# "kander + lots of tokens": one hype-riding post pushed a multi-agent kanban board up 130 stars in two days · GitHub open-source project

> Source: https://stringers.io/en/github/gh_dualface__kander.html (Stringers / 什么值得做, generated 2026-11-20)
> Stringers tracks which small products suddenly took off each day, investigates why one by one, and distils playbooks from them. Numbers come from 15 databases we collect daily, each with source and date.
> For agents: these investigations and the underlying databases (verified revenue, App Store / Google Play charts, ad libraries, GitHub, Product Hunt…) are queryable live over MCP — tools feed (daily breakouts) / plays (playbooks) / hot / products / data. Setup: https://github.com/komako-workshop/stringers (the user signs in with Google at stringers.io/en/start.html to get a personal URL). Site index: https://stringers.io/llms.txt

> The author is a long-established open-source developer in the Chinese community. Three times he personally posted on V2EX and X, bringing three waves of stars. His latest post caught the heat around Magpie's fully automated dev workflow and gained 130 stars in two days. The production table on the README's first screen (90 days, 2,477 cards, 51% of problems caught by review as mandatory fixes) keeps these posts from reading as pure ads.

[GitHub open-source project](https://stringers.io/en/github/index.md) · written 11-12

dualface/kander · dualface

# "kander + lots of tokens": one hype-riding post pushed a multi-agent kanban board up 130 stars in two days

**What it is** A kanban board that runs in the terminal. You talk through requirements with Claude Code or Codex, the agent creates cards itself and writes code in isolated git worktrees, and every finished card must be reviewed by a different agent. Review issues have to be fixed with evidence before the card can close.

**Why it took off** The author is a veteran open-source developer in the Chinese dev community. He posted about it himself three times on V2EX and X, and each post brought a wave of stars. The latest one rode the buzz around Magpie's fully automated dev workflow and gained 130 stars in two days. The README's opening screen shows a production data table: 90 days, 2,477 cards, and 51% of review batches caught must-fix issues. That data keeps the posts from looking like pure ads.

**C** **How you could follow this** Don't build another kanban board in the same category. Copy the playbook instead: use your own tool on real work for a few months, put verifiable production numbers at the top of the README, and when a topic is hot in your circle, post "my tool can do this too" the next day, promising a tutorial to catch the next wave.

Grade C: the multi-agent kanban category already has a long list of projects with tens of thousands to a few hundred stars. Even with the author repeatedly promoting it himself, Kander has only 300 stars, so copying it will most likely get you buried. The two things worth learning are using production data as the README's opening content and riding a hot topic the next day. What's being passed around in small circles

Data (updated daily) · Repo created 2026-09-05 · 331 stars · 7 days +26

GitHub · stars

Data 2026-10-15 → 2026-11-20 (37 days, 36 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

Kander is a terminal kanban board: one person assigns development tasks to six or seven AI coding agents running in parallel, and when they finish, another agent reviews the work. It has been open source for 69 days and has barely passed 300 stars. Top projects in the same category range from a few thousand to over a hundred thousand stars, so it hasn't broken through. Still, every wave of new stars lines up with one of the author's own posts, and the latest wave came from a single post that added 130 in two days. Below: what it is, where the stars came from, why people trust it, and what you can take from it if you want to build an open-source project that gets stars.

### A post at 11 pm on Nov 11

On Nov 10 a long post went viral on Chinese X. @xulijian1994, the author of openai-translator, described how he used Claude Code with three `/loop` commands to build a "fully automated dev workflow" for his own project Magpie, which has 35k stars. It watches issues, PRs, and feedback on X by itself, fixes bugs, and ships releases. The post got 114k views and 869 likes.

The next day, Beijing time 23:05, Kander's author dualface posted a follow-up on the same topic:

> Lots of people say Magpie's automated development is amazing. Honestly, it's easy to get too. Only two things needed: kander + lots of tokens

The post got 36k views, 213 likes, 20 reposts, and 33 comments. The comments mostly said the same thing: "looking forward to the tutorial," "share it so we can learn," "I've been looking for software like this." The author replied: "Currently grinding out the tutorial."

Kander's stars gained only 3 in the week before, then 11-11 ... 51,11-12 ... 78 (daily count from Trendshift). 11-13 On GitHub it's 309 stars and 30 forks. The 10 new forks from those two days all came from old Chinese developer accounts, registered for years, with followers and their own repos. They don't look bought.

### What it is

A single-file terminal program written in Go. On macOS you install it with one Homebrew command; on Linux and Windows you download the binary directly. It works the opposite way from tools like Vibe Kanban, where people drag cards around a board:

- **You only talk through requirements.** You clarify the requirements in a Claude Code or Codex session, and an agent with the Kander rules installed creates cards, starts work, and schedules progress on its own. The board is only for watching progress. In the author's words: "The board is just for convenience. The core is the rules and the tooling."

- **Each card gets its own isolated workspace.** Every card opens a separate git worktree and runs in its own tmux or Herdr window. Before scheduling, it checks whether several cards would touch the same files, and cards that would collide don't run at the same time.

- **The writer and the reviewer can't be the same agent.** When a card is done, it goes to a different agent for review, and that can cross vendors, for example Claude writes and Grok or Codex reviews. You can also use a cheaper model to review an expensive one's work. Each must-fix issue from the review has to be handled one by one: fixed, rejected with evidence, or deferred. Only after all of them are resolved can the card pass the completion gate.

It supports six agents: Claude Code, Codex, Grok, Cursor, Pi, and Devin. The community has since added Kimi CLI and droid.

It's a side product of the author's other project. He's building QuickTUI, an app for remotely controlling AI agents on your computer from your phone. It's free for single-machine use, and the cloud relay is a subscription. Kander is a tool he built for himself while developing QuickTUI. The task cards in the README screenshots are QuickTUI's, and QuickTUI is linked at the end of the README and on the website quicktui.ai. QuickTUI itself is small: 2 ratings on the US App Store and 7 on the China store (11-13).

### The stars came in waves

| Timeline | Stars gained | What happened those days |
|---|---|---|
| Repo created 09-05, first release 09-09 | By mid-September, about 38 | Posted "Built a multi-agent orchestration system that runs in the terminal" in V2EX's "Sharing Creations" section, asking for stars at the end; 12 hours later it had 458 views and 2 replies |
| 10-14, 10-15 | +31, +29 | 10-13 The author posted "From Aug 1 to now, I've completed nearly 2000 dev tasks" on V2EX, which drew 103 replies; 10-14 Dan Kornas, an overseas AI tool repost account, posted an English introduction on X |
| 10-16 to 11-10 | 0 to 5 a day | 10-29 released v1.0.0, and a group of overseas users filed detailed issues; 11-10 it was added to the awesome-agent-orchestrators list |
| 11-11, 11-12 | +51, +78 | The post riding Magpie's wave, above |

The October wave deserves more detail. In that V2EX post the author said he hadn't written a single line of code himself since March, QuickTUI had 20k commits and over 1.2M lines of code, and he casually mentioned a project for his company whose "user count is close to 9 million." The replies were almost all skeptical: "Even if it's sharing, it still smells like an ad," "The website is so ugly, 80% chance it's an ad," and a string of comments challenging the "9 million users" claim. The author answered bluntly: "Ha, yes it's an ad, but it's sharing too," and "I now think marketing matters most. The technical barrier is practically 0 now." The post drew 103 replies, mostly criticism, and gained 31 stars the next day.

Beyond these waves, the author regularly posts his usage stats and output on X: "Completed 1700+ tasks since August," "1500 task milestones," "running 40 Claude Code sessions at once." He says the WeChat group he runs has passed 200 people, which I can't verify.

Hacker News, Product Hunt, and the media have never covered it, and it has never appeared on GitHub Trending. Nearly all the stars came from the author's own posts in the Chinese community.

### What holds these posts up: a production data table

If you just say "I directed AI to finish 2000 tasks by myself," anyone would read it as bragging, which is exactly why the V2EX post got roasted. Kander's README opening shows a table, with the numbers as of 10-29:

| One person + 6 agents | 90 days | 5 projects |
|---|---|---|
| 2,477 task cards completed | 2,865 independent reviews | 51% of review batches caught must-fix issues |

Below is also a stats doc that spells out how each number is counted (read-only, from the `kanban/done/` folders of five projects), broken down by project, by agent, and by model. For example, cards written by Claude Opus 5.5 were sent back an average of 0.78 times each; Codex's gpt-6-sol averaged 0.36; and the median card took 57 minutes. There's also a long-form production retrospective and a PDF slide deck.

Numbers like these, from things that were actually run, are persuasive to people frustrated by agent code that passes all its tests yet still has bugs. 10-12 A Japanese author wrote a long analysis on note titled "49% of review batches caught Must-fix," and the README has Chinese, English, and Japanese versions, so overseas readers can share it straight away.

So it's worth separating two things. The direct source of the stars is the author's posts, not the product itself. But each post is backed by a data table you can check, which keeps it from being dismissed as a pure ad.

### Who the author is

dualface's V2EX profile reads "author of quick-cocos2d-x, QeePHP, FleaPHP, QuickTUI," a veteran open-source developer in Chinese game development and PHP circles. His GitHub account was registered 2009 years ago and has 754 followers, and his older project gbc-core has 363 stars. In 2026 he shifted to "one person + a team of agents" to build apps and games, and during the same period he opened a series of small tools: Ullage, ste-zh, ask-user-question, claude-switch, codex-loop, and others, mostly with single digits to a couple dozen stars. Kander is the only one that passed 100.

In other words, an author with some established reputation, an audience in the Chinese community, and a habit of posting his usage numbers personally pushed each time, one post at a time, and got Kander to 300 stars. Someone without that name probably wouldn't reach even this number (my inference).

### Who's competing for this spot

The combination of "worktree isolation + kanban + parallel multi-agent" was already being built by others as early as 2025, and the space is now crowded (GitHub, 11-13):

| Project | star | How Kander differs |
|---|---|---|
| stablyai/orca | 124k | A GUI for running a team of agents |
| herdrdev/herdr | 51k | A terminal container for agents; Kander uses it as a backend |
| BloopAI/vibe-kanban | 29k | A web kanban where people drag cards and review diffs |
| getpaseo/paseo | 24k | Desktop and mobile orchestration, a competitor to QuickTUI |
| smtg-ai/claude-squad | 8.8k | The closest terminal multi-agent manager to Kander, but without review |
| cline/kanban | 1.4k | A local web kanban from Cline |
| emanuele-em/kanzo | 1.2k, +145 over 7 days | Created 10-25, grew even faster than Kander in the same week |
| kdlbs/kandev | 1.1k, +60 over 7 days | Kanban plus approvals plus a web IDE |

Below these are a dozen or so similar small libraries with a few hundred stars each. Claude Code's subagents, Codex App's parallel worktrees, and Cursor's parallel agents are all making "parallelism within a single vendor's agent" a default feature (launch dates for each not verified). Kander's remaining differences come down to two points: review must be done by a different agent, and issues must be resolved with evidence before a card can close; and it supports cross-vendor peer review. A commercial desktop app, Pane (runpane.com), has already published "Best Kander Alternatives in 2026" to capture its search traffic.

### If it were you

**C: worth learning from, not worth copying.** The multi-agent kanban category is already crowded, and even Kander, pushed repeatedly by a veteran author himself, has only 300 stars. Building another tool in the same category, even in a different language or ecosystem, will most likely get lost among those dozen small libraries with a few hundred stars each. What's worth taking are its two moves: using your own production data as the README's opening content, and riding a hot topic the very next day.

**Who it suits:** developers who have an agent workflow they've really used for months, have accumulated real numbers, and have some presence on Chinese X, V2EX, or an English community.

**Who it doesn't suit:** anyone wanting to build yet another "worktree + kanban + multi-agent" tool from scratch, since this space has no shortage of tools; and anyone without any posting channel, since Kander's stars came almost entirely from the author's own posts.

**How to do it:** don't copy the topic, copy these steps. ① Use your own tool on real work for a few months and collect the numbers. Put "how much was done, how many issues were caught" at the top of the README, with the counting method and a reproducible way to verify it, and break it down by model or scenario in a table, since that kind of table is the easiest for others to write articles about. ② Watch the hot posts in your circle. When someone is flooding the feed with a certain "fully automated" approach, post "my open-source tool can do this too" the next day, as Kander did with a one-day gap. ③ If people in the comments ask for a tutorial, promise to write one, and attach the next wave of traffic to the tutorial. ④ Have an English version of the README ready from the start (ideally Japanese too), so overseas repost accounts and writers can share it directly.

**How to get users:** Your own posts are the channel. Start with V2EX "share what you built" and X in the Chinese dev circle. Getting called out as an ad isn't necessarily bad: Kander's 103-reply post gained 31 stars the next day. But it only works if you have data to back up the claims. Tossing out "9KW users" off the cuff just burns trust.

**When was the timing:** The "fully automated dev / one-person army" wave peaked in November, and the tutorial the author mentioned hadn't come out yet. Riding this topic, the window is measured in weeks.

**What would prove this read wrong:** ① Once the author's tutorial is out and Kander passes 1000 stars within a week, that would show the "rules + cross-review" setup pulls people in on its own, not just because the author is pushing it; ② If an unknown author in the same small-library cohort (kanzo, kandev) also hits several thousand stars, the category isn't saturated; ③ Conversely, if the tutorial ships and stars stall at three or four hundred, then "production data + riding the hype" only sustains that level, and tools in this category aren't worth chasing.

[GitHub repo ↗](https://github.com/dualface/kander)

**Research notes: competitor table, acquisition evidence, full research report**

Competitors

| Name | Where | Launched | Scale | How it differs from this one |
|---|---|---|---|---|
| stablyai/orca | Open source | 2026-03 | 123.6k star | GUI that runs a team of agents; Kander is a TUI with a mandatory cross-agent review gate |
| herdrdev/herdr | Open source | 2026-03 | 50.7k star | Runs agents in the terminal; Kander uses it as a backend and handles orchestration and review on top |
| BloopAI/vibe-kanban | Open source | 2025-06 | 28.7k star | Web kanban where humans drag cards and review diffs; in Kander, agents create and schedule cards by the rules, and a second agent reviews |
| getpaseo/paseo | Open source | 2025-10 | 23.9k star | Desktop and mobile UI orchestration, also a competitor to QuickTUI; no evidence-based review gate |
| smtg-ai/claude-squad | Open source | 2025-03 | 8.8k star | The closest TUI multi-agent manager, but it only manages sessions, not review or completion gates |
| stravu/crystal(Nimbalyst) | Open source | 2025-06 | 3.1k star | Parallel worktree sessions with a diff panel; humans do the review |
| phodal/routa | Open source | 2026-02 | 1.8k star | A multi-agent collaboration platform built around specs and kanban, leaning toward web and teams |
| cline/kanban | Open source | 2026-03 | 1.4k star | A local web kanban from Cline that runs CLI agents; no independent review chain |
| kdlbs/kandev | Open source | 2026-01 | 1.1k stars, +60 over 7 days | Kanban with approvals, a web IDE, and PR opening; Kander is terminal-first and rule-driven |
| emanuele-em/kanzo | Open source | 2026-10-25 | 1.2k stars, +145 over 7 days | A lightweight kanban shared by humans and agents, which grew faster than Kander in the same period |
| Claude Code subagents / agent teams, Codex App worktree parallelism, Cursor parallel agents | Big-company features | 2025-10–2026 (unverified) | Installed by default | Parallelism within a single vendor's agent; Kander runs across vendors and does mutual review (Claude writes, Grok / Codex reviews) |

Where users come from

- **Verified** **Riding the hype (X Chinese dev circle)** : 11-10 @xulijian1994 talked about yetone's Magpie fully automated dev workflow, 114k views; 11-11 23:05 dualface posted "the magpie setup can be done with kander + lots of tokens", 36k views, 213 likes, 20 reposts; 11-11 +51, 11-12 +78

- **Verified** **V2EX posts (launch and the controversial thread)** : 09-09 launch post in "share what you built" asking for stars (458 views); 10-13 "Completed nearly 2000 dev tasks" 103 replies, called out as an ad, author sparred with commenters; 10-14 +31

- **Verified** **Overseas AI tool repost account** : Dan Kornas posted an English introduction at 10-14 23:44 UTC (on X and LinkedIn); 10-15 +29 the same day, plus 7 Southeast Asian forks

- **Verified** **by the author's own serialized build data** : 10-07 "1700+ tasks", 10-10 "1500 task milestones", 11-07 "running 40 Claude Code sessions at once"; nearly all of Trendshift's social mentions come from the author himself

- **Verified** **README and site conversion** : a big star-begging banner at the top, a three-language README, long-form production stats and retrospectives, a dedicated quicktui.ai page, and a one-line brew install

- Inferred **WeChat groups** : 11-11 The author's post said the group had passed 200 members and was inviting people; spread within the group can't be verified

- **Verified** **Lists and secondary writing** : listed on awesome-agent-orchestrators (around 11-10); a long-form Japanese note 10-12; several auto-aggregated pages

## dualface/kander: a terminal kanban for one person running 6 agents. Why did it suddenly gain stars this week?

> Research date 2026-11-13 (why.md says 11-07, data card refreshed to 11-13). Channel: GitHub. Data card: /home/ljr/fanout/why/cards/gh_dualface__kander.json

### Verdict

Kander didn't really "break out." It's a multi-agent orchestration tool that dualface, the author of QuickTUI (an app for remotely controlling AI agents on your computer from your phone), built for their own use. It's open source and has 69 stars after 305 days, while the leaders in the same category sit at 1k to 120k stars. Its stars came in waves, and each wave lines up with a specific post the author made in the Chinese dev circle:

1. 09-09 launched on V2EX (458 views, 2 replies), plus the author serialized his own production numbers on X ("10-071700+ tasks", "10-101500 task milestones").

2. 10-13 posted "Since Aug 2, I've completed nearly 1 dev tasks" on V2000EX. It drew 103 replies (calling it spam, questioning the "9KW users" claim). The next day, overseas AI repost account Dan Kornas shared an English introduction on X. Together the two brought 10-14 +31 and 10-15 +29.

3. 11-11 23:05 (Beijing time). The author piggybacked on yetone's Magpie "fully automated dev workflow" hype (the previous day, @xulijian1994's post had 110k views), tweeting: "Lots of people say magpie's automated development is amazing. You can have it just as easily. Two prerequisites: kander + lots of tokens." That post got 36k views, 213 likes, and 20 reposts, driving 11-11 +51 and 11-12 +78, which is the surge the data card shows as "7 days +133, prior week +3."

What made these posts stick was the author's use of data from his own production projects as evidence: 90 days, 2,477 cards, 2,865 independent reviews, and 51% of review batches catching must-fix issues. The data is also reproducible (the repo has the counting method and scripts). The product's differentiation ("execution and review are physically isolated, and a card can only close once review produces evidence"; cross-vendor agents reviewing each other) is where those numbers come from, but it isn't the direct source of the stars. Stars followed the author's posts, not the product.

### 1. What it is

- A single-file TUI written in Go (Bubble Tea), running on macOS, Linux and Windows. Each task card is assigned to one agent, which executes it in its own git worktree and tmux / Herdr window. When it's done, **another** agent reviews it (a PMQA stage, plus a Security stage when needed). Every must-fix finding has to be handled one by one as "fix / reject with evidence / defer" before the completion gate lets the card through. It also has conflict-aware scheduling (static analysis of branch overlap and the file scope of changes), and it reuses the `gh` credentials to import GitHub issues. [README](https://github.com/dualface/kander)

- The workflow is the reverse of tools like vibe-kanban, where a human drags cards around a board. Instead, the human first talks through the requirements in an agent session. An agent with Kander's rules installed creates the cards, starts work and schedules itself. The board is only for watching progress. The author put it this way: "The board is just a convenient way to check in. The core is the rules and the tooling" ([X reply, 11-12](https://x.com/dualface/status/2120563470479589845)).

- Supported execution / review agents: Claude Code, Codex, Grok, Cursor, Pi, Devin. The community has also added Kimi CLI and droid.

- Kander is a side product of QuickTUI and also a funnel into it. The task cards in the README screenshots come from the QuickTUI project. The site has a dedicated page at [quicktui.ai/open-source/kander](https://quicktui.ai/open-source/kander/), and the end of the README links to QuickTUI and the author's other tools. QuickTUI is a free app with a Cloud relay subscription. It has 2 ratings on the US App Store and 7 on the China store (iTunes lookup, 11-13), so it's small.

### 2. Who the author is (this is the "asset")

- V2EX profile: "author of quick-cocos2d-x, QeePHP, FleaPHP, QuickTUI" ([V2EX API](https://www.v2ex.com/api/topics/show.json?username=dualface)). An established open-source developer in Chinese game dev and PHP circles. The GitHub account was registered 2009 years ago, has 754 followers, and older repos include gbc-core with 363 stars and go-gbc with 97 stars.

- For 2026 years he's been shifting toward "one person + a crowd of agents" for building apps and games. He frequently posts his own agent usage and output on X (running 40 Claude Code sessions at once, burning 75% of the Ultra weekly quota in a day, and so on). He's also started a WeChat group with 200+ people ([X 11-11](https://x.com/dualface/status/2120427337112309800)).

- A batch of related small tools came out around the same time: claude-switch, ullage, ste-zh, ask-user-question, easy-codex, codex-loop, cc-leader and others, mostly from single digits to a few dozen stars. Kander is the only one that took off 100.

### 3. Growth: three waves, and each one matches a post the author made

| Timeline | Stars gained | what happened | Sources |
|---|---|---|---|
| 09-05 repo created, 09-09 V2EX launch | About 9-14 stars before 38 | V2EX "share what you built" post: "Made a multi-agent orchestration system that runs in the terminal," with a star request at the end. 12 hours, 458 views, 2 replies. | [kohnews compiled](https://kohnews.com/blog/dualface-kander-terminal-multi-agent-orchestration-v2ex/), Trendshift estimate (cumulative 307,9-14 starting from +269) |
| 09-17 | +11 | The author pushed 5 review-related issues / releases that day. Reason unclear. | [Trendshift](https://trendshift.io/repositories/215828) |
| 10-07~10-12 | Single digits per day | The author's Chinese posts: "Completed 1700+ tasks since August", "1500 task milestones", "press c to start." A Japanese note author 10-12 wrote a long piece analyzing "49% of review batches catching Must-fix." | Trendshift social mentions; [note.com](https://note.com/totorotom/n/nd8e5eb7aaef5) |
| **10-14 / 10-15** | **+31 / +29** | ① 10-13 on V2EX, "Since Aug 1, I've completed nearly 2000 dev tasks." 103 replies, mostly calling it an ad and questioning the numbers (the author replied "Lots of people up there are so clueless" and "I've been driving a Porsche for years"). ② At 10-14 23:44 UTC, overseas AI tool repost account Dan Kornas posted an English introduction (synced to LinkedIn). 10-15 7 forks appeared, mostly older accounts from Thailand and Malaysia that fork any new repo they see. | [V2EX t/1248529](https://www.v2ex.com/t/1248529);[Dan Kornas X](https://x.com/DanKornas/status/2110517015488204973);GitHub forks API |
| 10-16~11-10 | 0–5 per day | 10-29 released v1.0.0. The same day, jamesdwilson and other overseas users filed a batch of in-depth bug and feature issues. 11-10 joined the awesome-agent-orchestrators list. | GitHub releases / issues;[awesome-agent-orchestrators](https://github.com/andyrewlee/awesome-agent-orchestrators) |
| **11-11 / 11-12** | **+51 / +78** | 11-11 23:05 (Beijing time) the author piggybacked on the Magpie hype (see below), 36k views, 213 likes, 20 reposts and 33 comments. From 11-11 16:37 UTC to 11-12, 10 new forks, all from real Chinese-speaking developer accounts. | [Original X post](https://x.com/dualface/status/2120533312150388945); x-radar `x replies` (11-13 pulled live); forks API |

**The Magpie hype** : yetone's (author of openai-translator) [yetone/magpie](https://github.com/yetone/magpie) (35.5k stars, repo created 09-23) is a menu-bar tool for routing agent models. 11-10 @xulijian1994 posted a long thread on how yetone built a "fully automated dev workflow" with Claude Code and three `/loop` jobs that monitor issues, PRs and X feedback, then fix bugs and ship releases automatically. The post got 114k views and 869 likes (x-radar `x live "magpie agent"`). The next day the author replied "kander + lots of tokens gets you there," which caught the curiosity of people who wanted fully automated dev too. The replies were mostly "looking forward to the tutorial," "post it so we can learn," and "been looking for software like this." The author answered "currently slogging through the tutorial," which suggests the next wave of growth will hang on that tutorial.

**Star-farming check** : The stargazers API (REST and GraphQL) returns no detail for this repo. The fork accounts we can see (the 11-12 batch: peixl, registered 2011 years ago, 109 followers, 280 repos; iotsecty, caishuai, bsqsbylt and others are all multi-year accounts) show no farm-account traits. Star timing lines up exactly with the launch tweet. **Verdict: real follows, not bought.**

**Channels we didn't find** : No Hacker News or Product Hunt (our hackernews / producthunt databases return 0 for both kander and quicktui), no press coverage, no one in the core tier of our five x-radar circles mentioned it, and it never made GitHub trending (the repo wasn't 60 days old, so it couldn't qualify anyway).

### 4. Competitors: crowded space, Kander is the small one

GitHub repo data, 11-13; release dates for the big-company features are from memory, not verified.

| Product | Form | Launched | Scale | How it differs from Kander |
|---|---|---|---|---|
| stablyai/orca | Open-source desktop | 2026-03 | 123.6k star | Runs a team of agents on your own subscription, GUI; Kander is a TUI with strict review rules |
| herdrdev/herdr | Open-source terminal runtime | 2026-03 | 50.7k star | A terminal container for agents; Kander runs on top of it (Herdr backend supported) as the orchestration layer |
| yetone/magpie | Open-source menu bar app | 2026-09 | 35.5k star | Model routing, not orchestration; the "automated dev workflow" it got popular for is where Kander's traffic came from |
| BloopAI/vibe-kanban | Open-source web kanban board | 2025-06 | 28.7k star | Humans drag cards and read diffs; Kander has agents create and schedule cards, with another agent reviewing |
| manaflow-ai/cmux | Open-source terminal | 2026-01 | 28.8k star | Ghostty-based terminal with tabs and notifications; no review gate |
| getpaseo/paseo | Open source desktop + mobile | 2025-10 | 23.9k star | Desktop and mobile app orchestrating multiple agents with a UI; also a direct competitor to QuickTUI |
| smtg-ai/claude-squad | Open-source TUI | 2025-03 | 8.8k star | The closest TUI multi-agent manager to Kander, but it doesn't handle review or completion gates |
| stravu/crystal(Nimbalyst) | Open-source desktop | 2025-06 | 3.1k star | Parallel worktree sessions with a diff panel; humans do the review |
| coder/xum | Open-source desktop | 2025-09 | 2.1k star | Desktop app for isolating parallel agents |
| phodal/routa | Open source | 2026-02 | 1.8k star | Multi-agent collaboration platform with shared specs and a board, by a Chinese-speaking author |
| cline/kanban | Open-source Web | 2026-03 | 1.4k star | Local web board for running CLI agents, from a big company (Cline) |
| kdlbs/kandev | Open-source Go | 2026-01 | 1.1k stars, +60 over 7 days | Board plus approval steps plus web IDE, opens PRs; Kander is more "rules first" and terminal-first |
| emanuele-em/kanzo | Open source Rust | 2026-10-25 | 1.2k stars, +145 over 7 days | Lightweight board shared by humans and agents, growing faster in the same period |
| EricTechPro/super-board, TechDufus/openkanban, herdr-board, kanban-md and others | Open source | 2026 | 150~670 star | Similar small repos; Kander's review evidence chain is more complete |
| Claude Code subagents / agent teams, OpenAI Codex App worktree parallelism, Cursor parallel agents | Big-company features | 2025-10 ~ 2026 | Installed by default | Parallelism inside a single vendor's agent; Kander does cross-vendor review (Claude writes, Grok / Codex reviews) |
| runpane.com(Pane) | Commercial desktop app | — | Unknown | Already published an SEO page, "Best Kander Alternatives in 2026", to capture Kander's search traffic |

**Bottom line** : The combo of "worktree isolation + board + multiple agents" was already overdone by mid-2025. What Kander can actually claim is two things: "review must be done by a different agent, and issues must be resolved with evidence before a card can close", and "use a cheap model to review an expensive one" ([author's blog](https://dualface.dev/posts/cheap-model-for-code-review/)). There are many similar small repos, so this difference isn't enough to drive large-scale growth on its own.

### 5. Why this one, and why now

1. **Timing** : In November, "fully automated dev / one-person army" was hot on Chinese X (Magpie's workflow post got 114k views). People wanted to know "how to actually do it", and Kander was packaged as "the answer + lots of tokens".

2. **The author brings his own credibility and audience** : author of quick-cocos2d-x, an old V2EX account, frequently posts his usage on X, and runs a WeChat group with 200+ people. All three star waves were started by him personally.

3. **Use production data as the hook** : 2,477 cards, 51% mandatory-pass rate, and a rejection-rate table by model (Claude Opus 5.5 rejected 0.78 times per card, Codex gpt-6-sol 0.36 times...). Numbers from real runs like this are persuasive to people fighting false-green code from agents, and a Japanese note author has already written a long piece on it.

4. **Controversy also brought exposure** : The V2EX post on 10-13 got called an ad, the author fought with commenters, and that pushed it to 103 replies, with +31 stars the next day.

5. **README conversion design** : A big "please star" image at the top, a trilingual README (Chinese, English, Japanese), a long production retrospective plus a PDF slide deck, a one-line Homebrew install, and fast iteration (the author made 858 commits; from 09-24 to 11-11 shipped 30 releases, v0.8.2 to v1.3.0).

### 6. How many people can see it

**Spreading in small circles.** On Chinese X's AI coding circle there was a tweet with 36k views, two V2EX posts (one with 103 replies, mostly insults), a repost from an overseas aggregator account, a Japanese note post, inclusion in an awesome list, and an SEO "alternatives" page from a competitor. No HN, PH, or press, and no one among x-radar's core accounts mentioned it. No revenue involved (MIT open source; the QuickTUI behind it hasn't disclosed revenue).

**Copycats** : Didn't see anyone copying Kander. runpane's "Best Kander Alternatives" is competitor traffic capture, not copying.

### 7. What to borrow

- **Riding a hot topic** : While others were flooding feeds with "automated workflows", the author posted the next day that his open-source tool plus tokens could do the same, and that one tweet brought 130 stars. Timing was key; the gap was only one day.

- **Use your own production data as the README's first screen** : Spell out the data method, provide a reproduction script, and break tables down by model. That gets more stars than a feature list, and it's easier for others to turn into articles.

- **Open source as a funnel, paid main product** : Kander ships with QuickTUI (free app + Cloud subscription), and its website has a dedicated tutorial page on "running Kander on your phone".

- **Chinese launch first, then spread to three languages** : V2EX and Chinese X built the initial traction; the README has English and Japanese versions so overseas aggregators (Dan Kornas) and Japanese writers could repost directly.

- **Next step to watch** : The author says he's writing a tutorial. When it comes out, that's the next growth point.

### 8. Uncertain

- Stargazer details aren't available, so I couldn't check each starring account; I only verified fork accounts, and they're all real.

- Who retweeted the 11-11 tweet 20 times, and whether any big accounts did, I couldn't find. The author's X follower count parsed as 754, exactly matching his GitHub followers, so the parse may be wrong; unverified.

- The source of the +11 on 09-17 is unknown.

- How many people actually use it can only be inferred from issues: about ten people such as jamesdwilson, ttlbb and zhouatie filed in-depth issues, which points to a small real user base, but there's no quantified data.

- Release timing for the big-company features is from memory, not verified.

This article was written by AI from public data and the data we collect daily. Numbers are sourced, and inferences are labeled.
