crewai-persistence-dynamodb

Amazon DynamoDB persistence backend for CrewAI Flows. Persists flow state across runs, with optional built-in message pruning via agentstate-reducer.

Implements CrewAI's FlowPersistence interface, so it drops straight into the @persist decorator.

Installation

pip install crewai-persistence-dynamodb
# with message pruning:
pip install "crewai-persistence-dynamodb[reducer]"

Requires Python 3.10–3.13 (matches CrewAI's supported range).

Usage

from crewai.flow.flow import Flow, start
from crewai.flow.persistence import persist
from crewai_persistence_dynamodb import DynamoDBFlowPersistence

backend = DynamoDBFlowPersistence(table_name="crewai-flows", region_name="us-east-1")

@persist(backend)
class MyFlow(Flow[MyState]):
    @start()
    def begin(self):
        self.state.counter += 1

The table is auto-created (on-demand billing) if it doesn't exist, keyed by flow_uuid.

With message pruning (conversational flows)

from agentstate_reducer import MessageReducer
from crewai_persistence_dynamodb import DynamoDBFlowPersistence

backend = DynamoDBFlowPersistence(
    table_name="crewai-flows",
    reducer=MessageReducer(min_messages=10, max_messages=20),
    messages_key="messages",   # state field holding the message list
)

API

DynamoDBFlowPersistence(table_name, *, region_name=None, boto_session=None, endpoint_url=None, ttl_seconds=None, reducer=None, messages_key="messages")

Authentication

Standard boto3 credential resolution — env vars, ~/.aws/credentials, AWS_PROFILE, SSO, or IAM roles. Needs permission to create the table (if absent) and read/write items.

Data model

One item per flow, keyed by flow_uuid. The flow state is stored as a JSON string in a data attribute; method_name and saved_at are stored alongside as metadata. Each save upserts, so load_state returns the latest state.

License

MIT

Long-term memory via on_prune (agentstate-reducer >= 0.4.0)

Messages pruned from the flow state are exactly the ones leaving the model's view. The persistence layer forwards a memory namespace to the reducer, and any on_prune hook receives (pruned_messages, namespace) — so pruned turns can flow straight into CrewAI's unified Memory (or any store), with no package coupling:

from agentstate_reducer import MessageReducer, ReducerConfig, Background
from crewai.memory import Memory

memory = Memory(storage=...)                       # e.g. crewai-memory-dynamodb

def remember(pruned, namespace):
    text = "\n".join(m["content"] for m in pruned)
    memory.remember_many(memory.extract_memories(text), scope=namespace)   # LLM calls -> run off-path

reducer = MessageReducer(config=ReducerConfig(max_messages=20, on_prune=[Background(remember)]))

class SupportState(BaseModel):
    id: str = ""
    memory_namespace: str = "/user/kamal"          # long-term scope: the USER, not the flow
    messages: list = []

@persist(DynamoDBFlowPersistence(..., reducer=reducer))
class SupportFlow(Flow[SupportState]):
    ...

The persistence layer reads memory_namespace (or whatever ReducerConfig.namespace_key names) from the flow state on every save_state and passes it through untouched. If the state never sets it, the namespace falls back to "/flow/<flow_uuid>". Each pruned message reaches the hooks once.