How the Stack Builder Scores

The rules come first

The builder is a decision tree, not a ranking. Each job has a ladder of plain rules, read from the top, and the first rule whose conditions hold and whose plan fits the money left for that job makes the pick: a tool you already pay for; under privacy, a tool that keeps your work private or runs on your machine; under a set amount of money, a cheaper or free plan; otherwise the tool built for the job. Jobs run in your priority order, the first with all the money. The rules are public in stack-rules.json, every card shows the step behind it, and the result lists every decision in order under "Why this stack". The formula below only breaks a tie when one step names more than one tool.

The formula

Every tool gets one score: coverage × budget fit × privacy fit × (0.8 + 0.2 × signal). The builder picks the highest score, adds that tool to your stack, and scores every tool again against what the stack now covers. It stops when no tool improves anything you ticked. Tools you already pay for go in first, at no cost. The same answers always give the same stack, and the score is printed under every pick so you can check it.

Coverage: what the tool does for your jobs

Each tool on the list is rated 1, 2 or 3 for each job it can do: 3 means the tool is built for it, 2 a solid feature, 1 it can manage. Coverage adds up how much better the tool does each of your ticked jobs than the stack so far, weighted by your priority order: with five jobs ticked, the first counts five times and the last once. A tool that only repeats what the stack already does well scores nothing and is left out.

Budget fit: the plan a small team starts on

Each tool has the plan a small team usually starts on, read from its pricing page with the date it was read. If that plan fits what is left of your budget, the builder uses it; if not, it uses the tool's free plan, if there is one, and otherwise skips the tool. Budget fit is 1 for a free plan and falls toward 0.5 as the plan's price nears your whole budget, so a cheaper tool wins a tie.

Privacy fit and signal

If you prefer tools that run on your machine or can be self-hosted, cloud-only tools score 0.6 of what they would; otherwise privacy fit is 1. The signal is built only from public counts: the App Store rating weighted by how many people rated it, GitHub stars, and Hacker News stories in the last 30 days. Hacker News leans to developer tools, so a mention can lift the signal and never lowers it; a tool with no public counts gets 0.4 out of 1, a little under a tool with modest counts, since no counts is no evidence. The signal can move a score by a fifth at most, so it breaks ties between tools that do the job equally well rather than deciding the stack.

How the budget is spent

Once every job you ticked has a tool, the builder spends what is left rather than leaving it idle, on the job you put first: it raises the plan of the tool that serves that job one tier at a time (when that tool has no bigger plan, such as a free engine, the raise passes to the tool for the next job), and only to a plan whose extra it can say in one line, such as five times the usage or 7,000 credits a month instead of 1,500. Every other tool stays on the plan a small team starts on. That line shows on the card as "Why this plan". A plan with no such line is never an upgrade target. When no raise fits or helps, the result says how much is left and that nothing else on the list would improve the stack for those jobs; adding a job is the way to use it.

When the tool for the first job did not move up (it has no bigger plan, none fits, or you already pay for it), the builder adds a second tool built for that job, on its starting plan, and the card says why. That happens only for work such as writing, coding, images or video: a second store, domain, site, analytics tool, email tool or scheduler would only duplicate the first. A pick that no longer does any job best once the stack is in is dropped. If money is still left, the result names the cheapest step up that fits, priced as the extra per month, with what it gets, rather than buying it for you; with ads on your list it also says the money left can go to ad spend, since ad clicks are paid on top of the account.

Some tools are bought once rather than by the month, such as Reaper or an Ozone edition. Their price shows as paid once and stays out of the monthly total, and the builder offers one only when a month of your budget would cover it, so a $0 budget never gets a paid licence.

The model line

If your jobs include writing copy, coding or building an app, the stack ends with a model for your agents and automations, paid per use. The builder estimates a rough monthly volume of tokens for your jobs, prices it at OpenRouter's live prices, and picks, among four models, the one rated best for that kind of work (coding, writing or analysis) in LMArena's public category leaderboards whose cost stays under a quarter of what is left. The card names the rating, the rank and the leaderboard's date. Aider's and SWE-bench's coding leaderboards would be the first choice for coding, but neither lists these models yet. The model line is only for calling a model from your own code or automations, so it is always marked optional and kept out of the total, and it is not shown at a $0 budget. If a plan in your stack already gives you a model in its own app, the card says so without claiming that plan includes a different company's model. The model price board shows every price it reads from.

How much of your week

The builder asks how much of your week goes to the first job. A few hours keeps every tool on the plan a small team starts on; most days lets the first job's tool move up one plan; all day lets it move up as far as the money allows, which is the only way to reach the largest plans, such as Claude Max 20x. Before each step up it compares, at that same price, the other tools built for the job, and takes one that does better instead. A step-up plan that already covers another job you ticked replaces that job's pick (ChatGPT Pro includes Codex, Claude Max includes Claude chat), and two picks on one subscription are merged.

A pick that costs more than twice a cheaper tool built for the same job stays only with a line in our list saying what it does that the cheaper one does not; otherwise the cheaper tool takes the job. With privacy asked for, a free plan that makes what you create public, such as Ideogram's or Leonardo's, is passed over; without it, the card says so.

AI tools only

The builder picks AI models, agents and AI-powered tools. A platform or channel, such as a game engine, a store builder, a domain registrar or an ad network, is never a pick: it shows on a card as what the tool works with (a coding agent works with Godot, Unity and Unreal; an ad creative tool works with Meta, Google and Microsoft ads). A job that no AI tool on the list does well enough to pay for gets one line naming the platforms people run it on, instead of a pick. Nameryn does not recommend tools that generate finished music: Make music offers tools that help you compose, mix, master and transcribe, or a library of licensed tracks, and sound effects are a job of their own.

What never counts

• Commissions. A few Try it buttons are affiliate links, and they say so. The formula has no input for them; a test checks that giving every tool a commission, or none, changes no score and no pick.
• Notes on what people say. Each pick's card shows short notes filed by hand from Hacker News comments, each with a date, a link and the account that wrote it; the account was at least a year old when it posted and has no affiliate links in its profile. They are there to read, not to score.
• Ads, sponsorships and reviews. No tool pays to be on the list or to rank.

Limits

The list is hand-checked, so it misses tools and its prices age; each tool shows when its price was read. The ratings of what each tool does well are judgement, applied the same way to every tool. The builder cannot see your contracts, your team size or the volume you need, so treat the stack as a starting point and open each tool's own pricing page before you buy. Try it with your own budget on the AI stack builder.

Open data: use it, credit Nameryn

The tool list and the rules are published as they are, under CC BY 4.0: ai-tools.json, every tool with its plans, prices, source page, checked date and confidence, and stack-rules.json, every job's ladder of steps. Each file carries its version, which is the date it last changed. Use them in your own work, a model's answer included, and credit Nameryn with a link to nameryn.com/ai-agents/tools/.