Seven ways to start a hunt

What happens

1. Turn your words into searchesAnything you pin, such as a field to stay in or a category to avoid, is held exactly as you wrote it.
2. Search GitHub, a repository index and the webFor engines, products and the complaints around them.
3. Group the findings into marketsThen look up who already sells in each, because a decision needs names and a count.
4. Read what the leaders’ own users complain aboutWhere products charge and users complain, demand is proven and the complaints say what is missing.
5. Go further, if you ask it toIt invents answers to the needs it found, borrowed from other fields, and checks each against what already exists. You choose how far to reach, in one to three rounds.
6. Keep up to three and research eachWho serves it, what would set it apart and how it could win. The markets it passed on are kept with what it saw there.

What you get

• Up to three candidates, each in one sentence: what to build and for whom.
• Who already sells a fix, what they charge and what their users complain about, with a link to each source.
• A verdict on each: start, narrow or stop.
• The markets it passed on, and why.

A first hunt typically costs about 41 credits, and a kept candidate about 15 more to check. From a first hunt to a project and a build brief is typically about 110 credits. These are typical figures, not a promise.

What happens

1. Find apps with proof of demand, or the one you namedStore figures, extension stores, for-sale listings and “alternative to” threads. If a name fits several apps, it asks which one.
2. Cut at the gatesA network, licensed content, certification, hardware or anything on your never list ends it, whatever the demand.
3. Scan each survivor for weaknessThe date of the last version, complaints about price rises and paywalls, platforms it does not cover and what its own users ask for.
4. Work out where the value livesLocal logic a rebuild can match, a wrapper around someone else’s service, or a server core that cannot be rebuilt.
5. Size the rebuildThe core job in one sentence, and any open source parts that already do some of it, against your size limit.
6. Write the case and the verdictFour factors, demand, weakness, feasibility and angle, each graded verified, observed, estimated or unknown, with the link that shows it.

What you get

• Find mode: up to three apps, each with a rebuild case and a verdict.
• Check mode: one app, one verdict and the evidence behind each factor.
• A legal line in the brief: build from public behaviour, never from the app’s code, name, icon or text.
• The unknowns, each with what would change the verdict.

Finding apps typically costs about 65 credits and checking one named app about 24. A check, a shaped plan, a project and a brief typically come to about 60. These are typical figures, not a promise.

What happens

1. Write searches for engines only programmers can runAnything you pin is held exactly as you wrote it.
2. Search GitHub, a repository index and the webThe index is searched by meaning and by words, so an engine is found by what it does as well as by its name.
3. Group what came back into marketsA free engine is not a market. What matters is whether people pay for the result, and who sells it today.
4. Read what the leaders’ users complain aboutThat says what an interface on a free engine would have to do better.
5. Keep up to three and research eachWho serves it, what would set it apart and how it could win.

What you get

• Up to three candidates, each built on a named engine.
• Who would pay for the interface, who sells one today and what they charge, with links.
• The engine’s licence and when it was last updated.
• A verdict on each: start, narrow or stop.

A first hunt from an engine typically costs about 41 credits, the same as From anywhere, and a kept candidate about 15 more to check. These are typical figures, not a promise.

What happens

1. State the idea back in one sentenceThe buyer, the job and how it would be paid for. If your message leaves the buyer or the job open, it asks one question and waits.
2. Find who serves the need todayPaid products, free tools, open source projects, and the spreadsheet or manual routine people use instead.
3. Read what they charge and who forFrom pricing pages. The price says whether this buyer pays for software at all.
4. Look for people asking for itRequest boards, forums, review complaints and “is there a tool that” threads.
5. Scan the closest products for gapsThe gap your idea would live in has to be one their own users name.
6. Size the build and name what is unknownThe core job, the parts that already exist and what to find out before you build.

What you get

• A verdict: worth building, worth building for a narrower buyer, a real need served the wrong way, or no evidence of a need.
• Who serves the need now, what they charge, who is asking for it and what it would take to ship, each with evidence or marked unknown.
• A table of who serves the need today: product, price, platform and buyer.
• The unknowns, each with what would change the verdict and where to look.

A check typically costs about 32 credits. A check, a shaped plan, a project and a brief together typically come to about 80. These are typical figures, not a promise.

What happens

1. Find the products that serve the marketThe web, app stores, directories and open source alternatives, searched in the words a buyer would type.
2. Read price, platform and buyer for eachFrom pricing and product pages.
3. Scan the five leaders for weaknessesIn their users’ own requests, complaints and paywall grumbles.
4. Write the map as a documentOne table you can sort and argue with, kept in your project when you name one.
5. Place your idea on it, if you gave oneWhere it fits and who it competes with directly. A conclusion the evidence cannot carry is marked as inference.

What you get

• One row per product: price, platform, buyer and weaknesses, each with the link that shows it.
• The segments the market falls into, and the one least served.
• What a newcomer would have to do better, according to the evidence.
• Where your idea sits, if you described one.

A market map typically costs about 45 credits. A broad market can cost more. These are typical figures, not a promise.

What happens

1. Read your projectIts candidates, documents, buyer and what the product already does.
2. Find the products this buyer compares it withThey are what a feature has to win against.
3. Read what each one offers and chargesFrom pricing and feature pages.
4. Sort the featuresMost rivals have it: you need it to compete. Few have it and your buyer wants it: a wedge. Your buyer asks and nobody offers it: an opening.
5. Check the effortWhether a free engine already does a feature, or it means new work.
6. Write the plan into your projectBuild now, wedge, later or skip, with the evidence and effort beside each.

What you get

• A document named Feature plan in your project.
• A table of features against the competing products, with each product’s price.
• Every feature sorted as must-have, wedge, opening or skip, then planned as build now, wedge, later or skip.

A feature plan typically costs about 50 credits. These are typical figures, not a promise.

What happens

1. Restate the problem and its limitsThe language, the licence and where it has to run decide what counts.
2. Search the repository index and GitHubBy meaning and by words, so describing what the code does is enough.
3. Read the strongest five to tenLicence, last activity, form (library, command line or API) and fit, read from the repository itself.
4. Rank them and recommend oneWith what integrating it would take. Name a project and the shortlist is written into it.

What you get

• A ranked shortlist of repositories.
• For each: licence, last push, form, fit and what integrating it takes.
• One recommendation.

A component search typically costs about 20 credits, the cheapest of the seven. These are typical figures, not a promise.

Every claim has a link

Each page the agent reads is saved to your library, and a result may only cite pages on file.

A stop is a result

When the evidence says nobody pays or the market is served, the verdict says so. See the sample verdict, which is a no.

Public pages only

It reads what anyone could read. Private sales, quotes on request and word of mouth leave no trace, and an empty search is never taken as proof.

You can steer it

Pin the buyer, rule out a field or ask why it kept something, in the chat beside the hunt, while it runs.

You pay for what it used

The maximum is set aside before a run, only what it used is taken, and a failed or cancelled run returns everything.

The full method, including where it goes wrong, is on how it works.