Standalone tools: the plan to let friends use the strategy tools without a client

The problem behind it

I wanted to run the strategy and brand tools without creating a YuktiBOS client: for testing screens, for friends, and later maybe as a lead magnet or a paid product. Without YuktiBOS, a user gets the plan and does the work themselves. With YuktiBOS, the same tools feed real clients.

What exists today

The Market Finder has a public version, behind a switch that is off by default. A visitor describes a business in a few words, waits about three minutes, and gets 16 markets, a shortlist of three, a score each and messages to test. They can leave an email to get the report. By construction a public run has no client, so it reads no client data. The Strategy tools (Offer, Pricing, Assumptions, Pack) can also run standalone, with no client, for the super admin or the person who started the run.

What is a plan

The "Studio" plan (version 2, 31 October) is a document for my review. Nothing in it was built when it was written. I have not checked later notes to see how much has been built since, so treat it as a plan. It was critiqued by an independent agent that read the code, and this is what changed because of that review.

The decisions, and why

Shutting workspaces out needs more than an "active" filter. Background jobs were safe, but seven list queries would still show a test workspace: the client directory, the work list, Yukti's company snapshot, the MCP client list and three queries with no filter at all. So the first step is to close them, with tests.

No new user role. The existing client login is already tied to one client and has an email invite flow. A half-built new role would fail open and see every client. Friends use a client login plus a Studio module that shows only Studio.

AI spend per client could not be counted. The AI log did not save the client for every call, and Market Finder never set it. The only caps were per worker and per provider, so one friend could use up the day's cap and block real client work. So every call gets tagged with a client, and workspaces get their own cap and budget pool. The default I proposed is Rs 300 per friend workspace.

Privacy both ways. A friend's text must stay out of the AI learning loops, so one person's material cannot change real clients' work. Friends need consent text, "delete my workspace" and an export.

Public sign-up must not live in Mabbit's database. That was decided on 18 October: self-serve users go to a separate shared database. Invite-only friends inside the main database are a recorded exception. The lead magnet and payments wait for the separate database.

"Connect to YuktiBOS" cannot be a one-click status flip. It would instantly start planning jobs and digests. It must go through normal client onboarding.

The real goal of version 1 is UI testing. So every Studio screen gets a feedback box.

The steps

Invites would expire after 7 days and work once.

What went wrong

The first version of this plan, v1, assumed too much. The critique found the leaks above and the uncounted AI spend. Had we built v1, a friend's test could have blocked a client's work.

What's next

I need to confirm how far A0 and A1 got. The brand voice tool is listed as a candidate for sale on its own, with a selling page; I have not decided pricing.

Numbers at a glance