Meeting assistant · Codebase expert · Works with Claude Code
Your meeting ends. Claude already has the plan.
NotCluely joins your meetings as a visible teammate that knows your codebase. It answers code questions live, takes the notes and turns what was decided into tasks for Claude Code. Go get a coffee. When you're back, Claude asks if it can start. Or turn on autopilot and the work is already done.
Sam asked: "Do we retry failed payments anywhere?"
Yes. Failed charges are retried 3 times, 10 minutes apart, in payments/retry.py. Claude added this two weeks ago. There is no retry for refunds yet.
Concept design · after the meeting · try the buttons
Claude · after your 10:30 product sync
Your team agreed on two bug fixes. I read the full transcript and wrote a plan for each.
Bug
Checkout button can be clicked twice and charges twiceRaised by Priya at 10:41 · plan: 3 steps, 2 files
Disable the button while a payment is in flight
Add an idempotency key to the charge request
Add a test that double-clicks checkout
Bug
Invoice dates show in the wrong time zoneRaised by Sam at 10:52 · plan: 2 steps, 1 file
Format invoice dates in the customer's time zone
Add a test for a customer in a different time zone
Both fixes are done and ready for your review. Each one links back to the moment in the meeting that asked for it.
How it works
Before: a brief on what changedA short summary of the code that shipped since your last meeting, including everything Claude wrote. Nobody walks in out of date.
During: a teammate who knows the codeIt joins as a named participant, so everyone can see it. Ask how something works and get an answer from your actual repo, with the file it came from. Say "look at my screen" and it saves a screenshot right then, pinned to that moment in the transcript. Short answers are spoken. Code, diffs and diagrams show up as cards on screen instead of being read aloud. You can also ask it to do things by name: "NotCluely, run the tests on this branch" or "summarize Priya's last commit."
After: the work, ready to goBugs, features and decisions get pulled out with who raised them and why. A summary, every context point and the full transcript go to Claude Code, which writes a plan for each.
You approve, or it just gets doneIn ask-first mode nothing runs until you say yes. On autopilot, Claude starts the moment the meeting ends, tests its own work and hands it to whatever comes next in your workflow. Each change links back to the moment in the meeting that asked for it.
Why we're building it
Meeting tools write notes that nobody acts on. Coding agents wait for someone to type the task. We connect the two. Claude now writes a lot of the code. Engineers still go into meetings and get asked how it works. Then someone has to remember what was decided, write it up and explain it all over again. We want Claude in the room for both halves.
Knows your codebase.Answers questions in the meeting from the code itself, not from memory.
Knows why, not just where.Tickets, design docs and team chat history too, so it can answer "why did we build it this way?"
Visible to everyone.It shows up as a participant. No hidden recording.
Sees what you shared.Say "see my screen" and the screenshot goes to Claude with what was being said at that second. The error on screen and the words about it stay together.
Notes and a full transcript.Every meeting, saved and searchable. Ask what was decided last Tuesday.
Tasks, not just summaries.Each decision becomes something Claude can plan and act on.
You set the rules.Read-only during meetings. Ask first by default. Autopilot only where you turn it on.
Two ways to work
Ask first.Claude plans while you're away and waits for your yes before touching anything.
Autopilot.Finish the meeting, get a coffee, come back to tasks that are done and tested. They go wherever your team's process sends finished work: a pull request, QA or staging. Our goal is a working preview link you can click before the call even ends.
Autopilot to production.For fast-moving apps that want it. Claude ships only after its own tests pass, and only on the projects you allow.
Safe by design
The meeting can't run code.What people say is treated as notes, never as commands. A separate, locked-down step decides what is allowed to happen.
Only your team can assign work.Guests and clients on the call can ask questions, but they can't start changes.
Risky steps need a tap on your phone.Database changes, new packages and production deploys ask the repo owner to approve on their own device. A voice saying "yes" never counts.
Sandboxed and tested.Every change is built and tested in an isolated sandbox before it goes anywhere.
A full activity log.One page shows everything the agent did, each item linked to the moment in the meeting that caused it.
Your Claude, your choice
Claude Code on your computer.Hand tasks to the Claude Code you already use, signed in with your own Claude account.
Your own Claude API key.Connect a key from the Claude Console and usage is billed to you directly.
Your code stays put.The repo is read on your machine. We don't copy it to our servers.
Where we are
This is the next product from the team behind NotCluely, which has paying customers today. It's in development now. People on the waitlist get access first.