A memory store that flags its own conflicts before your agent does.

What atlaso is

Atlaso is a Python SDK that stores plain-English deposits and tells you, at recall time, whether they agree, disagree, or conflict. Most memory libraries return a list of hits and trust you to figure out which ones to believe. Atlaso returns a list of hits plus an is_confident flag, a has_disagreement flag, an agreement_score, and a list of conflict_peers. Your agent gets data back; it also gets a verdict on whether the data is settled.

The store is local SQLite + FTS5, single-file, zero-telemetry, Apache-2.0. The only runtime dependency is httpx.

Thirty-second tour

Install:

pip install atlaso

Write a deposit. Recall it. Inspect the verdict.

from atlaso import Memory

m = Memory()
user = m.for_user("alice")

user.add("Alice prefers oat milk in lattes")

results = user.recall("milk")
print(results.explain())            # bag-level verdict
for r in results:
    if r.has_disagreement:
        print("WARN", r.content, "conflicts with", r.conflict_peers)
    elif r.is_confident:
        print("OK  ", r.content)
    else:
        print("?   ", r.content)

Why it exists

Agentic systems accumulate facts. Most of those facts agree with each other. Some don't. The ones that don't are the most expensive to ignore — a stale preference, a deprecated API, a contradicting observation from a redteam run. Conventional vector stores will happily return both versions ranked by similarity and let your model pick the wrong one.

Atlaso's gate rejects under-evidenced writes at the door. Its aware retrieval pipeline groups hits by scope and surfaces directional-polarity conflicts before they reach your prompt. The result: your agent can branch on is_confident instead of guessing.

Where to go next