# Rebuilding paid GTM tools in the open, on one key

Eleven open-source rebuilds of paid GTM tools on one API key: the quoted cost of a run, what each leaves out, and why none has had a paid test yet.

## Summary

A paid GTM tool is three things in one box: data, a workflow and a screen. The screen and the workflow are software, and software is cheap to write now. The data is the part you cannot write, and most of it is for sale by the call.

So I tried the obvious experiment. I took eleven paid GTM tools, rebuilt the data half of each as an open-source repo that runs on one API key, and wrote down what a run is quoted at and what I had to leave out. This post is the result, limits first.

Disclosure. I founded looot, and every one of these repos calls looot for its data. Each repo is free and MIT licensed, and each one is also a reason to buy looot credit. Judge the claims with that in mind.

Warning: No paid run was made on any of these repos. Every parser is built from provider documentation and fixtures, and every cost below is a quote at list price, not a measured bill. Treat each repo as unverified until a real run confirms it.

## What was rebuilt, and what does a run cost?

Each repo stands in for the data half of one or two paid tools. The table gives the quoted cost of a typical run at looot list prices and the features that did not make it.

Eleven open-source rebuilds: what a run is quoted at, and what each leaves out. Read on 2026-10-10 from [the README of each repo at github.com/loootai](https://github.com/loootai).

| Repo | Paid tool it stands in for | Quoted cost at list price | Left out |
|---|---|---|---|
| looot-intent | 6sense | $0.1720 per account | First-party visitor tracking, bidstream intent data, CRM sync, ad orchestration |
| looot-score | MadKudu | $0.83 to $1.67 per 100 leads | Behavioural scoring, a model trained on closed-won deals, CRM sync |
| looot-reveal | RB2B, Clearbit Reveal | about $0.08 per 100 new IPs | Person-level identification, email finding, hosted multi-user mode |
| looot-techstack | BuiltWith, Wappalyzer | $1.89 per 100 domains on the default three providers | List filters by country or spend, history trends, a browser extension, an own fingerprint database |
| looot-signals | TheirStack, Harmonic | up to $0.531 per run of five triggers | Founder, board and investor data |
| looot-lookalike | Ocean.io | up to $0.2138 for 6 seed companies | Contact discovery, CRM sync, scoring on revenue |
| looot-ads | Foreplay | $0.1057 for 5 competitors on 4 platforms | A discovery library of brands you did not name, analytics on your own ad account, AI briefs, creative storage |
| looot-reviews | ReviewTrackers | up to $0.015 for 5 companies on Google Maps | Trustpilot, G2 and Yelp, replying to reviews, continuous monitoring |
| looot-battlecards | Klue, Crayon | $0.0153 per competitor | Win-loss interviews, CRM and call-recording insight, publishing to Slack or a CRM, change alerts |
| looot-mentions | Brand24, Mention | $0.00663 per keyword on 6 sources | Facebook and Instagram keyword posts, podcasts, share of voice, volume-spike alerts |
| looot-warm-leads | Gojiberry | up to $0.25 per run on the new-role trigger | Company followers, lookalike leads, automatic sending, AI-drafted replies, CRM sync |

Costs are each repo's own quote at looot list prices, before any call. No paid run was made on any of these repos, so no cost here is a measured bill. The left-out features are as each README lists them, not checked on the vendors' own sites.

The repos live at [github.com/loootai](https://github.com/loootai). Each has a test suite that runs on fixtures, a dry-run mode that prints the quote before any call, and a spend cap that stops a run before the call that would pass it.

## How were the cost figures made?

Each repo plans its calls, looks up the list price of each endpoint and adds them up before it runs. That quote is what the table shows. It is the most a run should cost when every call is made and every call is billed.

Three things make a real bill differ from the quote:

- Misses. Some providers bill a call that finds nothing and some do not.
- Fallbacks. A repo that tries a second provider when the first is empty spends more on hard rows.
- Price changes. The quotes are dated 2026-10-10, and list prices move.

The third one has already happened. When I read the catalog again on 2026-10-11, three ad library endpoints that looot-ads calls were no longer in it. That repo's quote is out of date until it points at other endpoints. I left its row in the table and I am telling you here, because a frozen quote that looks precise is the kind of number this article is meant to question.

## What does a paid tool have that these repos do not?

Reading the "left out" column from top to bottom, the gaps fall into four groups. This is the most useful part of the experiment.

**Data nobody sells by the call.** Bidstream intent, a first-party visitor graph, a library of ads from brands you did not name, review sites that offer no endpoint. looot-intent scores an account from hiring, funding, news and tech changes. It has nothing like a third-party intent pool, and no open rebuild will until someone sells that data.

**Models trained on your own outcomes.** looot-score grades a lead from firmographics and tech stack with fixed points per rule. Its README lists a model trained on closed-won deals as the part it left out. Rules are easier to read and worse at surprises.

**The loop around the data.** CRM sync, a shared inbox, automatic sending, win-loss interviews. None of that is data. It is product.

**People, not companies.** looot-reveal returns a company from an IP address and stops there. Naming the person needs a licensed identity graph. I wrote up [how visitor identification works](https://looot.ai/blog/how-website-visitor-identification-works) separately.

My conclusion: the data half of these tools can be rebuilt on public endpoints, and a run is quoted in cents, not in seats. The other half is where paid tools earn their price. If you only need the data, in a table, on your own machine, these repos cover that half.

## Why are vendor prices not listed here?

Because the picture is uneven. Some of these vendors publish a price list. Several publish none and ask for a demo, and for a few the only figures I found were third-party estimates. A table that is half fact and half estimate helps nobody. Each repo README says where its vendor figure came from and on which day, and labels third-party figures as estimates.

Compare against the price you pay, on your own invoice. That number is the one that matters.

## What is untested in each repo?

All of it, in the strict sense, since no paid run was made. These are the specific weak points I know about:

- **looot-techstack.** All provider parsers are unverified. The inputs of the reverse search, which finds companies that use a technology, are inferred from endpoint descriptions.
- **looot-signals.** News headlines carry no structured company, so the company is guessed from the headline.
- **looot-lookalike.** Industry names are compared as lower-cased text, so two labels for the same industry do not match.
- **looot-battlecards.** Pricing pages that load their numbers with scripts come back empty. Review themes are keyword rules on search snippets, which is a weak signal.
- **looot-mentions.** Sentiment is a word list. It misses sarcasm.
- **looot-reviews.** Themes are word counts on English text, and sentiment is the rating the reviewer gave.
- **looot-reveal.** The tracking snippet and its collect route were not load tested, and the rate limit lives in memory in one process.
- **looot-warm-leads.** Sending on LinkedIn goes through your own account with a third-party service, is off by default, and was never called.

A first paid test of every parser is planned. It runs one small real job per repo, compares the raw response with the parser, and records the real cost next to the quote. Until that is published, the right word for each repo is "unverified".

## Which rebuild is the easiest to check yourself?

Technology detection. It is one job, the answer is easy to check by opening the site, and several providers sell it at different prices.

Detect the technologies a company uses. Prices read on 2026-10-13 from the live looot catalog. They change when a provider changes its price.

| Option | Cheapest provider | Price per call |
|---|---|---|
| Technology lookup for one domain | Tomba | $0.0089 per call |

looot-techstack calls three providers for the same domain by default. That costs more than one call and may catch technologies a single provider misses. Whether it is worth the extra spend is what a paid test should measure. If you use a paid lookup tool today, the [BuiltWith alternatives page](https://looot.ai/alternatives/builtwith) lists which of its jobs the catalog covers.

## Try it: detect a tech stack with your agent

The same pattern, plan then quote then run, is what [an agent does in place of a Clay table](https://looot.ai/blog/can-an-ai-agent-replace-clay). The endpoints and their prices are in the [public catalog](https://looot.ai/public-catalog).

## What this article cannot claim

- That any of these repos matches the paid tool it stands in for. None was compared side by side.
- A measured cost. Every figure is a quote.
- Accuracy. No parser has met a live response.

Costs read from the repo READMEs on 2026-10-10, catalog checked again on 2026-10-11. Written by Walid Boulanouar, who founded looot.

## Questions

### Can you rebuild paid GTM tools with open source?

The data half, yes. Hiring, funding, tech stack, ads, reviews and company data are sold by the call. Proprietary data, trained models and the workflow around them are not, and the rebuilds leave those out.

### What does an open-source GTM tool cost to run?

Each repo quotes a run at list price before it starts. The quotes in this article range from under one cent per keyword to under two dollars for 100 domains, and none is a measured bill.

### Have these open-source rebuilds been tested on real data?

No. No paid run was made on any of them. The parsers are built from provider documentation and fixtures, so each repo should be treated as unverified.

### What can a paid GTM tool do that an open-source rebuild cannot?

Four things: data nobody sells by the call, models trained on your own closed deals, the workflow around the data such as CRM sync and sending, and person-level identity.

### Do I need a looot account to run these repos?

For real runs, yes: a key with a funded balance. The demo and the dry run need no key. The code is MIT licensed and runs on your own machine.

### Why does this article not compare vendor prices?

Some vendors publish a price list, several publish none, and for a few only third-party estimates exist. Each repo README says where its vendor figure came from and labels estimates as estimates.

## For agents

### Tech stack for a list of domains

```text
Use looot to detect the technologies these five company websites use.
1. Set up https://looot.ai/skill.md.
2. Search the catalog for technology detection endpoints and show me each price before you run.
3. Run the cheapest one for each domain. If it returns nothing, try the next provider.
4. Show one table: domain, technologies found, which provider answered, cost. Add the total.
```