ooai-persistence 0.2.2

Typed persistence utilities for LangGraph-based OOAI applications.

ooai-persistence

ooai-persistence provides typed persistence helpers for LangGraph-based OOAI applications.

Responsibilities

• checkpointer configuration and backend resolution
• store configuration and backend resolution
• graph cache configuration
• strict serializer allowlist support
• sync and async persistence contexts
• local infrastructure defaults for Postgres, optional Redis, and optional MongoDB

Quick start

pdm install -G :all
pdm run pytest
pdm run ooai-persistence smoke --backend memory

For a no-infrastructure memory bundle:

from ooai_persistence import AppSettings, open_sync_persistence

settings = AppSettings.memory()

with open_sync_persistence(settings) as persistence:
    persistence.store.put(("users", "will"), "profile", {"name": "Will"})
    profile = persistence.store.get(("users", "will"), "profile")

For async usage:

from ooai_persistence import AppSettings, open_persistence

settings = AppSettings.local_sqlite(".ooai/persistence/dev.sqlite3")

async with open_persistence(settings) as persistence:
    ...

CLI

The package ships a small diagnostics and smoke-test CLI:

ooai-persistence doctor --backend postgres --json
ooai-persistence smoke --backend memory
ooai-persistence smoke --backend sqlite --sqlite-path .ooai/persistence/smoke.sqlite3
ooai-persistence smoke --backend postgres --async
ooai-persistence env --output .env
ooai-persistence ensure-postgres

The async Postgres smoke command exercises both the LangGraph checkpointer and store through the public context API.

Local Postgres

make bootstrap
make infra-up
make infra-test-postgres
make infra-down

make bootstrap creates .env from .env.example when needed and installs all PDM extras. See infra/README.md for Compose details.

make infra-test-postgres runs the real async Postgres E2E path and the CLI smoke path against Docker Compose Postgres.

If Docker is not installed but .env points at a reachable Postgres server, make infra-up falls back to ooai-persistence ensure-postgres and creates the configured database if needed.

Default backend behavior

By default, checkpointer and store use backend="auto".

auto resolves in this order:

1. async Postgres when configured
2. async SQLite when configured
3. async MongoDB when configured
4. async Redis when configured
5. in-memory fallback

Serializer allowlist registry

The package includes a reusable strict-msgpack registry:

from ooai_persistence.serde.registry import MsgpackAllowlistRegistry

registry = MsgpackAllowlistRegistry()
registry.register_symbol("my_app.models", "WorkflowState")
registry.register_type(MyPersistedModel)
registry.register_import_string("my_app.models:AnotherPersistedModel")

That registry can be passed into the persistence context:

from ooai_persistence import AppSettings, open_persistence

settings = AppSettings()

async with open_persistence(settings, registry=registry) as bundle:
    ...

Documentation and release checks

pdm run sphinx-build -W -b html docs docs/_build/html
pdm build

CI runs formatting, linting, typing, tests with coverage, and the Sphinx build. Docs also publish from main to GitHub Pages.

Releasing

Releases are tag-driven:

pdm lock --check
make check
git tag v0.2.0
git push origin v0.2.0

The release workflow verifies that the tag matches pyproject.toml, runs the async Postgres E2E checks, builds the wheel/sdist, publishes through PyPI Trusted Publishing, and creates a GitHub Release with artifacts attached.