crewai-persistence-mongodb

MongoDB persistence backend for CrewAI Flows. Persists flow state across runs, with optional built-in message pruning via agentstate-reducer. Implements CrewAI's FlowPersistence interface.

Installation

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

Requires Python 3.10–3.13 and a reachable MongoDB (local, Atlas, self-hosted).

Usage

from crewai.flow.flow import Flow, start
from crewai.flow.persistence import persist
from crewai_persistence_mongodb import MongoDBFlowPersistence

backend = MongoDBFlowPersistence(connection_string="mongodb://127.0.0.1:27017/")

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

With message pruning:

from agentstate_reducer import MessageReducer
backend = MongoDBFlowPersistence(
    connection_string="mongodb://127.0.0.1:27017/",
    reducer=MessageReducer(min_messages=10, max_messages=20),
    messages_key="messages",
)

API

MongoDBFlowPersistence(*, connection_string="mongodb://127.0.0.1:27017/", database_name="crewai_flows", collection_name="flow_states", client=None, reducer=None, messages_key="messages")

Data model

One document per flow, keyed by flow_uuid (unique index); the flow state is stored in a data field, with method_name and saved_at alongside. Each save upserts, so load_state returns the latest.

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(MongoDBFlowPersistence(..., 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.