smara-crewai

CrewAI integration for the Smara Memory API -- persistent, semantic memory for AI agents.

Install

pip install smara-crewai

Quick start

1. Memory backend (automatic)

Plug SmaraMemory into CrewAI so all agent memory is automatically persisted and retrieved via Smara.

import os
from crewai import Agent, Crew, Task
from smara_crewai import SmaraMemory

os.environ["SMARA_API_KEY"] = "smara_..."

crew = Crew(
    agents=[
        Agent(
            role="Researcher",
            goal="Find and remember key facts",
            backstory="You are a diligent research assistant.",
        )
    ],
    tasks=[
        Task(
            description="Research the latest AI memory techniques.",
            expected_output="A summary of techniques.",
        )
    ],
    memory=True,
    memory_config={
        "provider": SmaraMemory(user_id="research-session"),
    },
)

crew.kickoff()

Every memory created during the run is stored in Smara. On the next run (even from a different machine), the same memories are available.

2. Agent tools (explicit)

Give agents SmaraStoreTool and SmaraSearchTool so they can decide when to remember or recall information.

from crewai import Agent
from smara_crewai import SmaraStoreTool, SmaraSearchTool

store = SmaraStoreTool(user_id="customer-support")
search = SmaraSearchTool(user_id="customer-support")

agent = Agent(
    role="Support Agent",
    goal="Help customers and remember their preferences",
    backstory="You are a helpful support agent with long-term memory.",
    tools=[store, search],
)

The agent will autonomously call smara_store_memory when it learns something important and smara_search_memory when it needs past context.

3. Direct client usage

from smara_crewai.client import SmaraClient

client = SmaraClient(api_key="smara_...")

# Store
client.store(user_id="u1", fact="User prefers dark mode", importance=0.7)

# Search
results = client.search(user_id="u1", query="UI preferences")
for mem in results:
    print(mem["fact"], mem.get("score"))

# User context
ctx = client.get_context(user_id="u1")
print(ctx)

Configuration

API reference

SmaraMemory (Storage backend)

• save(value, metadata=None) -- store a memory (called automatically by CrewAI)
• search(query, limit=5, score_threshold=0.0) -- retrieve relevant memories

SmaraStoreTool

Agent tool to store a memory. Accepts fact (str) and importance (float 0-1).

SmaraSearchTool

Agent tool to search memories. Accepts query (str) and limit (int 1-50).

Links

• Smara API docs
• Smara website
• CrewAI docs