Loom
Request access →

Your team's memory. In every AI you use.

Ask in Cursor, Claude, Copilot or Codex. Loom answers from what your company actually decided, cites the source, and shows only what you're allowed to see.

Request access →How it works
acme.itsloom.app · MCP
You · in Cursor
Loom
The problem

Every AI tool your team uses starts from zero.

Scroll. Watch what changes when they share one memory.

Cursor
What did we decide about the vendor?
No memory of your company. Plausible guess incoming.
Claude
What did we decide about the vendor?
Postgres. Decided in May after the MongoDB pilot.
Answered here, once. The other three never hear about it.
Copilot
What did we decide about the vendor?
No memory of your company. Plausible guess incoming.
Codex
What did we decide about the vendor?
No memory of your company. Plausible guess incoming.
Loom
The thing nothing else can do

Steer the organization without revealing what you know.

A board decision stays with the three people cleared to see it. Its effect still reaches everyone. Same question, two people asking.

Draft the Q3 roadmap for the platform team.
One source shaped this draft that you are not cleared to see.withheld
Q3 roadmap · draft
  1. 01Audit-log retention for regulated customers
  2. 02Region-pinned deployments, EU first
  3. 03Patient-data masking in the ingest path
  4. 04Marketplace connectors, deferred
The draft leans healthcare. The engineer never sees the decision. Loom knew.
A ledger, not a chat log

It catches contradictions before they ship.

Loom holds what your organization has actually decided. When a person or an AI drafts against a live decision, it says so before the work leaves the building.

Q1 strategy · marketplace vertical · draft
Expand into fintech in Q1 with two lighthouse accounts.
Before you finalize
This contradicts a ratified decision that restricted marketplace investment to healthcare through H1. Are you superseding it, extending it, or does the draft change?
SupersedeExtendChange the draft
Confidentiality survives

Everyone gets answers. Nobody gets answers they shouldn't.

Five sensitivity tiers, enforced where the data lives, not in the app. Pick who is asking.

public
Published docs, marketing
answers
internal
All-hands notes, general strategy
answers
restricted
Customer contracts, HR files
withheld
embargoed
Board decisions, M&A, personnel
withheld
personal
Your own drafts and notes
withheld
Auditable by construction

Every answer leaves a paper trail.

Not a report you run later. The receipt is written on the way to the answer: what was retrieved, which tiers applied, which policies fired, anchored to a signed daily root. When someone asks why your AI said that, you have the answer.

Your AI audit trail isn't a project. It's a side effect.

answer
The renewal terms with Acme are unresolved between sales and legal. Legal's last position was a twelve-month cap.
retrievedTwo passages: legal thread, sales call notes
tiers appliedinternal, restricted · one embargoed source withheld
policies firedPersonal data masked · contract terms need legal on the thread
signed roottoday · verifiable offline
No new tab. Ever.

It works inside the AI tools your team already pays for.

Loom speaks the open standard Cursor, Claude, Copilot and Codex already speak. Connect once. Swap your model vendor next quarter; your memory stays.

Cursor
Claude
Copilot
Codex
Loom
"loom": { "url": "https://yourcompany.itsloom.app/mcp" }
What Loom is not
Not a chatbot
Every chat starts blank. Loom is the layer underneath all of them.
Not enterprise search
Search points at documents. Loom returns cited answers and the decisions behind them.
Not a wiki
Nobody writes anything down. Loom reads what your team already produces.
Not another login
Your team never opens it. It shows up inside the tools they already use.

If only the company knew what the company knows.

Managed cloud, or deployed on your own database in your own region. Same layer either way.

Request access →
© Loom
ManifestoWhitepaperContact