getmnemo

Python SDK for Mnemo Memory — long-term memory infrastructure for AI agents.

pip install getmnemo

Quickstart

from getmnemo import Mnemo

memory = Mnemo(api_key="prfly_live_...", default_container_tag="user:jane")

# Store an atomic fact
memory.add("User prefers Japanese short-grain rice for onigiri.")

# Retrieve relevant facts — returns a list of ranked hits
results = memory.search("what kind of rice does the user like?")
for hit in results:
    print(f"{hit.score:.2f}  {hit.content}")

Every call is scoped to a container — a "<type>:<id>" string such as "user:jane" or "project:alpha". Set it once on the client (default_container_tag=) or per call (container_tag=). Searching a container that has never been written to returns an empty list.

Async variant:

import asyncio
from getmnemo import AsyncMnemo

async def main() -> None:
    async with AsyncMnemo(api_key="...", default_container_tag="user:jane") as m:
        await m.add("Trip to Costa Rica was 5 days, brought 7 shirts.")
        results = await m.search("how many shirts did I pack?")
        if results:
            print(results[0].content)

asyncio.run(main())

Configuration

The client reads from env vars when arguments are not passed explicitly:

API surface

All methods exist on both Mnemo (sync) and AsyncMnemo (async).

add_many items are content strings or mappings:

memory.add_many([
    "Standup moved to 9:30.",
    {"content": "Invoices are net-30.", "memory_type": "fact", "idempotency_key": "inv-policy-1"},
])

Upgrading from 0.1.x

0.2.0 tracks the current Mnemo API; 0.1.x requests are rejected by it.

• workspace_id is no longer required (the API key carries identity); pass a container instead: Mnemo(api_key=..., default_container_tag="user:jane").
• search("q") now returns a list[SearchHit] (was SearchResponse.hits).
• add(...) now returns an AddResponse (was a single Memory).
• actor_id was removed — use per-user container tags.

Development

pip install -e ".[dev]"
pytest
ruff check .
mypy src

License

MIT