EARLY ACCESS · 2026

Your AI is brilliant.It just doesn’t know your work.

ran connects the tools you already use and keeps a live memory of everything in them — so the AI you ask already knows, and can already act.

Free while we’re in early access.

app.ran-stack.com
  • Every answer cited
  • 92.6% retrieval accuracy
  • 100% honest abstention
  • nothing sent without your yes

Margarita runs on your work. And answers for it.

One live memory across your tools — asked, checked, and acted on.

1/4

Ask like it sat next to you all year

One question searches everything you’ve connected. Every claim carries a citation that opens the exact passage, enlarged.

What did I quote Meridian for the spring order?

You quoted $15/unit on Mar 121 and confirmed it by email two days later2.

1Notion · Pricing 2026
Meridian — spring order: $15/unit, net-30. Confirmed with Dana on Mar 12.

2/4

When your tools disagree, it says so

Two documents, two numbers — Margarita shows both instead of picking one, and proposes the fix.

3/4

She drafts. You release.

Margarita writes the reply and stops at the edge. Nothing leaves until you press Release — one button, no countdown.

4/4

Always up to date

Your sources sync on their own — nothing to re-upload, nothing to maintain.

Get early access

  • NotionSynced 2 min ago
  • GmailSynced just now
  • SlackSyncing…
  • DriveSynced 14 min ago

Live memory · kept current on its own

Minutes to set up. First answers today.

Connect your accounts. Margarita takes it from there.

  1. 01

    Connect

    Link Notion, Gmail, Slack and Drive. OAuth, not a migration.

    No manual setup needed.

  2. 02

    Margarita reads

    She builds a live memory of what’s inside — and keeps it current on her own.

    Nothing to re-upload, ever.

  3. 03

    Ask & approve

    Every answer cites its source. When something would send, you release it.

    One Release button. No countdown.

From question to sent — with your yes in the middle.

The whole loop: cited answer, drafted reply, one Release, one receipt.

Product demo: you ask a question. Margarita answers with citations to your own documents, drafts a reply, and holds it — nothing sends until you press Release. Afterwards the activity log reads “Sent — approved by you”, with an Undo.

Built to carry your whole memory.

Everything you’ve connected, one place that already knows it.

Knowledge

One memory across your tools

Answers

Citations at the claim level

Control

Full auto, or approve before it sends

Undo

Everything reversible names its undo

MCP

Your Claude, on your memory

One memory. Every tool you already use.

Notion, Slack, Gmail, Drive, Outlook — and the Claude or ChatGPT you already pay for, over MCP.

ran.
  • Notion
  • Slack
  • Gmail
  • Drive
  • Outlook
  • Claude
  • ChatGPT

Read the MCP connector docs →

Measured, not promised.

An open benchmark on a real corpus — always the overall numbers.

92.6%
right sources retrieved
90.7%
answers judged correct
100%
honest abstention

Open benchmark · 31 questions · LLM-judged

Simple while it’s early.

No tiers to study — you talk to the founder either way.

Early access

Free/during early access

Your own memory and operator, built from your own accounts.

NotionGmailSlackDrive

Get early access

What’s included

  • Your tools, connected in an afternoon
  • A live memory that keeps itself current
  • Margarita, waiting for your yes before anything sends
  • Every answer cited, every change undoable
  • The MCP server for your existing Claude

Custom

Talk with us

Bring your own Anthropic or OpenAI key — your privacy terms, your bill.

Talk to us

What’s included

  • Everything in early access
  • Inference on your own model account
  • Your data agreements, your retention

See it on your own work.

Bring your tools and watch Margarita learn them.

Free while we’re in early access.