Atlaso

Atlaso is a memory layer for AI tools — connect once and it automatically recalls the context that matters across Claude Code, Cursor, Codex, and every AI tool you use.

Documentation

Proof

211 questions from LongMemEval-S. Same model, with and without Atlaso.

69%

of benchmark questions answered correctly

Without memory: 5.7%. With Atlaso: 69.2%.

96%

of the memory the answers needed, retrieved

Evidence recall. The four filled circles are the four in a hundred it missed.

+13

points over keyword search

Keyword search finds your words: 55.9%. Atlaso finds what you meant: 69.2%.

Exact figures: 5.7% without memory, 55.9% with keyword search, 69.2% with Atlaso, 96% evidence recall. See how we tested

Capabilities

*Automatic in tools with hooks. In MCP-only tools, the model calls Atlaso itself, and a skill can make that routine.

*Automatic in tools with hooks. In MCP-only tools, the model calls Atlaso itself, and a skill can make that routine.

*Pro. Today in Claude Code and Codex.

What makes your memory work.

Fewer repeated memories.

Before it saves, Atlaso checks what it already has and skips a near copy. Your dashboard shows your exact duplicate rate, so you can see it stay low.

Sorted by what it means.

Memories are filed as helpful, a warning, a caution, or still open. You can see the filing on your dashboard and change it.

Numbers backed by activity.

Every figure on your dashboard comes from your own sessions. Where the evidence is thin, it shows a range or says not measured, and tells you what unlocks it.

Yours to inspect and correct.

Every memory is on your dashboard in plain text. Read it, refile it, fix it, or remove it. Housekeeping Atlaso does itself is never counted as your correction.

One score for your memory.

Memory Power runs 0 to 100: how much you have saved, how well it agrees with itself, how recently it has been used. One number to watch instead of a dozen.

Capture, counted.

Your dashboard shows how many of the moments it looked at became a memory, and which tools contributed. If the counters ever look wrong, it hides the number rather than clean it up.

Forget means forgotten.

Forget a memory and it stops coming back, in every tool. Other devices pick up the change the next time they sync.

What people say

From our Product Hunt launch.

Product Hunt

The re-explaining tax is real, I switch between Claude Code and other tools daily and re-establishing context every time is the part nobody talks about when they compare AI coding tools.

Product Hunt

I really like the distinction between thin, settled, contested, and superseded memories. I haven't seen many memory tools make uncertainty this explicit.

Product Hunt

The way you're answering people here is genuinely refreshing. Publishing the benchmark you lose on is not a small thing.

Product Hunt

The hands-off sync between Cursor and Claude Code sounds ideal because that was literally one of my biggest pain points. … Gonna try out the MCP workflow today.

Product Hunt

Love that you're going after the "portable memory" problem instead of building yet another chat wrapper. The idea that my context follows me from ChatGPT to Claude to whatever comes next is exactly the pain I feel daily as a founder juggling tools.

Safety, built in

Encrypted in transit and at rest

TLS on every connection, encrypted at rest by our infrastructure providers. Keys, tokens and passwords are redacted before a memory is stored, wherever we can recognise them. Each user’s memory lives in its own database.

Never trained on, never sold

We do not train AI models on your memory, and we never sell it or use it for ads. When a model is needed to work on a memory, it is sent only to produce that result, to a provider whose policy is not to store prompts or outputs. Our internal analytics show counts, not memory content.

Yours to control

Forget a memory and it stops coming back in every tool; other devices pick up the change when they sync. Delete your account and every memory goes with it.

Questions, answered

Which AI tools can I connect?

Atlaso supports Claude Code, Codex, Cursor, Antigravity, OpenCode and Claude Desktop. Other tools can connect if they support remote MCP with sign-in. The connection picker in the dashboard shows what is available and walks you through the right setup for each.

How do I set up Atlaso?

Do I have to tell Atlaso what to remember?

Where is my memory stored, and is it used to train AI?

How much does Atlaso cost?

Less explaining.More

Bring your context to the AI tools you use.