The operating system for AI-powered product teams.
ProductMaxxing is a product execution framework. It turns every customer call, Slack thread and ticket into sized work in your PM tool, puts each PM’s AI agent on the highest-value task it can finish, and waits for your review.
The difference
Same customer ask. Three weeks, or one day.
Without PMX
- Mon 11:00A customer explains what is blocking them.
- Mon 11:40Your notes go into a doc. Two more calls follow.
- WedA Slack thread with the same complaint scrolls out of view.
- Week 2Backlog grooming. Someone writes the ticket from memory.
- Week 3It gets a size by guesswork and waits for a sprint.
With PMX
- Mon 11:00A customer explains what is blocking them.
- Mon 14:00PMX proposes a sized task and says why it got that size.
- Mon 14:05You approve it. It reaches your PM tool at the next sync.
- Tue 09:00Your agent claims it: the best ready task within today’s budget.
- Tue 11:30You review the draft and accept it.
Illustrative timings.
How it works
Five steps from a customer’s words to reviewed work.
- 01
Collect
A scheduled task in your Claude app copies transcripts from your meeting apps, such as Granola or Fireflies. Customer calls and Slack threads are read. DMs and personal meetings are not.
- 02
Propose
PMX turns the day’s customer conversations into proposed tasks. Each one has a short summary in the model’s own words, the reason for its size, and the dates of its sources.
- 03
Approve
You see each task exactly as your PM tool will get it. Approve one, several or all, and drop the rest. Nothing reaches your PM tool before you do.
- 04
Pull
Each PM’s agent claims the best ready task that fits the budget and the review cap, and runs it. Only one agent can win a claim, so no task is done twice.
- 05
Review
Accept the draft, reject it, or send it back with a note.
pmx explainshows what the run used, what it left out, which model ran and what it cost.
What it runs on
- Machine
- A Mac with Node 22.12 or later
- Agents
- Claude, through a clean profile PMX keeps separate from yours
- PM tool
- The one your team already uses, such as Linear, Aha! or monday.com
- Sources
- Meeting apps connected to Claude, such as Granola and Fireflies, plus Slack
- Limits
- An optional daily cap of your own, a team cap, a review cap and your Claude plan’s limits. Set your cap to zero to pause PMX.
Architecture
Transcripts and task text are never stored on our servers.
PMX runs on your Mac, between the tools your team already uses. Pick a boundary to see what crosses it.
PMX runs on your Mac, between the tools your team already uses.
- Customer calls and threads are read and summarized on your Mac, with your own Claude plan.
- Only tasks you approve reach your PM tool, as summaries.
- Agents coordinate through a git repository your team owns.
- The optional hosted app keeps IDs and counts, never transcripts or task text.
Calls are handled on your Mac, with your own Claude plan, and most of what PMX produces stays there.
- A scheduled task in your Claude app copies meeting transcripts in. PMX’s own runs never get connectors.
- Only calls with someone outside your company become team tasks. Meetings about 1:1s, hiring, performance, pay or health are held back and discarded.
- Your own to-dos from meetings, including 1:1s, never leave your Mac.
- Proposed tasks and digests wait in staging until you approve them.
- Every digest and synthesis run leaves a receipt: owner-only, kept 14 days, never committed to git.
- Agents run in a clean Claude profile, limited to the repo, with the sandbox on and permissions never skipped.
Only tasks you approved, as summaries, with their status and links.
- Each task is a short summary in the model’s own words, with why it got its size and when its sources were.
- PMX refuses any description that repeats eight words in a row from a meeting.
- Changes travel through your Claude app’s connector and arrive at the next sync.
- Nothing about the meetings themselves is written there. Processed meetings are recorded in your PMX repo under an opaque ID.
- Everything PMX wrote can be listed, and any task withdrawn: canceled, with its text replaced. Your PM tool may keep the old text in its history.
One git repository that your team owns.
- A claim is a git ref that only one push can win, so no task is done twice.
- Digests are committed only after you approve them.
- Team memory changes only through a pull request that a code owner reviews.
- Each agent’s output PR carries a run record. CI recomputes the run’s environment fingerprint and rejects records that don’t match.
- The run log is hash-chained per agent and checked by
pmx doctor. - Team settings change through one pull request, and the whole team moves to a new version together.
IDs and counts. Task text passes through and is never kept.
- The hosted app is optional, and nothing is sent until you run
pmx login. - It stores IDs, sizes, statuses, counts and timings. Never task titles, descriptions, transcripts, drafts or source text.
- Its board reads tasks from your PM tool and passes them to your browser without storing them.
- If you connect your PM tool there, it keeps an encrypted token for it, plus your members’ emails.
pmx analytics --showprints exactly what is waiting to be sent, andoffstops it.
Pricing
Free at launch.
PMX is free at launch, including the hosted team board. Pricing is coming soon.
Install
Install it in two commands.
On your Mac, with Node 22.12 or later, run:
npm install -g productmaxxing
pmx init
pmx init asks for your company’s email domain, sets up or joins your team’s PMX repo, keeps tokens in your keychain and saves nothing until you confirm. Then pmx doctor checks everything. You’ll need macOS, Node 22.12 or later from Homebrew (not nvm), a Claude plan and your PM tool.
Get an email about new versions and pricing.
Only for releases and pricing news. Unsubscribe in one click.