# CRHQ

Aiolum story: https://aiolum.com/story/crhq/ · added 2026-11-05 · updated 2026-11-06

**In short.** CRHQ (Hosted agent harness: AI coding agents on a dedicated server for each customer, Croatia): 619K agent messages across its servers in one month, July 2026. Evidence grade E: no revenue figure was checked. Team: 1 named founder, director of the Zagreb company that runs it; headcount not published. Launched March 2026; still running.

Hosted agent harness: AI coding agents on a dedicated server for each customer · Croatia · AI-native · solo founder · still running · group: Developer tools and coding agents

- Outcome: 619K agent messages across its servers in one month, July 2026 (evidence grade of the headline number: E; no revenue figure)
- Launch: March 2026; months from launch to the outcome: 4
- Team: 1 named founder, director of the Zagreb company that runs it; headcount not published
- Outside funding: None found
- Main factor: Founder runs his own businesses on it
- Meters, 1 to 4: evidence strength 1, copyable pattern 2, capital needed 1
- Tags: AI-native, Solo founder, Bootstrapped, Running, B2B, Developer tools

Links:

- Visit crhq.ai: https://crhq.ai/

**The product.** A rented server with an agent system already installed. The customer signs in with the AI subscriptions they already pay for (Claude Code, OpenAI Codex, Cursor, Z.ai GLM), then creates named agents with instructions, skills, memory and scheduled jobs. The agents message each other, delegate tasks, use a browser, and can build, test, deploy and monitor software on the same server. Partners can resell the servers under their own brand.

**Why it worked.** Coding agents such as Claude Code became able to run long tasks in 2026, but they run on one person's laptop and stop when the laptop closes. CRHQ puts them on an always-on server with a database, a scheduler and a shared skill library, and handles the server work. The founder runs his own SaaS products on two of these servers, and the company publishes what those agents did in a week.

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

CRHQ is an agent harness, the software around an AI model that gives it tools, files, memory and a schedule so that it can work on long tasks without a person typing each step. Each customer gets a VPS (virtual private server, a rented computer in a data center) with the harness installed. The agents run on the customer's own AI subscriptions, so CRHQ does not pay for model usage.

It is run by Zero Point Studio d.o.o. in Zagreb, Croatia, whose director is Andrej Šimunaj. The first Internet Archive copy of the site, from 11 March 2026, already lists a price, and the pricing page in November 2026 lists US$299 a month per server. The company blog reported that its servers handled 619,189 messages in July 2026, and that most of them were written by agents to other agents.

No revenue figure was found in any source that this page could download, so the evidence grade is E. What is checked is that the product is live, sells paid plans, has named customers quoted on its home page, and publishes its own usage statistics.

## Pricing

CRHQ prices on its pricing page, downloaded 5 November 2026 (grade C)

| Plan | Per month, billed yearly (US$) | Per month, billed monthly (US$) | Limit |
|---|---|---|---|
| Trial | 19 for 7 days | 19 for 7 days | Standard server, one trial per organization |
| Standard | 239 | 299 | 4 CPU / 8 GB RAM server, unlimited agents, skills and jobs |
| Pro |  | 399 | 8 CPU / 16 GB RAM server |
| Enterprise |  | 499 in Europe, 599 in the US | 16 CPU / 32 GB RAM server |

The pricing page lists one price per server, billed monthly in advance, with a prorated refund for unused days. Paying yearly saves 20%; the page states the yearly figure only for the Standard plan (US$239 a month, US$2,868 a year). A first server can start with a 7-day trial for US$19, which is not refundable. AI subscriptions are billed separately by the AI providers.

## How much revenue does CRHQ make? The numbers, graded

A Croatian agent-hosting service, live with paid plans since March 2026, with usage numbers published by the company; no revenue figure checked

| Fact | Value | Grade | Note | Sources |
|---|---|---|---|---|
| Status | Live, selling paid servers | C | Home page and pricing page downloaded 5 November 2026. | [1, 2] |
| Earliest archived home page | 11 March 2026, with "$299/month per satellite" | C | Internet Archive copy. The about page says the product has been "In production since early 2026". Two customers on the home page are labelled "Customer since March 2026". | [10, 3, 1] |
| Company and founder | Zero Point Studio d.o.o., Zagreb; director Andrej Šimunaj | C | CRHQ's contact page and terms name the company, its address and its tax number (OIB 52438945902). The terms of TranscriptAPI.com, another product of the same company, name "Director: Andrej Šimunaj". A customer quote on the home page says "Andrej doesn't just hand you a tool, he builds it with you". | [4, 12, 9, 1] |
| Team | headcount not published | C | The company writes "we" and "Built by operators". Its week-of-work page describes the founder's two servers as "run by one person". No source gives a staff count or suggests more than 10 people. | [3, 6] |
| Prices, November 2026 | US$299 a month per server (Standard); US$399 (Pro); US$499 or US$599 (Enterprise) | C | Pricing page. 7-day trial for US$19. Yearly billing saves 20%. | [2] |
| Messages on all servers, July 2026 | 619,189 | C | Company blog, 2 August 2026. Also 66,889 agent sessions and 367.8 billion tokens. 55.3% of the messages that someone started were started by an agent, not a person. | [5] |
| Model mix, July 2026 | Claude 91.6% of model spend, OpenAI Codex 8.4% | C | Company blog. The cost is "API-equivalent token value", what the tokens would cost at list prices, not what anyone paid, because much of the fleet uses subscription sign-in. | [5] |
| Share of tokens that were cache reads, July 2026 | 96.3% | C | Company blog. A cache read is a part of the conversation that the provider has stored and bills at about a tenth of the normal input price. | [5] |
| Founder's own servers, 30 July to 6 August 2026 | 45 named agents, 195 sessions, 60 scheduled jobs | C | Company page "One week of agents at work", counted from the two servers' databases. | [6] |
| Hosting | Hetzner Cloud in Germany and Finland | C | Security page. Each customer has a separate server with its own PostgreSQL database. | [8] |
| Outside funding | none found | C | No downloaded source mentions investors. | [3, 4] |
| Time from the first archived page to the July 2026 figure | 4 months | calc | Calculated: March 2026 to July 2026. | [10, 5] |

How to read this table: a grade C value is stated by the company on its own site or blog, or read from an Internet Archive copy of that site. A "calc" value was computed on this page. No value comes from a payment processor, a filing or a journalist.

## When did CRHQ launch, and how did it grow? Timeline

From the first archived home page in March 2026 to monthly usage reports, read from the company's site, its blog and the Internet Archive.

- **11 March 2026** (grade C [10]): First Internet Archive copy of crhq.ai: "One human. A team of AI agents.", with Research, Developer, Content, SEO and BI agents and a price per "satellite" (server).
- **March 2026** (grade C [1]): The earliest paying customers quoted on today's home page start, according to their labels "Customer since March 2026".
- **10 April 2026** (grade C [8]): The security page is published: one dedicated Hetzner server per customer, database bound to localhost, encrypted credentials.
- **July 2026** (grade C [5]): 619,189 messages across all servers; daily volume roughly doubles within the month; agent-to-agent messages pass human ones for the first time.
- **30 July to 6 August 2026** (grade C [6]): The founder's two servers run 45 agents for his own products while he works a few hours a day on vacation; the company publishes the log.
- **November 2026** (grade C [2, 1]): Pricing page lists four server sizes and a US$19 trial; a white-label program lets partners resell servers under their own brand.

## Why did CRHQ work?

### Founder runs his own businesses on it

The company's week-of-work page says the founder's own servers run agents that manage marketing, billing checks and support for several of his SaaS products, build new ones, and maintain CRHQ itself. The product is sold as the tool he already depends on.

### Bring your own subscription

Customers connect the Claude, ChatGPT, Cursor or Z.ai plans they already pay for. CRHQ charges for the server and the harness, so it carries no model cost, and the blog lists what the July usage would have cost at list prices only to show its size.

### One server per customer

Each customer gets a separate server with its own database, files and browser. The pricing page lists one price per server, and the white-label program lets agencies give each client a server under the agency's brand.

### Published usage numbers

Instead of revenue, the company publishes fleet statistics every month and a day-by-day log of what agents did, which is its main public evidence of use.

## How it was built

These details come from the company's own pages. They are grade C.

- Each customer server is a Hetzner Cloud VPS in Germany or Finland with its own PostgreSQL database, a firewall, Fail2ban and key-only SSH.
- The agents run on the customer's own Claude Code, OpenAI Codex, Cursor or Z.ai GLM access; the customer picks a model for each session.
- Agents keep instructions, skills and memory as files on the server, run scheduled jobs, use a browser, and can message and delegate to each other.
- Canvas lets a customer replace the interface with a branded one, built by a Theme Builder agent; an API lets a customer's own app call configured agents.
- Billing runs through Stripe. Updates are tested on a canary server before they go to the whole fleet, according to the week-of-work page.

Sources: [8, 2, 1, 6]

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

Three companies in our database of failed startups sold a team that does routine work for a busy person or company, before AI agents could do the work themselves. 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.

### Zirtual (2011 to 2015; funding US$5.5M; cause: Ran Out of Cash)

- Promise: Remote human assistants on a subscription, for email, scheduling and social media.
- Why it died: It employed many assistants without enough cash for payroll and ran out of money during expansion.
- Record: 

### Niles (2017 to 2020; funding US$2.8M; cause: Competition)

- Promise: A bot in Slack that answered staff questions from the company's knowledge.
- Why it died: Slack added its own features and cheaper rivals followed.
- Record: 

### Operator (2015 to 2020; funding US$30M; cause: Unit Economics)

- Promise: A personal shopper in an app, mixing human staff and software.
- Why it died: The human work cost too much per request to compete with Amazon.
- Record: 

## What changed

Zirtual and Operator sold work done by people, so every new customer needed more paid staff, and the cost per task did not fall. Niles used software, but the language technology of 2017 could only answer questions, not carry out tasks. CRHQ sells a server on which AI agents do the tasks, and the customer pays the AI provider directly, so CRHQ's own cost per customer is mainly the server.

The risk has moved: CRHQ depends on AI providers' subscription plans and their rules for third-party use, which it does not control.

## Can one person copy CRHQ now?

**Verdict: Partly: the hosting is copyable, the provider rules are the hard part.**

Putting coding agents on an always-on server with a scheduler and memory is a known technique, and open-source harnesses exist. What a copy would need is the same careful reading of each AI provider's rules on third-party use of subscriptions, and a founder who runs real businesses on it to show what it does.

1. Run your own products on the system first, and publish logs of what it did.
2. Give each customer a separate server, so their data and credentials stay apart.
3. Let customers sign in with their own AI accounts, so model cost is not yours.
4. Offer a white-label version to agencies that want to sell agents to their clients.

**What is still open.** Hosted agent servers for one industry (accounting firms, dental clinics, local agencies) with skills prepared for that industry's routine work.

## What could go wrong for CRHQ?

- **What is not verified.** No revenue figure was found in a source this page could download. The company site, its blog, its about and pricing pages and an Internet Archive copy give prices and usage statistics but no revenue or customer count. A founder profile with a monthly revenue and margin figure exists on FounderBase, but that site's robots.txt disallows Claude's crawler, so it was not read and is not used. No TrustMRR profile was found (trustmrr.com/startup/crhq.md answered 404). A reader could look for a payment-processor profile or an interview in which the founder states revenue with a date.
- **Account rotation and AI provider rules.** The home page says: "Smart auto-rotation moves work to an available connected account when five-hour or weekly limits are reached", for customers who connect several Claude, Codex or Z.ai accounts, and adds "It rotates accounts within a provider; it does not remove provider limits. Availability and provider terms apply." Anthropic's Claude Code legal page (downloaded 5 November 2026) allows a user to sign in to the unmodified Claude Code binary with their own subscription where a platform hosts Claude Code, and says advertised usage limits "assume ordinary, individual usage". Whether account rotation fits each provider's terms is not settled by any source here. Claude was 91.6% of model spend in July 2026, so a change in Anthropic's rules would affect most use.
- **Outreach and social-posting features.** Use-case templates include "Outreach campaigns and inbox work: agents draft, send and follow up on replies the way you set them up", LinkedIn engagement with messages "drafted in your voice, using your saved login", and social posts "planned and scheduled across your platforms". The founder's own marketing agent runs "a backlink outreach queue firing on a 20-minute cadence within pre-approved limits". No bulk-sending infrastructure (sender or inbox rotation) was found, and the terms forbid users to "Send spam, phishing messages, or unsolicited communications".
- **The owner's other products.** The same company runs TranscriptAPI.com, whose terms name Zero Point Studio d.o.o. CRHQ's week-of-work page also lists the founder's products YouTubeToTranscript, Recapio, ZillAPI and StayingAPI, and a client project with "scraping infrastructure with anti-detection browsers". This page judges only CRHQ; CRHQ's own pages do not sell those products.
- **Usage figures are the company's own.** The July 2026 message and token counts come from the company's blog. It also says the busiest single server made 28.9% of token value and the top ten made 77%, so use is concentrated in a few customers or in the founder's own servers.

## Sources

1. CRHQ home page, downloaded 5 November 2026 — https://crhq.ai/ (grade C). Read in full. Source for the product, the supported AI subscriptions, the auto-rotation text and the customer quotes.
2. CRHQ pricing page, downloaded 5 November 2026 — https://crhq.ai/pricing (grade C). Read in full, including the questions and answers. Source for the prices, the trial and Stripe billing.
3. CRHQ about page, downloaded 5 November 2026 — https://crhq.ai/about (grade C). Read in full. Source for "In production since early 2026" and the company's account of its origin.
4. CRHQ contact page, downloaded 5 November 2026 — https://crhq.ai/contact (grade C). Read in full. Source for the company name, Zagreb address and tax number.
5. CRHQ blog, 2 August 2026: CRHQ in July 2026: agents now talk to each other more than we talk to them — https://crhq.ai/blog/crhq-in-july-2026 (grade C). Read in full. Source for the July message, session, token, model and cache figures.
6. CRHQ: One week of agents at work (30 July to 6 August 2026) — https://crhq.ai/agents-at-work (grade C). Read in full. Source for the founder's own servers, his other products and the outreach agent.
7. CRHQ use cases page, downloaded 5 November 2026 — https://crhq.ai/use-cases (grade C). Read the use-case list, including the outreach, LinkedIn and social media templates.
8. CRHQ security page, last updated 10 April 2026 — https://crhq.ai/security (grade C). Read in full. Source for Hetzner hosting, the server hardening and the database setup.
9. TranscriptAPI.com Terms and Conditions, downloaded 5 November 2026 — https://transcriptapi.com/terms (grade C). Read the company section. Names Zero Point Studio d.o.o., its Zagreb address and "Director: Andrej Šimunaj".
10. Internet Archive copy of crhq.ai, 11 March 2026 — https://web.archive.org/web/20260311111209/https://crhq.ai/ (grade C). Read in full. The earliest capture in the archive's index; source for the March 2026 product and price.
11. Claude Code docs: Legal and compliance, downloaded 5 November 2026 — https://code.claude.com/docs/en/legal-and-compliance (grade C). Read the sections on hosting Claude Code, authentication and usage limits. Context for the auto-rotation note; it does not mention CRHQ.
12. CRHQ terms of service, downloaded 5 November 2026 — https://crhq.ai/terms (grade C). Read in full. Source for the operating company and the ban on spam and unsolicited messages.
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 CRHQ? Send corrections or a verified figure to aiolum@livyden.com (subject: Correction: CRHQ).