# AgentKit (formerly ClaudeKit)

Aiolum story: https://aiolum.com/story/agentkit/ · added 2026-10-26 · updated 2026-10-27

**In short.** AgentKit (formerly ClaudeKit) (Paid kits of skills, workflows and subagents for AI coding agents, Vietnam): US$82.2K in verified revenue since launch, October 2026. Evidence grade A: disclosed by a company or a filing, or verified by a payment processor. Team: 1 named founder; headcount not published; the founder also leads TOP Group, a Ho Chi Minh City agency that lists 40+ members. Launched 7 October 2025; still running, revenue falling.

Paid kits of skills, workflows and subagents for AI coding agents · Vietnam · AI-native · solo founder · still running, revenue falling · group: Tools for using AI models

- Outcome: US$82.2K in verified revenue since launch, October 2026 (evidence grade of the headline number: A; headline in US dollars: 82195)
- Launch: 7 October 2025; months from launch to the outcome: 12
- Team: 1 named founder; headcount not published; the founder also leads TOP Group, a Ho Chi Minh City agency that lists 40+ members
- Outside funding: None disclosed
- Main factor: Timing
- Meters, 1 to 4: evidence strength 3, copyable pattern 2, capital needed 1
- Tags: AI-native, Solo founder, Bootstrapped, Running, B2B, Developer tools

**The product.** Ready-made sets of instructions and helper programs for AI coding agents such as Claude Code, Codex and Cursor. A buyer installs a kit into a software project, and the coding agent then follows prepared roles (planner, tester, code reviewer), slash commands and workflows from planning to release. Each kit costs US$99 once; a bundle of the engineering and marketing kits costs US$149, and a desktop app costs US$49 or US$99 a year.

**Why it worked.** It launched in October 2025, when Anthropic's Claude Code had made skills, slash commands and subagents a normal way to steer a coding agent, and when most developers were still writing their own. A founder who presents themself as a CTO with 18 years of experience sold a tested set as a one-time purchase with a money-back promise. Verified revenue reached US$16,040 in March 2026 and US$82,195 in total, but it fell to under US$2,000 a month by September 2026.

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; calc = computed on this page from graded facts.

## Summary

AgentKit sells kits for AI coding agents, programs such as Claude Code, Codex or Cursor that read a code base, write and run code and edit files on a developer's computer at the developer's request. A kit is a folder of plain-text instructions and small scripts that the agent loads: skills (instructions for one kind of task, loaded when that task appears), subagents (separate agent roles, for example a planner or a code reviewer, that the main agent can hand work to), slash commands such as /ak:plan and /ak:cook, and hooks, scripts that the agent runs before or after a tool call, for example to ask for approval before a force push. The home page lists "108+ Skills", "95+ Commands" and "45 Subagents". The product launched as ClaudeKit, "Built on Claude Code by Anthropic", and was renamed AgentKit in 2026 when it added other agents.

The founder is Duy Nguyen from Vietnam, who writes as Goon Nguyen and signs the sales page as "Founder & CTO, AgentKit 18+ years building systems at scale". TrustMRR, a service that reads revenue through the seller's own payment account, gives a founding date of 7 October 2025 and reads the revenue from Polar, the payment platform that sells the kits. It shows US$10,411 in November 2025, a best month of US$16,040 in March 2026, then US$4,529 in May, US$2,641 in August and US$1,819 in September. By 27 October 2026 the total was US$82,195. The kits are sold once, not by subscription: the profile shows a monthly recurring revenue (MRR, the monthly value of active subscriptions) of only US$27.77.

The revenue figures are grade A: TrustMRR reads them through an API key from Polar. Sales paid by Vietnamese bank transfer go through another processor, Sepay, and are not in that figure, so the real total may be higher. The team size, user counts and ratings are the founder's own statements, grade C. The page also shows earlier companies that sold tools to make programmers faster, and why they failed.

## Pricing

AgentKit prices in October 2026, from the company's home and product pages (grade C); monthly equivalents calculated on this page

| Plan | Per month, billed yearly (US$) | Per month, billed monthly (US$) | Limit |
|---|---|---|---|
| AgentKit Engineer | 99 one-time |  | 100+ skills, 30+ workflows and AI subagents for software engineering |
| AgentKit Marketing | 99 one-time |  | 12+ MCP integrations and workflows for research, sales and content |
| AgentKit Bundle | 149 one-time |  | Engineer and Marketing kits plus a 1-year, 3-device desktop app licence |
| AgentKit App, 1 device | 4.08 |  | US$49 a year; monthly figure calculated |
| AgentKit App, 3 devices | 8.25 |  | US$99 a year; monthly figure calculated |

The kits are one-time purchases with lifetime updates and a 14-day money-back guarantee (30 days in November 2025). The bundle includes a one-year, three-device licence for the desktop app. The buyer also needs a paid plan or API account with the AI provider of the coding agent; that cost is not part of these prices. In November 2025 the only product, ClaudeKit Engineer, cost US$99, shown as reduced from US$149.

## How much revenue does AgentKit (formerly ClaudeKit) make? The numbers, graded

US$82,195 in verified revenue in about one year from one-time kit sales, with a best month of US$16,040 in March 2026 and a fall of 89% by September; the money is read from the payment platform, the team size is not published

| Fact | Value | Grade | Note | Sources |
|---|---|---|---|---|
| Verified revenue, all time, to 27 October 2026 | US$82,195 | A | TrustMRR, read from the company's Polar account through an API key; revenue last synced 27 October 2026 at 02:22 UTC. Last 12 months: US$80,601. Sales by Vietnamese bank transfer through Sepay are not read by TrustMRR. | [1, 7] |
| Revenue by month | Nov 2025 US$10,411; Dec US$7,246; Jan 2026 US$5,060; Feb US$11,507; Mar US$16,040; Apr US$8,743; May US$4,529; Jun US$4,143; Jul US$4,854; Aug US$2,641; Sep US$1,819; Oct US$1,921 to 27 October UTC | A | TrustMRR monthly table. The public table shows only the last twelve months. | [1] |
| Revenue before November 2025 | US$3,281 | calc | Calculated: the all-time US$82,195 minus the US$78,914 that the twelve months in the table add up to. This is the revenue of the first weeks, from 7 October 2025. | [1] |
| Best month | US$16,040 in March 2026 | A | TrustMRR monthly table. | [1] |
| Fall from the best month | September 2026 was 88.7% below March 2026 | calc | Calculated: (16,040 - 1,819) / 16,040 = 88.7%. | [1] |
| Verified revenue, last 30 days | US$2,365, up 32.5% from the 30 days before | A | TrustMRR. Last 7 days: US$208. Last 3 months: US$6,826. Last 6 months: US$19,985. | [1] |
| Monthly recurring revenue | US$27.77 from 8 active subscriptions | A | TrustMRR snapshot. Almost all revenue is from one-time kit purchases; the subscriptions are probably the yearly desktop app licences, which the profile does not say. | [1] |
| Where the revenue figures come from | Polar, through an API key; merchant of record: no | A | TrustMRR's verification section. The terms of use name "Polar.sh for global credit card and PayPal payments" and "Sepay for Vietnam bank transfer payments (VND)". | [1, 8] |
| Kit sales implied by the revenue | about 550 to 830 | calc | Calculated: US$82,195 divided by US$149 (551) or by US$99 (830). The real number is unknown, because discounts, refunds, referral commissions and desktop licences change the result. | [1, 2] |
| Founding date | 7 October 2025 | C | TrustMRR profile, a field written by the founder. A copy of claudekit.cc saved on 1 November 2025 shows the product on sale at US$99. | [1, 4] |
| Users and community | "Estimated user count: 7,104"; "3,800+ builders on Discord"; "4.9/5 349 ratings" | C | TrustMRR founder fields and the home page, read 25 October 2026. The home page also says "12 customer reviews" and does not say where the 349 ratings were given. In November 2025 the site said "Join 500+ developers". | [1, 2, 4] |
| Founder | Duy Nguyen, who writes as Goon Nguyen | C | TrustMRR: "Founder: Duy /zuey/" with the handle goon_nguyen. The blog author page: "Goon Nguyen Founder of AgentKit". The GitHub profile of "mrgoonie (Duy /zuey/)" links agentkit.best and says "Just launched ClaudeKit.cc". | [1, 10, 11] |
| Team | 1 named founder; headcount not published | C | No source gives a number of people. The sales page is written in the first person: "I guarantee it—I've proven it in production." The founder's GitHub profile also says "Founder @ wearetopgroup.com", a creative agency; see the watch section. | [3, 11, 15] |
| Location | Vietnam | C | TrustMRR "Country: VN". The privacy policy: "AgentKit is based in Vietnam." The terms are governed by the laws of Vietnam, with courts in Ho Chi Minh City. | [1, 7, 8] |
| Outside funding | none disclosed | C | No source mentions investors. This is an absence of information, not a statement by the company. | [1, 2] |
| Referral program | 20% to 40% commission per sale | C | The product page: "Get paid 20% recurring commission for every friend who buys AgentKit", with "$19.80 per referral". The leaderboard page lists tiers from 20% up to 40% for referrers with US$15,000+ in referred revenue. | [3, 9] |
| Rename | ClaudeKit (ck) became AgentKit (ak) by August 2026 | C | Blog post of 4 August 2026: "ClaudeKit ( ck ) is now legacy, and AgentKit ( ak ) is the supported path going forward." The post does not give a reason for the new name. | [5] |
| Time from launch to the October 2026 figure | 12 months | calc | Calculated: 7 October 2025 to 27 October 2026 is 385 days. | [1] |

How to read this table: a grade A value was read by TrustMRR from the company's Polar account through an API key. A grade C value is what the founder or the company's own pages state. A "calc" value was computed on this page from the rows above it.

## When did AgentKit (formerly ClaudeKit) launch, and how did it grow? Timeline

From ClaudeKit, one US$99 kit for Claude Code, to AgentKit with three products for seven coding agents, in about twelve months. Each entry carries its grade.

- **7 October 2025** (grade C [1, 11]): Founding date of ClaudeKit on the TrustMRR profile. The founder's GitHub profile announces "Just launched ClaudeKit.cc".
- **October 2025** (grade A [1]): US$3,281 in verified revenue in the first weeks.
- **1 November 2025** (grade C [4]): claudekit.cc sells ClaudeKit Engineer, "12 specialized AI agents" built on Claude Code, for US$99 with a 30-day money-back guarantee and a 20% referral commission, and says 500+ developers use it.
- **November 2025** (grade A [1]): US$10,411 in verified revenue in the first full month.
- **January 2026** (grade A [1]): Revenue falls to US$5,060.
- **2 March 2026** (grade C [12]): A blog post by the founder answers the most frequent question, why pay when free kits exist, with "QUALITY > QUANTITY" and "54 skills. 80+ connections."
- **March 2026** (grade A [1]): Best month: US$16,040.
- **May to July 2026** (grade A [1, 6]): Revenue falls to between US$4,143 and US$4,854 a month. The changelog shows separate release lines for the Engineer kit and the Marketing kit.
- **4 August 2026** (grade C [5]): A blog post tells users to move from the old ck command to the new ak command: ClaudeKit is now AgentKit.
- **September 2026** (grade A [1]): US$1,819 in verified revenue, the lowest full month.
- **October 2026** (grade C [6, 2]): Version 3.0 betas of the AgentKit command-line tool; the home page lists support for Claude Code, Codex, Cursor, Antigravity, Pi, Oh My Pi and DeepSeek Harness.
- **27 October 2026** (grade A [1]): US$82,195 all time and US$2,365 in the last 30 days.

## Why did AgentKit (formerly ClaudeKit) work?

### Selling a tested set of instructions

A coding agent works better with a written plan, fixed roles and checks, but each developer had to discover these by trial and error. The kit sells that work: the November 2025 page called its author a "18+ Year CTO" who "burned millions of Claude tokens finding what works". The buyer gets plain-text files that they can read, change and keep.

### One payment, no subscription

A kit costs US$99 once, with updates included. This removes the objection of another monthly bill next to the AI provider's plan, and the money-back guarantee if the kit "doesn't save you 10+ hours" lowers the risk further. The cost is that revenue depends on new buyers every month, and the verified MRR is only US$27.77.

### Referrals from buyers

From the first month the site offered buyers 20% of each sale they refer, later up to 40%. The product page shows the arithmetic: "$19.80 per referral" and "$1,980 with 100 referrals". No source says what share of sales came from referrals, so this page leaves the channel unknown.

### From one agent to seven

ClaudeKit was built for Claude Code only. As other coding agents spread, the product was renamed AgentKit and adapted to seven of them, and a marketing kit and a desktop app were added. The changelog shows frequent releases, many of them automated.

## How it was built

These details come from the company's own pages, its changelog and its legal pages. They are grade C.

- The kits are files for the coding agent: skills, subagent definitions, slash commands and hook scripts, installed with the company's command-line tool, first called ck and now ak.
- The first version was built on Claude Code with Anthropic's Sonnet 4.5 model, according to the November 2025 page. The current kits also target Codex, Cursor, Antigravity, Pi, Oh My Pi and DeepSeek Harness.
- Polar takes card and PayPal payments, Sepay takes Vietnamese bank transfers, and the home page promises "Instant license-key access". Early buyers received access through invitations to a GitHub repository, which the privacy policy now calls legacy.
- The website is hosted on Vercel. Resend sends the transactional email. Google Analytics, Umami and Google Tag Manager measure visits.
- Releases are published by GitHub Actions; the changelog page lists each release with the author "github-actions[bot]".
- The site also sells a partner product, ReelCrew, a video toolkit for Claude Code built by another team.

Sources: [4, 6, 7, 8, 14]

## Who tried this before AgentKit (formerly ClaudeKit), and why did they fail?

Four companies in the Loot Drop database sold tools that promised to make programmers faster or their code better, before AI coding agents existed or as they arrived. The values below are from that database and are not independently verified.

The failed startups come from Loot Drop (https://www.loot-drop.io/), a free database of dead startups, retold here in our own words; Loot Drop describes its entries as AI-assisted summaries of public sources, so they carry no evidence grade.

### Kite DevTool (2014 to 2024; funding US$21M; cause: Competition)

- Promise: AI code completion inside the programmer's editor.
- Why it died: The database summary says GitHub Copilot, built on a much stronger OpenAI model, arrived in June 2021 and made Kite's older models obsolete, while Kite never found a way to make developers pay.
- Record: https://www.loot-drop.io/startup/2491-kite-devtool

### Fig DevTool (2020 to 2024; funding US$2.4M; cause: Unit Economics)

- Promise: Autocomplete and visual command builders for the terminal.
- Why it died: The database summary says users loved it and used it daily, but too few of them paid, because developers have a very low willingness to pay for terminal tools.
- Record: https://www.loot-drop.io/startup/2502-fig-devtool

### Sider (2012 to 2024; funding US$7M; cause: Competition)

- Promise: Automated code review and static analysis for development teams.
- Why it died: The database summary says the market split between free open-source tools and large enterprise platforms, and Sider was squeezed between them.
- Record: https://www.loot-drop.io/startup/2494-sider

### Light Table/Eve (2012 to 2015; funding US$3M; cause: Competition)

- Promise: An editor that shows code results in real time, then a new way for non-programmers to build apps.
- Why it died: The database summary says professional developers stayed with the editors they already used, and the pivot to Eve lacked focus and resources.
- Record: https://www.loot-drop.io/startup/1560-light-table-eve

## What changed

Kite, Fig and Sider each built a large piece of software, an AI model, a terminal layer or an analysis engine, and needed millions of dollars before they could sell it. Fig shows the other problem of developer tools: people who use a tool every day still do not pay for it. All three ended when a free or bundled alternative became good enough.

AgentKit builds almost no software of its own. The coding agent and the model come from Anthropic, OpenAI and others; the kit is instructions and scripts on top. That is why one founder could ship it within weeks and charge from the first day. The same fact is the risk: the instructions can be copied, free kits exist, as the founder's own blog post of March 2026 admits, and the agent makers add their own skills and workflows. Verified revenue fell 88.7% from March to September 2026.

## Can one person copy AgentKit (formerly ClaudeKit) now?

**Verdict: The method, yes. The timing, no.**

The opening was the months after Claude Code made skills and subagents common and before free collections were everywhere. Today buyers can compare paid kits with many free ones, and the verified revenue shows the effect. What can still be copied is the structure: package expert know-how for a new tool as files the buyer owns, sell it once with a refund promise, and pay buyers to recommend it.

1. Use the new tool every day until you have a set of instructions that clearly works better than the default.
2. Package it so that it installs with one command and can be read and changed by the buyer.
3. Sell it once with a short refund period, and connect a payment platform that can later prove the revenue.
4. Offer buyers a stated commission for each sale they refer, and keep the kit updated as the tool changes.

**What is still open.** Kits for one industry or one stack are still open: for example rules and checks for medical-device software, banking code or a single framework, where generic free kits do not know the requirements.

## What could go wrong for AgentKit (formerly ClaudeKit)?

- **Revenue fell by most of its peak.** Verified revenue was US$16,040 in March 2026 and US$1,819 in September 2026, 88.7% lower. The last 30 days were 32.5% above the 30 days before, at US$2,365. One-time sales need new buyers every month, and there is almost no recurring revenue.
- **Income wording around the referral program.** The product page heads its affiliate section "Turn AgentKit Into Passive Income" and shows "$19.80 per referral", "$198 with 10 referrals" and "$1,980 with 100 referrals", which is the stated 20% commission on a US$99 kit multiplied out. Aiolum accepts an affiliate program that states its commission. The same page has a "First Feature ROI" box that values 10 saved hours at an assumed US$100 an hour and shows "Net profit: $901"; this is a claim about saved working time, backed by a refund "If it doesn't save you 10+ hours", not a claim about money earned from the product. A blog post of 19 August 2026 on extra income says vibe coding does not "guarantee a passive income stream".
- **The founder's other company.** The founder's GitHub profile says "Founder @ wearetopgroup.com". That site describes T.O.P Group as "doanh nghiệp cung cấp giải pháp sáng tạo được thành lập năm 2014 và có trụ sở chính tại Thành phố Hồ Chí Minh" (translation: a creative-solutions business founded in 2014 with its head office in Ho Chi Minh City) and shows "40 + Thành Viên" (translation: 40+ members). No source says that its staff work on AgentKit, and the AgentKit pages do not mention it. The GitHub profile also lists other products of the founder: GoClaw.sh, TOSE.sh, Dewee.sh, AgentBrain.sh and AgentWiki.cc.
- **Built on other companies' agents.** The kits only work inside coding agents made by Anthropic, OpenAI and others, and the buyer pays those providers separately. The first name, ClaudeKit, used Anthropic's product name; the 2026 rename post does not say why the name changed. Another developer also published a project called claudekit, which the founder's blog mentions.
- **Unclear proof on the sales page.** The home page shows "4.9/5 349 ratings" next to "12 customer reviews" without saying where the ratings come from, and the user counts (7,104 on TrustMRR, 3,800+ on Discord) are the founder's own figures.
- **Part of the revenue is not verified.** TrustMRR reads only the Polar account. Vietnamese buyers can pay by bank transfer through Sepay, and those sales are not in the US$82,195. The partner product ReelCrew is also sold on the site; whether its sales pass through the same Polar account is not stated.

## Sources

1. TrustMRR profile: ClaudeKit (markdown version), revenue synced 27 October 2026 — https://trustmrr.com/startup/claudekit.md (grade A). Read the markdown profile, on 25 and 27 October 2026. Revenue and MRR are read from Polar through an API key (grade A). The description, founding date, user count and pricing are written by the founder (grade C).
2. AgentKit home page, read 25 October 2026 — https://agentkit.best/ (grade C). Read in full. Source for the products, prices, supported agents, counts of skills and the ratings line.
3. AgentKit Engineer product page, read 25 October 2026 — https://agentkit.best/engineer (grade C). Read in full. Source for the price box, the ROI box, the refund terms, the referral section and the founder's signature.
4. claudekit.cc home page, Wayback Machine copy of 1 November 2025 — http://web.archive.org/web/20251101043222/https://claudekit.cc/ (grade C). Read in full. Source for the first product, its price, the 30-day guarantee, the referral offer and the 500+ developers claim.
5. AgentKit blog, 4 August 2026: ClaudeKit AgentKit migration: How to safely upgrade to ak CLI — https://agentkit.best/blog/claude-code-guides/claudekit-agentkit-migration-how-to-safely-upgrade-to-ak-cli (grade C). Read in full. Source for the rename.
6. AgentKit changelog, read 25 October 2026 — https://agentkit.best/changelog (grade C). Read the release list. Source for the release lines and the automated releases.
7. AgentKit privacy policy, read 25 October 2026 — https://agentkit.best/privacy (grade C). Read in full. Source for the service providers and the location.
8. AgentKit terms of service, read 25 October 2026 — https://agentkit.best/terms (grade C). Read in full. Source for the payment processors and the governing law.
9. AgentKit referral leaderboard, read 25 October 2026 — https://agentkit.best/leaderboard (grade C). Read in full. Source for the commission tiers.
10. AgentKit blog author page: Goon Nguyen, read 25 October 2026 — https://agentkit.best/author/goon (grade C). Read the profile line and the list of posts.
11. GitHub profile of mrgoonie (Duy /zuey/), read 25 October 2026 — https://github.com/mrgoonie (grade C). Read the profile text. Source for the launch note, the founder's other products and the link to wearetopgroup.com.
12. AgentKit blog, 2 March 2026: What Makes ClaudeKit Special Compared to Open Source Kits? — https://agentkit.best/blog/uncategorized/what-makes-claudekit-special-compared-to-open-source-kits (grade C). Read in full. Source for the founder's answer to free competitors.
13. AgentKit blog, 19 August 2026: Vibe code extra income: Building profitable AI products fast — https://agentkit.best/blog/vibe-coding/vibe-code-extra-income-building-profitable-ai-products-fast (grade C). Read in full for the purpose check. Quoted in the watch list.
14. AgentKit partner page: ReelCrew, read 25 October 2026 — https://agentkit.best/reelcrew (grade C). Read in full. Source for the partner product.
15. T.O.P Group home page (Vietnamese), read 26 October 2026 — https://wearetopgroup.com/ (grade C). Read in full. Source for the founder's other company, its founding year and its member count. Quotes are in the original Vietnamese with a translation by this page.
16. Loot Drop database export, 1,749 failed startups — https://www.loot-drop.io/ (grade C). Local export. AI-written summaries, not independently verified. Source for the failed predecessors.