Work on what matters most. Catapult handles the rest.

Built uniquely for you, Catapult understands your work as if it's been with you since day one. Every email, message, meeting and document.

So it knows your tone, your priorities, the context behind your decisions. It anticipates what you need before you think to ask, and helps you get it done.

It drafts your replies in your voice. It preps you for the meeting, then follows through on what you promised. It catches what you'd have missed. It thinks through the hard problems with everything you've ever worked on in front of it. And the more it learns, the more it does that you never thought to ask for.

No explaining yourself. No uploading context. No markdown files, no workflows to build. It just works.

We're a small, obsessed team of repeat founders, researchers, and engineers — from Meta, Amazon Research, Oxford, Cambridge, and Imperial building Catapult for ourselves, and for people like us.

We can't wait for you to try it.

Get early access

A few of the things Catapult does

Drafting emails in your voice, with the right context, ready to send. Especially the thorny ones.

Reply to David from Jupiter about Catapult's US bank account

David needs the missing information to open your US bank account. I’ve found everything and drafted your reply.

Email draft
· Replying in thread
To David@jupiter.com

Managing your schedule around your priorities: creating events, suggesting times, fixing conflicts.

Schedule 3 year service review with Carla
  • Carla followed up asking to schedule Catapult's Year 3 Service Review for your office space.
  • I've created the invite for Friday June 5, 10:00-11:00am BST, which is a clear free slot. If that works, I'll also draft the email reply back to her for you.
Fri 5 Jun BST
6 am 7 am 8 am 9 am 10 am 11 am 12 pm 1 pm 2 pm 3 pm 4 pm 5 pm 6 pm 7 pm 8 pm
Investor intro
Maya 1:1
Product roadmap
Rav x Carla - Jora service review
Priya 1:1
Candidate debrief
Strategy block
Sam 1:1
Customer escalation
Board prep
Greg 1:1
GTM pipeline
Finance review
Calendar event
Sources

Preparing you for the meetings that matter, with everything you need to know, ask, and do.

Cian Murphy research engineer interview - Meeting prep ready

Here's the context

  • You have Cian / Rav - 30min Meeting on Fri June 24, 12:30 to 13:00 EST.
  • He's already shown initiative, asking for an earlier slot, and sending you his Github repo, and ideas for Catapult.

What to focus on in the call

  • Understand CV gaps: I analysed his CV, and there's an 11-month gap between his time at Meta, and joining OpenAI. His LinkedIn profile implies one job started after another. Dig deeper on this.
  • Systems depth: he said the project he's most proud of was his memory system. Push him on this, given Catapult's built a long-term, dynamic memory system that updates in real time. How would he build this from scratch?
  • Career focus: he's worked in pure research roles as well as pure engineering roles. Both at startups he's built, as well as for the likes of Meta and OpenAI. Why Catapult, and why now? Look out to see where his eyes light up around the work he enjoys.

Beyond these, ask your usual questions. Let me know if there's any other research I can do ahead of the call. Whether it's finding his latest tweets or blog posts.

Sources

Handling every promise and action item, the moment the meeting ends.

Send Greg the product feedback from the call with Sam

Here's the context

  • From your Granola meeting note, Sam said he's really enjoying Catapult. He's surprised by how well it knows him, has cancelled his Claude subscription, and is using Superhuman less for email.
  • He shared a few product ideas: a Daily Briefing of work Catapult did overnight, importing his Claude memories, and keyboard shortcuts.
  • You said you'd share the feedback with Greg. I've drafted the message for you.
Slack message → Gregory Wilsenach
hey Greg! Sam's loving Catapult, he's even cancelling his claude sub. He had 3 requests for the roadmap: a daily briefing, a way to import his memories from claude, and keyboard shortcuts. I know we're already working on the daily briefing, defo worth you chatting to Sam further on this.
Sources

Thinking through the hard problems, with everything you've ever worked on in context.

Commitments you've made across your work
What have I dropped the ball on?

Highest priority

  • Eric Jones: lock the Friday onboarding time. He replied asking to "shoot for Friday," but I don't see a confirmed slot or reply yet.
  • Mercury: complete onboarding to get your US bank account set up.
  • Landing page: update the text based on the team's feedback.
  • 3 invoices due today: pay Goodrise, Anthropic and Google Cloud.

Hiring

  • Inderjit Singh: send the interview outcome. Seems like a reject based on your Granola notes.
  • Mark Cristoph: move him to the loop with the rest of the team - you're all free next Tuesday.
  • Founder's Associate: review the 29 pending applicants.

Go to market

  • Approve new Alpha users: Greg mentioned we have capacity for 20 new users to onboard.
  • Review your outreach metrics: you've got a couple weeks of data to look at.
Sources Gmail Google Calendar Granola Notion Slack Google Docs

Catching what you'd miss. Users about to churn, leads going cold, failed payments. Surfaced before they slip.

Things I've handled for you

Today

Yesterday

Do your life's best work, and help the world do theirs

We're building at the frontier of AI, solving problems that look impossible: dynamic long-term memory, proactive agents, context engines that understand your work like you do.

Between us, we've built and scaled startups, published in top journals, and shipped everything from quantum random number generators to local coding agents.

Join us, and spend your best years changing how the world works.

Things we get asked

Which models do you use?

Two things make Catapult work.

The first is the live model we build of your work: a deep understanding of everything you've ever worked on. That's the part we've pioneered, and it's what sets Catapult apart.

The second is how we answer your questions. There we stay model-agnostic. We continuously test the best agents available and match each task to whichever one handles it best, so you always get the strongest result.

Does Catapult have memory?

Yes, and it's real memory, not the shallow kind.

Most tools "remember" by re-fetching a handful of recent messages the moment you ask, then starting from zero next time. Catapult builds a living model of your work and keeps it current in the background. It learns what's top of mind this week, the people you work with, your tone and preferences, what's coming next, and what's starting to slip.

Because it holds the context of every conversation, you never repeat yourself or re-explain what you meant last time.

Ask what's on your plate and Catapult can write your week back to you, often more clearly than you would yourself. That's when it stops feeling like a tool and starts feeling like someone who already knows you.

How does Catapult compare to Hermes, OpenClaw and Claude Cowork?

Catapult is different in two ways that matter.

It's proactive, not prompted. Cowork and Hermes, and OpenClaw wait for you to assign a task: you open the app, point it at a folder, write a prompt. Catapult watches what's happening across your work, from emails to Slack to meetings, and starts on things before you ask. The bug ticket is drafted before you've seen the Slack message. The investor follow-up is waiting when you sit down.

It's built on deep, pre-existing context. This is the bigger one. Cowork, Hermes and OpenClaw rely on MCPs (Model Context Protocol connectors) or CLIs to pull in information the moment you ask. That sounds fine in theory. In practice it means:

  • They start from zero every time. Each task is like opening an incognito window: a few live API calls to Gmail, Slack, or Drive, a thin slice of recent data, whatever fits in the context window. No memory of you, your work, or your patterns.
  • The context is patchy. MCPs return whatever the underlying API exposes, which is usually far less than what's actually in your work. Attachments are the obvious one: most can't read them, so a thread about a contract becomes a thread about "the attached PDF." Same for images in Slack, linked docs, embedded tables, and anything behind a reference.
  • They can't see across sources. A call to Gmail returns emails. A call to Slack returns messages. Nothing joins them up. If the answer lives half in an email, half in a Slack DM, and half in a Notion doc from three months ago, you're the one stitching it together.
  • Retrieval is shallow and recent. MCPs are good at "the last 20 messages in this channel," bad at "what did we decide about pricing six months ago, and who pushed back." The model never learns your history. It just keeps re-fetching fragments of it.

Catapult works the other way around. Instead of pulling context at query time, it builds a persistent model of your work in the background. It ingests everything: emails and their attachments, Slack threads and the images in them, docs, meetings, the calendar, the lot. It keeps that model current as you work, so when Catapult acts, it already knows the context: your tone, your priorities, the commitments you've made, the threads that matter. It doesn't have to go look.

How secure is Catapult?

Your data is yours. It's encrypted in transit and at rest, and it's never used to train third-party models. We comply with GDPR, and we're currently completing SOC 2 and ISO 27001 certification with Vanta.

Does Catapult act without my permission?

No. We believe you should have the final say on everything Catapult does. Catapult prepares work for you to review, so you can hit send or take the action.