Your first memory
Prerequisites
Use Python 3.10–3.13 in a virtual environment. This exercise uses synthetic data, does not contact a model, and does not enable agent injection.
python -m venv .venv
source .venv/bin/activate
python -m pip install atmem==2.3.6On Windows, activate with .venv\Scripts\activate instead.
Store and retrieve
Run this Python example. Its temporary database disappears when the block finishes.
from tempfile import TemporaryDirectory
from pathlib import Path
from atmem import Memory
with TemporaryDirectory() as directory:
memory = Memory(Path(directory) / "demo.db", auto_vectors=False)
try:
memory.remember("demo-user", "My preferred editor is Vim.", session_id="demo")
records = memory.recall("demo-user", "preferred editor", limit=5)
assert any("Vim" in record["content"] for record in records)
assert memory.recall("other-user", "preferred editor") == []
print("Memory found; another subject cannot retrieve it.")
finally:
memory.close()Expected output: Memory found; another subject cannot retrieve it.
This demonstrates embedded memory retrieval, not proof that a model received context.
auto_vectors=False isolates the basic example; normal persistent use can keep
the default vector sidecar enabled.
Open the dashboard
For a persistent local installation, outside this disposable example:
atmem init
atmem dashboardKeep the Administrator temporary password printed by initialization, sign in and change it. The dashboard is local, not the public documentation website. See accounts and evidence.
Connect your agent
Choose an integration. Start in shadow mode, verify identity, capture and delivery readiness, then explicitly activate. Optional intelligence setup is explained in AtBot.
If it fails
An import error usually means the active Python environment differs from the one
where you installed AtMem. Check python -m pip show atmem.
An empty result is not permission to broaden scope: inspect subject identity,
record lifecycle and retrieval configuration.