Everything the memory layer does.
Lookout is not another notetaker. It is the layer underneath the ones you already use — reading across them, threading what was decided into one private brain, and handing that brain to your agent as a tool it can call. Here is every capability, in plain terms.
Six things, done properly.
Each one exists because a founder loses something real without it — a decision, a commitment, or the ability to prove either one was ever made.
Vendor-neutral ingestion
Lookout reads across Fireflies, Granola, and Otter at the same time — plus the calendar you already keep. Bring the notetakers your team already pays for and already trusts. There is no rip-and-replace, no migration weekend, and no new habit anyone has to learn for this to start working.
Cross-meeting memory graph
Every transcript is broken down into what actually matters — decisions, commitments, people, and dates — and woven into one private graph. The thread that runs from a question raised in January to the call where it was finally settled in March exists as a connection, not as four transcripts you have to reread.
Agent-native MCP recall
Your memory arrives as a tool your agent can call, over the Model Context Protocol. Recall from Claude Code, Claude Desktop, Cursor, Windsurf, or any MCP-capable client. The recall runs on your own Claude tokens inside your own session — we serve the memory, we never see the conversation.
Grounded answers
Every answer comes back traced to the exact source moment it came from: which meeting, which speaker, which date. You can follow any claim back to the sentence that produced it. Memory you cannot audit is just a confident guess, so Lookout never returns one.
Source ownership
The memory is yours, not a hostage of whichever notetaker you signed up for first. It stays portable, stays private, and stays isolated per user by Row-Level Security at the database. If you leave, you leave with your brain intact.
Read and write
Most memory tools stop at recall. Lookout lets you correct a decision, add a note, or fix an attribution straight from Claude Code, with a confirmation on every write. A brain you can only read drifts out of date. This one lets you keep it true.
Built so you'd trust it with the room.
Your tokens, not ours
Recall runs inside your own Claude Code, on your own tokens. We serve the memory; we never see the conversation. The only spend on our side is ingesting your meetings — never your questions.
Isolated by design
One browser consent (OAuth 2.1 + PKCE), a short-lived scoped token, no key on disk. Every row is walled by per-user Row-Level Security at the database — so one account reaching another's memory isn't restricted, it's structurally impossible.
Read and write
Most memory tools only let you read. Lookout lets you edit — correct a decision, add a note, fix an attribution — straight from Claude Code, with a confirmation on every write. This one lets you keep it true.
The tools you already run.
Connect once over OAuth. Lookout reads across all of them together, so the answer does not depend on which notetaker happened to be in the room that day.
Recalled anywhere MCP is supported — Claude Code, Claude Desktop, Cursor, Windsurf.
Put it in your Claude Code.
Lookout is in private beta. Tell us which notetakers you use and we'll bring you in as we open seats.