Small agents with big memory

Tiny footprint, infinite memory

The Problem

Smolagents from Hugging Face are beautifully minimal. Small, fast, and surprisingly capable. They prove you don't need a massive framework to build effective agents.

But small doesn't mean you want amnesia.

Your smolagent handles tasks efficiently, then forgets everything. The next session, it's back to "Hello, I'm an AI assistant." All that efficiency, none of the memory.

Minimal framework doesn't have to mean minimal context.

How Stompy Helps

Stompy gives your smolagents persistent memory without adding bloat.

Your lightweight agents gain: - Cross-session context without heavy dependencies - Knowledge that persists across runs - Semantic search for finding relevant past interactions - All without sacrificing the "smol" philosophy

Stay light. Remember everything.

Integration Walkthrough

Add Stompy to your smolagent

Smolagents support custom tools. Add Stompy memory tools directly.

from smolagents import CodeAgent, ToolCallingAgent
from smolagents.mcp import MCPClient
import os

# Connect to Stompy via SSE
stompy = MCPClient(
    url="https://mcp.stompy.ai/sse",
    headers={"Authorization": f"Bearer {os.environ['STOMPY_TOKEN']}"}
)

agent = ToolCallingAgent(
    tools=[*stompy.get_tools()],
    model="Qwen/Qwen2.5-72B-Instruct"
)

# Agent now has persistent memory