Memory Substrate · For AI Agents

Your memory, everywhere you think.

A shared memory layer across every AI tool you use.
Claude, Cursor, Codex, ChatGPT, Antigravity: they all know what you know.

Get started → Read the docs

Vertiso Memory is a Verified connector in the Anthropic Connectors Directory.

Free plan, no card LongMemEval R@5 98% MCP · REST · CLI

Runs where your agents already work

One memory, every agent

You're speaking to Claude. But your other agents are in the room.

What you settle with one agent is on hand for the rest, typed and waiting for your confirmation.

Claude Desktop 9:41 AM
You Heads up: the billing cutover moves to Oct 14. Stripe webhooks stay on v2 until then.
Claude Noted. I saved that as a decision with the v2 constraint attached, dated today.
Remembered · decision · constraint
Vertiso Same room
Cursor · new session 9:42 AM
You Start the webhook migration branch.
Agent Your memory says the cutover is Oct 14 and v2 stays live until then. I'll scaffold the v3 handler behind a flag and leave v2 untouched.
Recalled · decision · nothing re-explained

0

times you repeat yourself

8 types

decision, constraint, action, and more

You confirm

nothing closes itself

How It Works

Four verbs. One memory.

Write it once. Ask for it in plain language. Reach it from any client. Nothing closes itself.

  1. 01

    remember

    Write structured memories from any agent, session, or file. Facts, rules, projects, and the work in flight.

  2. 02

    recall

    Semantic + full-text hybrid retrieval. Ask in plain language; get ranked, relevant context back instantly.

  3. 03

    connect

    MCP server, REST API, CLI. Meets Claude, Cursor, Codex, and Antigravity wherever they live.

  4. 04

    confirm

    Actions inside your notes become live checkboxes. Report one done and it turns amber, not closed, and stays on your list until you click. Evidence proposes. You decide.

A note in Vertiso Memory with an action inside it reported done and awaiting confirmation.
An agent reported it done. It waits for you.

Structured Memory

Eight types. Zero noise.

Every memory has a type. Types shape how Vertiso Memory stores, retrieves, and surfaces context — so the right thing is on hand the moment you need it.

observation

What you've observed or recorded — about yourself, your work, or what you noticed.

constraint

Rules and guidance, graded by importance — from soft style preferences to non-negotiable, always-on requirements.

action

A concrete task or next step that needs to be done.

event

Something that occurs at — or between — specific points in time.

intent

A goal with a measurable outcome — achieved, or maintained in range.

project

A named body of work — actions roll up into it; it serves your goals.

decision

A choice you're weighing or have already made — with the rationale that drove it.

lesson

What you learned, why it matters, and what should change next time.

Session skills

Four skills. Your session survives the tool switch.

One install teaches your agent to save progress, hand work off, pick it up in the next tool, and close it out cleanly.

/checkpoint

Save work in progress without ending the session. Decisions and their reasons are kept verbatim.

Mid-session · before compaction

/handoff

Write a claimable handoff before you switch tools or stop for the day.

End of session · tool switch

/handoff-resume

Pick the handoff up in the next tool. It reconciles against current reality before you resume.

Session start · any client

/wrap-up

Close out finished work: complete actions, archive done intents, keep follow-ons tracked.

When work stabilizes

Ships with Claude Code Codex Cursor VS Code JetBrains The claude.ai connector delivers the memory tools today. The skills reach it when MCP prompts support ships.
All four skills, explained →

Fair question

“Doesn't Claude already have memory?”

Claude keeps a summary of you inside Claude. Vertiso Memory is yours to keep, and every tool you use can read it and write to it. Here is what that lets you do.

Cross-tool

Decided once. Known everywhere.

Claude, Claude Code, Cursor, Codex, ChatGPT, and any MCP client share one memory. Switch tools mid-task and nothing is re-explained.

Supersession

New truth replaces old.

Update a fact and recall returns the current version. The older one is marked superseded and kept, so the history is never lost.

Confirmation

Nothing closes itself.

An agent reports an action done. It turns amber and waits. It stays on your list until you confirm.

Graph

A graph, not a pile.

Memories link to the people, projects, and decisions they touch. Recall follows those edges outward, so context arrives connected.

Cited recall

Answers with receipts.

Ask in plain language. Get a synthesized answer with the memories it came from, a confidence, and what is missing.

You

What did we decide about the billing cutover?

Recall

The cutover moves to Oct 14. Stripe webhooks stay on v2 until then, and the v3 handler ships behind a flag.

decision · billing cutover constraint · webhooks stay on v2

Confidence: high · Missing: who owns the rollback.

Before you connect

Questions, answered.

Is there a free plan?
Yes. 100 memories, 20 writes and 25 synthesized recalls a month, no card. Upgrade when your memory outgrows it. See the plans
What happens when a fact changes?
Update the memory. Recall returns the current version and marks the older one superseded, so the history is kept and the stale fact stops leading. Read the guide
Which tools get the session skills?
Claude Code, Codex, and Cursor ship them in the plugin. VS Code and JetBrains install them from the extension. The claude.ai connector delivers the memory tools today; the skills reach it when MCP prompts support ships. See the skills
Can Vertiso read my memories?
Your memories are yours. The Trust Center says exactly what is stored, who can see it, and how to export or delete it. Read the Trust Center
What happens in my first conversation?
Your assistant calls hello, receives your standing guidance and open work, and you tell it what to remember. The install guide walks through it. Install guide
Can I export or delete everything?
Yes. Export your memories at any time, and delete your account with everything in it. Read the Trust Center
Does it work in the claude.ai web app?
Yes. Add it from the Connectors Directory inside claude.ai. Claude Desktop and Claude Code use the same connector. Install guide
What is a synthesized recall?
Ask in plain language and get an answer written from your memories, with the memories it came from, a confidence, and what is missing. Plain search stays unlimited. Read the guide
Is there a refund policy?
Yes. The refund and cancellation policy is public, and cancelling keeps your memories readable on the Free plan. Refund policy

Trust

Built to be trusted with what you know.

Each claim links to the public document that makes it in full.

LongMemEval R@5

98%

Every evidence session in the top five results, on the published run. Every question and verdict is public.

See every question →

  • Encrypted, moving and at rest.

    TLS 1.2 or higher in transit. AES-256 or equivalent at rest. Encrypted backups.

    Security baseline, in the Terms

  • Your memories are yours.

    Vertiso does not use your memories to train, fine-tune, or develop any model. Every tier, unconditionally.

    Privacy Policy, section 7.2

  • Export it. Delete it.

    Export every memory whenever you like. Delete your account, and everything in it goes with it.

    Your controls

  • A published uptime target.

    99.5% monthly uptime target for the API and MCP endpoints, measured per calendar month.

    Service Level Agreement

  • The rules, in writing.

    What may be stored and what may not, in a public policy. Nobody has to guess.

    Acceptable Use Policy
Read the Trust Center →

Pricing

Start free. Pay when you outgrow it.

Free is a plan, not a trial. Upgrade when your memory outgrows it.

Free

$0 forever

Everything you need to start. No card, no expiry, until you outgrow it.

Get started free

No card. No expiry.

  • 100 stored memories
  • Up to 4,000 characters per memory
  • 20 memory writes / month
  • 25 synthesized recalls / month
  • Automatic memory decomposition
  • Full knowledge-graph traversal

Basic

$24.99 /mo

Portable memory across every AI tool you use.

Upgrade

Cancel any time. Refund policy

  • 1,000 stored memories
  • Up to 4,000 characters per memory
  • 75 memory writes / month
  • 150 synthesized recalls / month
  • Multi-account switching (up to 1 linked account)
  • Automatic memory decomposition
  • Full knowledge-graph traversal
Most popular

Savant

$119 /mo

For wired-in agentic developers who live in their tools.

Upgrade

Cancel any time. Refund policy

  • 10,000 stored memories
  • Up to 32,000 characters per memory
  • 250 memory writes / month
  • 250 synthesized recalls / month
  • Multi-account switching (up to 3 linked accounts)
  • Automatic memory decomposition
  • Full knowledge-graph traversal
  • Premium synthesis models

Pro

$349 /mo

Maximum headroom: unlimited memories and unlimited linked accounts.

Upgrade

Cancel any time. Refund policy

  • Unlimited stored memories
  • Unlimited memory size
  • Unlimited memory writes
  • Unlimited synthesized recalls
  • Multi-account switching (unlimited linked accounts)
  • Automatic memory decomposition
  • Full knowledge-graph traversal
  • Premium synthesis models
Compare every plan, plus Teams and Enterprise →

From the blog

Articles

August 12, 2026

What it costs to remember

We measured three ways of giving an agent memory. Fifty questions, same model, same judge. Here is what we learned, and what we improved.

July 31, 2026

Ninety-three point four, honestly

What three months of LongMemEval taught us about building a memory product

May 31, 2026

Why I built Vertiso Memory

The founding post: the problem, the product, and who it’s for.

See all articles →

Ready when you are

Start building your memory layer.

Connect Claude, Codex, Antigravity, or any MCP-aware client. Your memory follows you everywhere.

Get started →