Questions
Twenty-six answers, from what it costs to whether it runs inside Claude, and what the four controls in the app are for.
An execution coach for one goal at a time. You declare one task each morning, file proof of it each evening, and your idea moves through five phases — IDEA, VALIDATION, BUILD, LAUNCH, TRACTION — by clearing a bar of accepted evidence rather than by you saying it is done.
It is free, it runs in a browser, and there is nothing to set up.
Yes, and there is no paid tier waiting behind it: nothing to enter a card into, no trial clock, no plan to upgrade to.
If that ever changes, it will appear in the changelog before it appears in your way.
No. Nothing in the signup checks either one — a Google account or any email address works, from anywhere.
The product is written for first-time builders in India who never got a mentor, and the examples come from that world: a tiffin-delivery app, a hostel mess queue. That is who it is aimed at, not who it is limited to. If you are outside that, the loop works the same; some of the examples will not be yours.
Nothing. It is a website. You can add it to your home screen if you would rather it opened like an app, but that is optional and it is the same product either way.
No. If you do not have one there is a workshop that starts from what you have actually noticed — something that annoys you, something people around you keep doing badly — and ends with one goal written down.
If you arrive with an idea already, you can have it reworded instead.
Whichever way you arrive, the first thing he asks for once you commit is one evening at your desk.
No. Tell Masterji in the chat what you have already done — the problem you wrote up, the customers you spoke to, the thing people are using — and he writes it down as a list you press once. Each piece is read against the same bar as an evening's proof, and whatever clears it counts.
You do not skip anything. You arrive having cleared it, or you do not arrive — the count that opens the next stage is the same count, earned the same way. He reads work you did months ago a little harder than a Tuesday night, because one evening after a day job has earned some benefit of the doubt and last year's work has had as long as it needed. Expect to be asked which conversation, with who, and roughly when.
Two limits worth knowing before you start. Prior work carries you as far as building, and no further — if you are already launched, you will still earn the last stretch here the way everyone does. And it is a door at the start: once you have earned an evening of your own, it closes, and what comes next is tonight's work.
It does not build you a streak and it does not move you up a cohort board. Those count days you showed up, and an afternoon of remembering is not those days. Your own record shows the difference, marked as brought in wherever it appears.
It depends on the phase, and the bar is written out in front of you before you file anything — you are never guessing what would count.
A real rejection counts. I posted it is not a stranger acting.
Your streak goes back to zero and starts again that evening. Nothing else happens: no lockout, no penalty, nothing taken away. Proofs you have already banked stay banked, and your longest-ever run is kept on the record even after the goal is retired.
One thing worth knowing about how it counts: a day is complete only when it was both declared and proved. Today never shows as a broken streak while it is still in progress — an unfinished morning is not a missed day until the day is over.
No. The database will not hold a second one.
A second goal is the most comfortable way there is to avoid the first one. You can retire a goal and start another — you just cannot run two at once.
Yes, and the old one does not vanish. Retired goals stay on your record, read-only, with their day-by-day history and their longest streak. Nothing writes to a retired goal again.
You can close one whenever you like, and an idea that dies in front of real people reads as tested on your record — most first ones should.
It does not move.
Advancing a phase is a server decision: it counts accepted proofs in the database and refuses if they are not there. The coach can propose an advance, and the dashboard has a button, but both roads lead to the same check, and that check counts rows.
So leaving VALIDATION needs three conversations with three different people — the same willing friend three times is three rows and one person, and it will not clear. Leaving BUILD needs two proofs with at least one of them a real user touching the thing. You can argue with the coach for as long as you like; the count is still the count.
Nobody's track record. Masterji makes no claim to founder wisdom and has not built a company.
Its authority is procedural. The method is borrowed, and credited by name: The Mom Test for how to talk to people without leading them, The Lean Startup for what counts as evidence, and MAKE for shipping small and often.
The part that matters is that every refusal is a condition written down in advance rather than a judgement made on the day. That is what makes it arguable-with in principle and unmovable in practice — see what if I disagree with the gate. The gate conditions and the playbooks behind them are shared on request.
No. There is no repository integration, no editor plugin and no access to your files, and it does not produce your deliverable.
What it does: ask what you are doing today, help you think out loud when you are stuck, and draft tonight's proof out of what you told it — which you then file. If nobody does the work, nothing gets filed and nothing advances.
A commercial model from one of the major providers, hosted by them. Which one can change — and when it does, the changelog says so.
Not by Masterji. The coach is a commercial model reached over an API, so there is nothing here that could be trained, and nothing you write is gathered into a training set. What he knows about you he is handed at the start of each turn, out of your own rows.
Your chat does leave this server, though, and that is the part worth knowing: every message goes to the provider running the model, with the context built around it — your goal, your phase, your counts, and the proofs already banked. What that provider then does with it is set by their terms rather than by this code, and nothing on this page can promise you otherwise.
What Masterji stores, and where it goes names every recipient, and how long each thing is kept.
For most of it, you should — and if you can build a startup at all, you already do. Nothing Masterji knows about finding a problem or talking to customers is its own: the method is borrowed from published work and credited by name, and the models have read the same books.
What a chat cannot do is refuse you. Reword the question, open a new tab, or push for long enough, and it will help you plan a fourth evening instead. Here the count that opens the next phase is kept in rows the conversation cannot write, and each proof is judged on what you filed rather than on how hard you pushed. That holds when the conversation is with Claude itself: connecting Masterji to it opens a second door onto the same rows, not a softer one.
It also cannot be the stranger. A model makes building cheap; it does not make anyone talk to a person who did not have to be kind about it. That is why three conversations with three different people come before BUILD opens, and no model can have those for you.
Yes. Add https://helpindiebuild.com/api/mcp/ to Claude as a connector and sign in once, and the morning and the evening happen where you were already thinking out loud. Connecting it walks through where that button is on each Claude plan, because it is not where you would look. Claude gets five things it can do for you: see where you stand, declare today's task, file tonight's proof, put a question to Masterji, and ask to open the next phase.
Connect once you have a goal. Two things stay here, and they are the reason: the workshop, for a builder who does not have an idea yet; and screenshots. A link is proof Claude can file for you; an image is filed on the site.
It is the same product, not a copy of it — the same rows, the same judge, the same count. What Masterji stores has a line for the connection: what goes to Claude, how often, and the sign-in it holds for you.
No, and not because Claude is asked nicely. Every tool that writes anything goes through the same endpoint the website calls, so a proof filed from a conversation is the same row, read by the same judge, counted by the same code. What opens a phase has never been a conversation, in either place.
What the connection does ask of Claude is advisory, and a client is free to ignore it: check where you stand before helping you build, and pass Masterji's refusals on in his own words instead of softening them. A client that ignores that is a worse messenger. It is not a softer gate.
One thing is recorded differently: which door a check-in or a message came in by. That is so we can tell whether anybody actually files a proof from inside Claude, and it is on the privacy page with everything else.
By default, only you. Your goal title, your morning declarations, your proof text and everything you said in chat are yours. None of it is public, and none of it sits on a page search engines are allowed to index.
Two things can let something out, and both are an action you take: you generate a share link for your record, or you join a cohort. Both are below.
The phase you reached, the proofs you banked, and how many of them came from talking to real people.
The page carries no title with your goal in it, and it is marked noindex, so a crawler that finds the link is told to forget it. It shares a page with one person, not with a listing.
Counts, and nothing you typed.
Their board shows the phase you reached, your proofs from real contact, your proofs banked, and your day streak. It shows no goal title, no proof text, no reason you gave for closing an idea — not one word you wrote. A member between ideas shows as between ideas, unranked, rather than scored zero. What a coordinator gets is a page of its own.
You join with a code and you can leave. Nobody is counted inside somebody else's aggregate without having joined it.
And you can leave altogether, by asking: write to the address on What we store and your account is erased. Your email, username, name and any password are overwritten, so nothing identifying is left holding your address, and the record comes out of the product.
Continue with Google, or your email and password. If you signed up by email link, you chose the password there. A Google-only account has no password until you set one on your account screen — linked from your username in the corner (or "account" on a phone), or reach it directly from here.
Forgot password? on the sign-in screen mails a link to the address on your account, if it is confirmed. It works once, for twenty-four hours. Choosing a new password there signs you out everywhere else and signs you in; changing it from your account screen keeps the browser you're using.
Your account screen lists every browser you're signed into, and you can end any of them there.
The switch over the box you type in has two settings. The default one runs the daily loop: it asks what you are doing today, it wants proof of it tonight, and it will name it when you are hiding in research.
"Think with me" moves Masterji to your side of the table for that conversation — questions rather than assignments, one at a time, and two or three concrete options on the table when you are stuck rather than another instruction to be more specific. He will say which way he would lean and let you disagree with him. He will not demand a declaration in the middle of a half-formed thought.
What it does not change is anything that counts. The phase rules still hold, the bar is the same bar, and the gate belongs to the server in both modes. It is a way of talking, not a way past the door — so switching costs you nothing, and it is a setting on your account rather than on the tab you are looking at.
Until the first proof is accepted, yes, and it costs you nothing — your streak and your days on the goal are untouched. That is the point of it: rewording used to mean retiring the goal and starting again, which threw both away, so builders polished the sentence before committing instead of getting on with it. Masterji will sometimes offer a sharper version of the line; it fills the box and never saves itself.
After the first accepted proof the link is gone and the server refuses the edit. That is what makes the record worth having rather than a punishment: the evening you banked was work on that sentence, and rewriting it afterwards would rewrite what those evenings were for. The written brief behind the goal locks at the same moment, for the same reason.
It was never a way around the gate, and it cannot become one. A proof is judged against the task you declared that morning, frozen on that day's row — never against the goal's title.
At TRACTION, and nowhere earlier, you are asked which single number means somebody got the value — payments, completed actions, not visits. One, because four numbers is a dashboard, and with four you can always find the one that is going up.
A number that falls costs you nothing. Nothing counts it: the gate has never read it, no phase requires it, and an evening with a number and no proof banks nothing while an evening with a proof and no number banks everything it always did. Zero is a real reading, and the day the number did not move is the day Masterji most needs to see it.
What is on the record is the swapping. Each reading is stamped with the name it was read under, so changing what you watch shows up as what it is instead of quietly re-labelling the evenings you already counted.
It is the room before the goal, for the builder who cannot fill the commit box yet. Every other screen needs a goal to exist first, so without it your first conversation with Masterji would be the welcome message written after the commitment that frightened you.
It is metered, and the meter is server-side rather than a promise in a prompt: twenty turns, and then the only door left is Commit. You can park at most three candidate ideas — three makes collecting impossible and choosing the only move left — and a parked candidate is a bare one-liner with no links and no notes on it, because a candidate that can carry a reading list is a candidate you can hide behind.
It banks nothing and advances nothing; the gate has never read the room. And there is a second, smaller one — five turns, once per goal — for the day you still have your idea and have stopped believing in it, which is the same sentence four days later.