citadeldb-crewai

CrewAI memory backed by Citadel. Encrypted at rest, embedded in your process, and deletes that destroy the record's key, not just its row.

pip install citadeldb-crewai

Requires citadeldb>=2.2,<3 and crewai>=1.14.7,<2.

Provide a cancellation-aware embedder for the same model CrewAI uses. The Python semantic-embedder example shows the required protocol. In the examples below, my_embeddings is your application module exporting that configured embedder; it is not part of this package.

Route CrewAI's default memory backend through Citadel at startup:

from citadeldb_crewai import use_citadel
from my_embeddings import embedder

use_citadel("crew_memory.cdl", key="your-passphrase", embedder=embedder)

Enable memory on the crew:

from crewai import Crew

crew = Crew(agents=[...], tasks=[...], memory=True)

use_citadel handles CrewAI's default storage, storage="lancedb", and storage="citadel". Other backend names and LanceDB paths are left unchanged.

To route one crew instead of the whole process, hand the backend over directly and skip the startup call:

from crewai import Crew
from crewai.memory.unified_memory import Memory
from citadeldb_crewai import CitadelBackend

backend = CitadelBackend("crew_memory.cdl", key="your-passphrase", embedder=embedder)
crew = Crew(agents=[...], tasks=[...], memory=Memory(storage=backend))

Direct use

from citadeldb_crewai import CitadelBackend
from crewai.memory.storage.backend import MemoryRecord

backend = CitadelBackend("crew_memory.cdl", key="your-passphrase", embedder=embedder)

backend.save(
    [
        MemoryRecord(
            content="the deploy failed because the disk was full",
            scope="/team/ops",
            categories=["incident"],
            metadata={"env": "prod"},
            importance=0.9,
        )
    ]
)

embed_query = getattr(embedder, "embed_queries_with_cancel", embedder.embed_with_cancel)
query_embedding = embed_query(["Why did the release break?"], None)[0]
hits = backend.search(query_embedding, scope_prefix="/team", limit=5)
for record, score in hits:
    print(f"{score:.3f}  {record.content}")

Deletes destroy the key

Every record is sealed under its own key. Deleting destroys that key, so the bytes on disk stay unreadable. A backup taken before the delete carries its own copy of the wrapped key and is out of scope.

from datetime import datetime, timedelta, timezone

cutoff = datetime.now(timezone.utc) - timedelta(days=30)

backend.delete(record_ids=["abc123"])  # one record
backend.delete(scope_prefix="/team/ops", categories=["incident"])
backend.delete(scope_prefix="/team", older_than=cutoff)
backend.reset("/users/alice")  # a whole subtree

reset on a per-user scope destroys the key of every record in that subtree.

Importance

MemoryRecord.importance is retained in the record. Backend search returns vector similarity; CrewAI applies its composite scoring, including importance, after the storage call.

Notes

CrewAI embeds queries itself and hands the backend a vector, so search runs vector recall plus the scope, category, and metadata predicates the protocol defines.

Citadel is embedded and one process owns the file. A path already open on this thread, under the same passphrase, is shared, so this can sit on the same database as another Citadel adapter; construct them on the same thread.

Supplied vectors are stored as-is. Records without a vector are embedded by the configured model. An update without a supplied vector preserves the stored vector when the text is unchanged.

The embedder must expose dim, metric, and model_id, plus embed_with_cancel(list[str], cancel_token) -> list[list[float]]; asymmetric models may also provide embed_queries_with_cancel. Accept None as the token; otherwise poll cancel_token.check() between bounded batches. Pass the same model to CrewAI and Citadel so supplied and generated vectors share one space. A cosine metric is required. Regions persist the model identity and dimension; switching models requires re-embedding or a new region.

License

Apache-2.0