# Reprompt AI

Aiolum story: https://aiolum.com/story/reprompt-ai/ · added 2026-11-08 · updated 2026-11-08

**In short.** Reprompt AI (AI research agents that keep business and location datasets up to date, United States): Acquired by Rippling, November 2026. Evidence grade E: no revenue figure was checked. Team: 2 co-founders in late 2023; team size 3 on its Y Combinator profile, read November 2026. Launched 15 March 2024; sold 32 months after launch.

AI research agents that keep business and location datasets up to date · United States · AI-native · team of 2 to 10 · sold · group: Documents, data and slides

- Outcome: Acquired by Rippling, November 2026 (evidence grade of the headline number: E; no revenue figure)
- Launch: 15 March 2024; months from launch to the outcome: 32
- Team: 2 co-founders in late 2023; team size 3 on its Y Combinator profile, read November 2026
- Outside funding: Y Combinator, Winter 2024 batch (amount not stated in the sources read)
- Main factor: Domain experience
- Meters, 1 to 4: evidence strength 1, copyable pattern 1, capital needed 2
- Tags: AI-native, Small team, Funded, Exit, B2B, Business data

Links:

- Visit repromptai.com: https://repromptai.com/

**The product.** AI agents that research businesses and places on the web and return structured facts: whether a shop is open or closed, its hours, its official website, its category, its building entrance, and checks for know-your-business (KYB) and fraud. Customers send their own lists through an API or an AI table, or use Reprompt's own database of more than 200 million businesses.

**Why it worked.** Map, delivery and payment companies hold lists of millions of businesses that go out of date every day, and people who check them by hand are slow. The co-founders came from Mapbox search and Robinhood growth, and built agents that check each fact against several web sources.

Evidence grades: A = disclosed by a public company, a filing or a payment processor, or verified by one; B = two or more independent reports agree, or a journalist confirmed it; C = a founder claim that no one else verified; D = reported, not verified: one founder claim or a third-party estimate with no second source; E = no revenue figure found, so the headline is a non-money fact and the other facts carry their own grades; calc = computed on this page from graded facts.

## Summary

Reprompt AI sold AI research agents, programs that use a large language model to plan web searches, read the pages they find and return a structured answer. Its agents keep data about points of interest (POI), the shops, restaurants and other places that appear on a map, up to date, and they run KYB checks (know your business: checking that a company is real, registered and operating at its address) for payment companies.

Lukas Martinelli and Rob Balian founded it in late 2023 and joined Y Combinator's Winter 2024 batch. Their first launch, on 15 March 2024, was a tool to test and fix AI apps; in January 2025 it announced an agent for Foursquare's place data. On 6 November 2026 the software company Rippling announced that the team behind Reprompt AI was joining it, and Reprompt's own home page now says that Reprompt has been acquired by Rippling.

This is a grade E story: no source we read gives Reprompt's revenue or a sale price. Its Y Combinator profile shows a team of 3 people. Rippling named Visa, DoorDash and Tripadvisor as customers, and Reprompt's profile named the mapping companies Radar and TomTom.

## Pricing

Reprompt's only published prices, from its blog post of 27 July 2026 (grade C); the main data products are sold through demos and sales calls

| Plan | Per month, billed yearly (US$) | Per month, billed monthly (US$) | Limit |
|---|---|---|---|
| Search API (ranked web results) |  | 5 per 1,000 requests | pay per request |
| Ask API (short answer with citations) |  | 5 per 1,000 requests | pay per request |

The address repromptai.com/pricing returned page not found on 8 November 2026. The enrichment, fraud and KYB, last-mile and dataset products show "Get a Demo" and "Book a call" buttons and no public price. The blog post says the Ask endpoint has no separate token surcharge.

## How much did Reprompt AI make, and what was it sold for? The numbers, graded

A three-person AI data company whose team Rippling took in 32 months after its first launch; no revenue figure or price was published

| Fact | Value | Grade | Note | Sources |
|---|---|---|---|---|
| Company start | late 2023; 2024 on its Y Combinator profile | B | Rippling's announcement: "Lukas Martinelli and Rob Balian founded Reprompt in late 2023". The Y Combinator profile gives 2024 as the year founded. The earliest Internet Archive copy of repromptai.com is from 23 March 2024. | [1, 3, 11] |
| First public launch, 15 March 2024 | a tool to evaluate and improve AI apps without code | C | Y Combinator launch post: trace AI calls, flag hallucinations as bugs, and write prompt overrides. The place-data agents came later. | [4] |
| Team | 2 co-founders; team size 3 | C | Y Combinator profile, read 8 November 2026. Martinelli is CEO and Balian CTO on Reprompt's about page. | [3, 12] |
| Funding | Y Combinator, Winter 2024 batch | B | Rippling's announcement and the Y Combinator profile. No source we read states an amount or a later round. | [1, 3] |
| Customers named | Visa, DoorDash and Tripadvisor; Radar and TomTom | B | Rippling, 6 November 2026, names the first three. Reprompt's Y Combinator profile says "Mapping companies like Radar and TomTom use our agents". The customers themselves were not read. | [1, 3] |
| Scale of the agents | "over 150M AI agents" run; "over 4 trillion LLM tokens" processed | B | Rippling, 6 November 2026, for the agents. Reprompt's blog, 29 May 2026, for the tokens. | [1, 7] |
| Business database, November 2026 | "200M+ businesses", "+220M rows", 50 enrichment attributes | C | Reprompt's home page, read 8 November 2026. | [2] |
| Published price, July 2026 | US$5 per 1,000 requests for the Search and Ask APIs | C | Reprompt's blog post of 27 July 2026 by Rob Balian. | [5] |
| Own benchmark, July 2026 | 36 of 39 closed places found (92%) | C | Reprompt's test on human-checked places. It reports 25 of 39 for OpenAI Deep Research (o3) and for Gemini 3.5 Flash, 23 for GPT-5.5 and 17 for Claude Opus 4.8. Not checked by anyone else. | [8] |
| Outcome, 6 November 2026 | team joins Rippling; Reprompt's site says it was acquired | B | Rippling: "the team behind Reprompt AI is joining Rippling". Reprompt's home page: "Reprompt has been acquired by Rippling." No price was published. Outcome exit under casebook rule 10, because a source says the company was acquired. | [1, 2] |
| Time from first launch to the sale | 32 months | calc | Calculated: 15 March 2024 to 6 November 2026. | [1, 4] |

How to read this table: grade B is Rippling's own announcement as the buyer; grade C is Reprompt's own pages and its Y Combinator profile. A "calc" value was computed on this page. No revenue figure exists in any source we read.

## When did Reprompt AI launch, and how did it grow? Timeline

From a tool that tests AI apps to place-data agents bought by Rippling, in about two and a half years. Each entry carries its grade.

- **Late 2023** (grade B [1]): Lukas Martinelli, earlier a general manager on Mapbox's executive team, and Rob Balian, earlier head of growth at Robinhood, found Reprompt.
- **15 March 2024** (grade C [4]): Launch on Y Combinator during the Winter 2024 batch: a tool that lets AI teams trace, test and fix their apps without code.
- **5 January 2025** (grade C [6]): Partnership with Foursquare: Reprompt's agent works inside the FSQ Places Engine and accepts a fact only when three separate sources agree.
- **27 July 2026** (grade C [5]): Search and Ask APIs for other companies' agents. Reprompt's blog lists both at US$5 per 1,000 requests.
- **6 November 2026** (grade B [1, 2]): Rippling announces that the Reprompt team joins it to work on Rippling's AI agents. Reprompt's home page says the company has been acquired by Rippling.

## Why did Reprompt AI work?

### A narrow kind of research

General chat agents answer any question. Reprompt answered a few questions about places, again and again, for millions of rows: is it open, what are its hours, where is the entrance, who owns it. Its own July 2026 test claims it found more closed places than the general agents of OpenAI, Google and Anthropic.

### Several sources for each fact

The agents read search engines, business directories, official websites, social media and government filings, and in the Foursquare work a fact needed three separate sources. Reprompt measured fill rate, precision and stability, which is how often a repeated run gives the same answer.

### Expensive models only where needed

Reprompt's May 2026 post says it first runs a task at full depth, then finds the steps that a faster, cheaper model can do, such as summarizing one web page for the main agent.

## How it was built

These details come from Reprompt's own blog and pages. They are grade C.

- A planner agent resolves which place is meant, then chooses research branches: search and browse, routing, aerial and street-level imagery, property data and partner databases.
- The benchmark post says the agent harness has "over 40 tools at its disposal".
- Custom source ranking by country and a custom address formatter, because the models handle addresses poorly.
- Requests go into a queue and workers process them on their own, so millions of POI enrichments run without blocking.
- Customers reach it through an API, an AI table called the Reprompt AI Workbook, or data files from Reprompt's own business database.

Sources: [7, 8, 10, 2]

## Who tried this before Reprompt AI, and why did they fail?

Three companies in our database of failed startups tried to build a business on data about places before large language models could read the web for them. The values below are from that database and are not independently verified.

The failed startups come from our database of dead startups, retold here in our own words; Its entries are AI-assisted summaries of public sources, so they carry no evidence grade.

### Gowalla (2007 to 2012; funding US$10.5M; cause: Competition)

- Promise: Check in at coffee shops and other places on your phone and collect badges.
- Why it died: Lost to Foursquare's partnerships; Facebook then took the team, not the product.
- Record: 

### Desti (2012 to 2014; funding US$2M; cause: Unit Economics)

- Promise: An AI travel guide that read the web and suggested hotels and restaurants.
- Why it died: Users liked the suggestions, then booked elsewhere, so the technology never earned enough.
- Record: 

### Qraved (2013 to 2021; funding US$15M; cause: Ran Out of Cash)

- Promise: Restaurant discovery and booking for Indonesia, with an editorial team.
- Why it died: Google Maps took over discovery and super-apps took bookings; COVID closed dining rooms.
- Record: 

## What changed

The three predecessors built consumer apps around information about places, and the database records that Foursquare, Google Maps or other booking sites took their users. Reprompt sold to companies that already had the lists and needed them correct, and its agents did the reading that editors, raters and users had done before.

Gowalla's end matches Reprompt's: a larger company took the team. The difference is that Reprompt's site describes a sale of the company, and the buyer, Rippling, said it wanted the founders to build its own agents, not to run a place-data business.

## Can one person copy Reprompt AI now?

**Verdict: Hard for one person; the method is open.**

Reprompt needed founders who had run search and data-labeling teams, a large database and enterprise sales to Visa-sized customers, as Rippling and Reprompt's own pages describe. The method a small team can reuse is one narrow research question, asked by an agent across many web sources, measured against a human-checked sample.

1. Choose one fact that companies must check again and again in a large list, such as whether a business is still open.
2. Build an agent that needs agreement from several sources before it answers.
3. Measure fill rate, precision and stability on a sample that people have checked.
4. Sell it as an API priced per request, and run the cheap steps on smaller models.

**What is still open.** Other lists that go out of date: supplier and vendor data, store hours for delivery apps, licence and registry checks in countries with poor public data.

## What could go wrong for Reprompt AI?

- **What is not verified.** No revenue figure and no sale price were found. We looked for a payment-processor profile, a founder interview with a dated figure, a funding announcement and a buyer filing; Rippling's post gives no price. Rippling writes that the team is joining, while Reprompt's own home page says Reprompt has been acquired; this page uses outcome exit under casebook rule 10. The team size of 3 is the Y Combinator profile only. The customers are named by Rippling and by Reprompt, not by the customers, and the home page shows the logos of Data Axle and Matador, and the about page's "Trusted by Industry Leaders" row adds Y Combinator, whose batch Reprompt joined, Robinhood and Mapbox, where the co-founders worked before, and Retool, Instacart, AWS, Rebel Fund, TomTom, Radar, Kalibrate, HGreg and People Data Labs; none has a case study. A reader could look for a Rippling filing or a statement by a named customer.
- **Competition.** Ballerine, another story on this site, sells AI agents for merchant risk checks to payment companies, which overlaps with Reprompt's fraud and KYB product. Reprompt's own July 2026 benchmark compares it with the general research agents of OpenAI, Google and Anthropic, which also search the web.
- **Dependence on model providers.** Reprompt's May 2026 post says it had processed over 4 trillion model tokens, that the price of hosted small models was rising as providers retire older ones, and that it was testing self-hosted and open-source models. The sources we read do not name the providers it used.
- **Data about business owners.** The products research the businesses in public web pages, social media and filings, and two pages offer questions such as "who the owner is" and "Does owner own short-term-rental properties?". For a small business the owner is often one person. The sources we read do not say how such personal details are handled.

## Sources

1. Rippling blog, 6 November 2026, by Albert Strasheim: Reprompt AI is joining Rippling to build the future of agents — https://www.rippling.com/blog/reprompt-ai-joins-rippling (grade B). Read in full. The buyer's own announcement. Source for the founding date, the founders, the customers, the agent count and the team joining.
2. Reprompt home page, read 8 November 2026 — https://repromptai.com/ (grade C). Read in full. Source for the acquisition banner, the products, the database counts, the logos and the owner question.
3. Y Combinator company profile: Reprompt, AI Agents for Location — https://www.ycombinator.com/companies/reprompt (grade C). Read the profile data in the page: description, year founded, team size, location and founders.
4. Y Combinator launch post, 15 March 2024: Reprompt AI, Helping AI apps get to production quality 10x faster — https://www.ycombinator.com/launches/Kf6-reprompt-ai-helping-ai-apps-get-to-production-quality-10x-faster (grade C). Read in full. Source for the first product and the founders' earlier work.
5. Reprompt blog, 27 July 2026: Introducing Search and Ask APIs — https://repromptai.com/blog/search-and-ask-api (grade C). Read in full. Source for the only published prices.
6. Reprompt blog, 5 January 2025: Announcing Our Partnership with Foursquare — https://repromptai.com/blog/announcing-foursquare-partnership (grade C). Read in full. Source for the Foursquare agent and the three-source rule.
7. Reprompt blog, 29 May 2026: Building a location deep research agent — https://repromptai.com/blog/deep-research-agent-architecture (grade C). Read in full. Source for the architecture, the token count and the model costs.
8. Reprompt blog, 8 July 2026: Reprompt outperforms OpenAI, Claude, Gemini on hard places research tasks — https://repromptai.com/blog/deep-research-places-benchmark (grade C). Read the method and the closure results. The company's own test.
9. Reprompt product page: Fraud & KYB, read 8 November 2026 — https://repromptai.com/products/fraud-kyb (grade C). Read in full. Source for the KYB attributes and the owner question.
10. Reprompt blog, 26 January 2026: Web agents on 200M+ POI data — https://repromptai.com/blog/web-agents-200m-poi-data (grade C). Read in full. Source for the agents' web sources, the metrics and the queue of workers.
11. Internet Archive availability record for repromptai.com, earliest copy — http://archive.org/wayback/available?url=repromptai.com&timestamp=20200101 (grade C). Read the record: the closest copy to 1 January 2020 is from 23 March 2024.
12. Reprompt about page, read 8 November 2026 — https://repromptai.com/about (grade C). Read in full. Source for the CEO and CTO roles.
13. Database of failed startups, local export — /method/#dead (grade C). Local export. AI-written summaries, not independently verified. Source for the failed predecessors.

Contact: aiolum@livyden.com. Run Reprompt AI? Send corrections or a verified figure to aiolum@livyden.com (subject: Correction: Reprompt AI).