pydantic-ai-firestore-persistence

Google Firestore backend for pydantic-ai-persistence — a PydanticAI StepStore (durable events / snapshots / tool-effect ledger) and message-history store, on Firestore.

Install

pip install pydantic-ai-firestore-persistence
# or:  pip install "pydantic-ai-persistence[firestore]"

Usage

from pydantic_ai import Agent
from pydantic_ai_harness.step_persistence import StepPersistence
from pydantic_ai_firestore_persistence import FirestoreStepStore

store = FirestoreStepStore(
    project_id="my-gcp-project",
    root_collection="pai_persistence",
    max_snapshots_per_run=20,
)
agent = Agent("openai:gpt-4o", capabilities=[StepPersistence(store=store)])

FirestoreHistoryStore(project_id=…) provides save/load message history. Authentication uses Application Default Credentials (gcloud auth application-default login).

Data model

Each partition key maps to a document <root>/<enc(PK)> whose items subcollection holds one doc per sort key: <enc(SK)> → {SK, data}. Doc ids are URL-encoded (Firestore ids can't contain /); the raw SK is a field so prefix/range queries work within a subcollection without extra indexes.

Note: PydanticAI's StepStore is a beta/experimental harness feature.

License

MIT