row that cycles with ←/→, and the web UI renders a selector in the page. The default persistence mode is metadata, which keeps the choice in the session only. Use inline_discriminator to write a discriminator key such as class_name: slack alongside the selected model fields.

Drafts (auto-save + recovery)

from pydantic_studio import find_draft, load_draft, save_draft, delete_draft

# Mid-session
save_draft(tree, ".pydantic-studio.draft.json")

# On the next launch
existing = find_draft(".")
if existing is not None:
    tree = load_draft(existing, MyConfig)
    # ... user resumes editing ...
    delete_draft(existing)  # on successful submit

save_draft_yaml(tree, path) is the YAML equivalent — it skips to_instance() validation so partial trees can persist mid-edit.

Architecture

Frontends
  console prompts    Textual TUI      HTML browser      CLI helpers
  run_console_app    StudioApp        StudioServer      fill/run/check
        \                |                /                  |
         \               |               /                   |
          v              v              v                    v
FormTree (canonical state)
  - 24 node types, discriminated by `kind`
  - path-addressed mutations: set_value / add_item / rename_key / ...
  - snapshot ring for undo / redo
  - validate-first contract
        |
        v
I/O layer
  - load_config / save_config (extension dispatch)
  - YAML / TOML / JSON helpers
  - draft persistence

The full architecture doc is at docs/site/architecture.md. Run uv run mkdocs serve to read it locally.

Development

git clone https://github.com/invoker-bot/pydantic-studio
cd pydantic-studio
uv sync

# Tests
uv run pytest -q                          # 1753 default tests; skips tests/e2e/
uv run pytest tests/unit/test_yaml_io.py  # focused

# Lint
uv run ruff check
uv run pyright src/pydantic_studio       # production code only

# Docs
uv run mkdocs serve                       # 127.0.0.1:8000
uv run mkdocs build --strict              # also covered by test_docs_build.py

# Packaging
uv build
uv run twine check dist/*
uv run python -m venv .dist-smoke-wheel
.dist-smoke-wheel/bin/python -m pip install dist/*.whl
.dist-smoke-wheel/bin/pydantic-studio version
uv run python -m venv .dist-smoke-sdist
.dist-smoke-sdist/bin/python -m pip install dist/*.tar.gz
.dist-smoke-sdist/bin/pydantic-studio version

GitHub Actions mirrors this split: the default suite runs across Python 3.11-3.14, while browser and packaging gates run once on Python 3.13.

Project conventions are documented in CLAUDE.md — the guide for AI-assisted development sessions, but useful for any contributor.

Frontend development (Phase 2+)

The console and Textual frontends are pure Python. For example:

uv sync
uv run python examples/05_console_prompts.py
uv run python examples/02_server_config.py tui

The web renderer is a React SPA built with Vite, source under frontend/. End users do NOT need Node — pip install pydantic-studio ships the pre-built bundle. To modify the SPA:

cd frontend
corepack enable && corepack prepare pnpm@9 --activate   # one-time
pnpm install
pnpm dev              # Vite dev server with HMR; proxies /api/* to FastAPI on :8000

# in another terminal - dev-backend pins port 8000 to match Vite's proxy.
# (The packaged `examples/*.py web` flow binds an ephemeral port; great
# for end users, useless for a fixed-port dev proxy.)
uv run python frontend/scripts/dev-backend.py

# then open http://localhost:5173 (Vite's port; the proxy forwards
# /api/* to FastAPI on :8000).

# To refresh the committed bundle:
pnpm build            # or frontend/scripts/build.sh
git add ../src/pydantic_studio/renderers/html/static/dist

The bundled output (src/pydantic_studio/renderers/html/static/dist/) is committed to the repo so CI and downstream users don't need a Node toolchain.

Running the Playwright e2e suite (Phase 3+)

The unit-test default skips tests/e2e/ and disables the pytest-playwright plugin (its mere presence interferes with Textual's App.run_test() under asyncio_mode="auto"). To run the e2e suite explicitly:

uv run playwright install chromium                                  # one-time, ~150 MB
uv run python -m pytest tests/e2e -p playwright -o "addopts=-ra"    # 82 browser tests

License

MIT.