Log inGet Started

A closer look at how Atlaso takes shape.

The ideas, experiments, and stories behind Atlaso.

Four memory signals — recent, recurring, unsettled and next — converge on Atlaso before the first word.

Growth5 min read

Giving AI a subconscious.

Your AI wakes up blank every session. Ambient Memory hands it an orientation block built from your own memories: what changed, what you keep returning to, what's still unsettled, and where you're headed. All of it before you type a word.

Paired Atlaso and mem0 results using the same reader and questions across the study's four scoring measures.

Research10 min read

What memory systems actually do.

A reader-matched head-to-head on LongMemEval-S (n=500): the same Qwen 3.5-9B reader answers for every system. Atlaso beats mem0 by +6.6 points (p≈0.007) across three independent judges, with zero LLM calls at ingestion. Letta ties; the losses are published too.

One hundred dots show the lift decomposition: approximately 94 percent retrieval and 6 percent format.

Research12 min read

Where the 12-point lift comes from.

On 10 seeds × 300 tasks, the +11.87pp lift on a 9B model decomposes cleanly: ~94% comes from scope-matched retrieval, ~6% from the trained deposit format. All 10 seeds positive, sign-test p = 0.00195.

Independent signals pass through an adaptive gate before agents connect.

Research8 min read

When to start communicating.

An adaptive gate that turns on coordination only under population-level distress outperforms both always-on and always-off baselines by +34% on the hard task across 10 seeds, and the same principle transfers to LLM agent collectives.