Skip to main content

In-Memory Backend

The in-memory backend stores everything in Python dicts — no persistence, no external dependencies. It's the fastest option and ideal for testing and prototyping.

When to Use​

ScenarioRecommendation
Unit tests✅ Best choice
Integration tests✅ Fast and isolated
Prototyping✅ No setup required
Production❌ Data lost on restart

Usage​

from memharness import MemoryHarness

# In-memory — no files, no database
harness = MemoryHarness("memory://")
await harness.connect()

await harness.add_conversational("thread1", "user", "Hello!")
messages = await harness.get_conversational("thread1")

await harness.disconnect()

In Tests​

import pytest
from memharness import MemoryHarness

@pytest.fixture
async def harness():
    h = MemoryHarness("memory://")
    await h.connect()
    yield h
    await h.disconnect()

async def test_my_agent(harness):
    await harness.add_knowledge("Python is great", source="test")
    results = await harness.search_knowledge("Python")
    assert len(results) > 0

The in-memory backend implements cosine similarity in pure Python. It's not optimized for large datasets but works correctly for testing.

Limitations​

  • All data is lost when the process ends
  • Not suitable for multi-process use
  • No persistence between test runs (which is usually what you want!)