Stringers
Axium-Labs/AX · Axium-Labs

A terminal coding agent written in Rust gained 131 stars in four days: of the 21 accounts that forked it, 19 are the same batch of fake accounts

What it is A single-file terminal AI coding agent written in Rust, paired with a desktop console that has an Android app, allowing task orchestration across multiple machines with resume after disconnects.

Why it blew up It didn't, really: the stars were bought from a boosting service in two waves, with 17 days in between where nothing moved. Real downloads are in the double digits, and there's zero discussion of it anywhere online.

DHow you could approach it Don't build a generic terminal agent like everyone else. If you want in on this space, pick one small piece the big vendors' CLIs handle badly (resuming work across machines, approving from your phone), build it as a small tool that plugs into existing CLIs, and post a demo video on X and HN.

D grade: the stars were bought, so nothing proves anyone wants this feature mix. Building it means facing opencode, Claude Code, and Codex, all at 100K-star scale. The only thing worth taking away is how to spot bought stars. Hardly anyone knows about it yet
Data (updated daily) · Repo created 2026-09-22 · 410 stars · 7 days +20
GitHub · stars
020040010-0910-2311-0111-1011-20410
Data 2026-10-09 → 2026-11-20 (43 days, 37 observations). Points before 10-28 come from deeply / archive backfill, not every day has one

Early Nov 9, a GitHub account called ravenjudges8e forked AX, a terminal coding agent. The owner is Bernard Greco, whose bio reads "Teacher — reading, woodworking." The account was created in Oct 26 and has 0 followers. Over the same week, a dozen-plus other accounts with the same pattern of name and bio also forked it, including Mason Walker ("Working as Photographer. Love coffee & kayaking.") and Henry Lewis ("painting · woodworking · Pharmacist"). AX's stars went from 113 to 247 in just these few days. The project has a fair amount of features, but its stars are mostly bought, and the people who've installed it number in the double digits. What it can teach you is how to spot bought stars at a glance, and what buying them doesn't get you.

What it does

AX (github.com/Axium-Labs/AX, Rust, Apache-2.0) was created in Sep 22. It's a single-binary AI coding agent for the terminal. The README's first line reads "A fast, lightweight AI agent for the terminal, built in Rust." It can read and write files, run commands, connect MCP and Skills, log in with a Codex account or an API key, and keeps chats, memory, and config on the local machine. It also has a few advanced features: Goal (set acceptance criteria and it keeps editing until tests pass), Long Project (keeps a project progress file across sessions), extracting skills from execution logs, and Computer Use. The README itself admits "No benchmarks are published yet," so the "fast" claim has no test data behind it.

The same org also has AXCrew, a desktop control panel for AX with an Android app. On your local network you can approve actions and assign tasks, orchestrate AX across several machines as one distributed job, and resume from checkpoints after a disconnect. The site's tagline is "One agent system across every machine you own."

The author is active: from Sep 22 to Nov 3 they shipped 20 releases (v0.1.0 through v0.3.17), the README is bilingual Chinese and English, and there's a GitCode mirror. 103 commits, 102 of them from the same developer. That GitHub account now returns an error page, and I can't tell whether they deleted it or it was banned. It looks like an individual or small team in China.

The star curve looks like steps

DateStars (GitHub, daily)fork
10-09201
10-15521
10-181156
10-19 → 11-04115 → 112 → 1136–7
11-051167
11-0717110
11-09244 (live as of the same day 247)18 (live 20)

4 days gained 95 stars, then 17 days of nothing, then 4 days gained 131. Across those 17 flat days releases kept coming out (10-19, 10-23, 10-26, 11-03, one each), yet stars didn't move at all, and in Oct 25 it actually dropped by 3. A project that genuinely got noticed shows a decay tail after the spike, not a hard cut like this.

Who forked it

GitHub shows 21 forks of AX (11-09). Apart from Keenable, a search API company, and a deleted account, the other 19 look identical:

  • Usernames follow "random English word + two alphanumeric characters": ravenjudges8e, shindig-callous8j, decimal-lantern7f, 29daresimmer, integerintense-5t;
  • All were registered between late August and late October, with 0 followers and 0 following each;
  • They use common Western names, and their bios follow a template: occupation plus two hobbies;
  • The fork timing lines up with the two star waves: Oct 14 to 16 day 6 forks, and Nov 5 to 9 day 13 forks.

Looking at what else these accounts forked: nearly all are small Chinese developer repos created in the last month or two, such as a McDonald's savings guide, an AI coding collection, and a Claude desktop client. Those repos also gained stars in sudden spikes and then flattened or fell back. One lost 7 stars in 214 days, which looks like GitHub cleaning out a batch of fake accounts. It's clearly a service that sells stars and forks to small Chinese repos, and AX was one of its customers. Whether the author bought them directly or joined a mutual-boosting group, I can't tell, and the conclusion is the same either way.

I couldn't pull the full stargazer list, so the bought-star ratio is inferred from the fork list (an inference). Still, the two star waves and the two waves of suspicious forks line up exactly in time.

247 stars, roughly 120 downloads

AX's install script pulls from GitHub Releases. Across all 20 releases, total downloads come to about 120 (11-09). The most-downloaded, v0.3.16, got 20, and the latest, v0.3.17, got only 6. AXCrew's two releases total 23, and the GitCode mirror shows 0 downloads. The repo has zero open issues, and the only outside PR is the Keenable one.

Nothing dedicated to AX turned up on X, HN, Reddit, V2EX, Zhihu, or Bilibili. Searching "AX agent" mostly returns Google's Agent Executor (google/ax, 15k stars), which it released in March this year. The name collision makes it hard to find through search.

Who's getting stars in this space

Terminal coding agents are one of the most crowded open-source tracks right now (star count per 11-09 GitHub, checked live):

ProjectWhat it isstar
opencodeTop multi-model open-source CLIs226k
Claude CodeAnthropic's official CLI155k
OpenAI Codex CLIOpenAI's official CLI, also in Rust134k
Gemini CLIGoogle's official CLI109k
GooseOpen-source Rust agent with a desktop app56k
Vibe Kanban / claude-squad and othersMultiple agents in parallel, task orchestration3k–29k
HappyRemote control for Claude Code / Codex from your phone25k

Every feature AX offers has a more mature competitor. Rust plus a desktop app: Goose had that long ago. Phone approval is Happy's main selling point, and AXCrew only works on the local network and only controls AX. Multi-task orchestration: Vibe Kanban already does it. A "terminal agent + desktop + Android + multi-machine orchestration" bundle built in Rust is rare, but there's no benchmark, no demo video, and no user feedback showing anyone wants that combination.

What the bought stars actually got it

They got some automated attention. In Nov 7, a skills-directory site listed it when it had 211 stars (that site sells featured placements for $49 / $99 over 30 days, but there's no sign AX paid). Two other open-source radar sites that auto-scrape repos by star growth also have pages for it. In Oct 29, an employee of Keenable opened a PR adding their search MCP to the project, and that employee has opened 368 similar PRs, apparently scanning agent repos in bulk.

All of this followed the star count rather than bringing it in. Keenable's PR was merged, and for the 7 days after, stars didn't move at all. When the directory site listed it, the second wave of fake accounts was already underway. There's no sign that real users came in because of any of this, and the release download numbers confirm it.

If it were you

In one line: take a look, but don't copy it. AX's stars were bought, so it doesn't prove that a multi-machine terminal agent has demand. If you build one, you're up against opencode, Claude Code, and Codex, all at 100K-star level.

Two things worth learning:

  • How to spot bought stars: are the fork accounts "word + two-character suffix" with template bios? Is the star curve stepped, with a suspicious flat middle? Are release downloads absurdly low compared to stars (AX: 247 to 120)? Repos under 60 days old that never made GitHub Trending are the first ones to check this way.
  • Buying hundreds of stars gets you directory listings and bot PRs from integrators. It doesn't get you users or reputation, and anyone who knows the space can see it from the fork list in a glance.

Who it's for: people already building agent tools who want to see where real demand is in this space. The "run agents across several machines + approve from your phone" direction does have demand: Happy at 25k stars and Vibe Kanban at 29k stars are proof.

Who it's not for: anyone who wants to build a generic terminal coding agent from scratch to win stars. That slot is already taken by big vendors and a few top open-source projects, and no matter how long the feature list, it's hard to get noticed.

If you do build, go narrow: skip the all-in-one agent. Pick one small piece the big vendors' CLIs handle badly, like resuming long jobs across machines or approving Codex / Claude Code actions from your phone. Build it as a plugin or small tool that works with existing CLIs, and post a 30-second demo video on X and HN. That's how Happy got started: it didn't compete with the vendors on agents, it just does remote control.

What window? There's no time window here. AX didn't create any demand of its own.

What would show this read was wrong: if AX's downloads climb with its stars over the next month, real users open issues, or people seriously discuss its multi-machine orchestration on X or HN, the combination has real demand and deserves another look. On the other hand, if stars start dropping in batches or the fake accounts get purged, this read is confirmed.

GitHub repo ↗

Research notes: competitor table, acquisition evidence, full research report
Competitors
NameWhereLaunchedScaleHow it differs from this one
Claude CodeBig-company features2025-02GitHub 155k starsNot tied to a model, data stored locally, built-in multi-machine orchestration and desktop/Android apps; otherwise behind on nearly everything
OpenAI Codex CLIOpen source2025-04134k starsSame Rust single binary; AX logs in directly via Codex OAuth, close to a new wrapper around a Codex account
opencodeOpen source2025-04226k starsTop multi-model open-source CLI; AX adds Goal, Long Project, and desktop/Android apps
GooseOpen source2024-0856k starsRust plus a desktop app was already Goose's combo; AX's addition is distributed multi-machine orchestration
HappyOpen source2025-0725k starsPhone remote control for Claude Code/Codex; AXCrew's Android approval is the same selling point, but it only works on the local network and only controls AX
Vibe Kanban / claude-squad / CrystalOpen source20253k-29k starsParallel multi-agent and task orchestration; AXCrew adds cross-machine scheduling and resume after disconnects
google/ax(Agent Executor)Big-company features2026-0315k starsName collision; Google's AX is a K8s-style orchestration runtime, and nearly every "AX agent" search returns it
Where users come from
  • Confirmed star/fork farming service: of 21 forks, 19 come from fresh accounts with "word + digits/letters" usernames (registered late August to October), 0 followers each, and bios that are copy-paste templates. The timing lines up with two star waves (10-14~16 had 6 of them, 11-05~09 had 13). The same batch also forked a dozen-plus small Chinese repos like mcd-saver-strategist, AI-Coding and McOracle. Between the two waves, stars sat still for 17 days, and 3 stars were removed.
  • Verified auto-listed by directory sites: skillsllm.com listed 11/7 (back then 211 stars, selling a 49/99-dollar Featured slot; no sign AX bought it). ossradar.net and gitofanton.com also picked it up. All are auto-scraped based on star growth, so this is a result of star inflation.
  • Verified GitCode/AtomGit mirror (G-Star incubation): gitcode.com/axlab/AX has 72 stars, shows the G-Star incubation badge, and its download count reads 0
  • Checked: Keenable integration PR: a Keenable employee, 10-29, submitted a PR adding Keenable MCP as a search fallback. keenableai 10-31 forked it. It was one of the batch of integration PRs Keenable opened (368 in total). Stars did not rise afterward.
  • Inferred: social and community content: No dedicated AX coverage found on X, HN, Reddit, V2EX, Zhihu or Bilibili.

Axium-Labs/AX: a Rust terminal AI coding agent whose stars were inflated.

Bottom line: AX never really took off. Its 244 stars arrived in two waves (10/15–10/18, +95; 11/05–11/09, +131), with three weeks of no movement in between. Over the same period, almost all accounts forking it were batch-registered fake accounts, and the project's total downloads come to about 120. It is a feature-rich project from a small Chinese team that bought or bartered stars. No genuine community spread turned up.

Data as of 2026-11-09 (our GitHub database + live GitHub API queries).

---

1. What it is

  • AX (github.com/Axium-Labs/AX, Rust, Apache-2.0,2026-09-22 license): a single-binary terminal AI coding agent that reads and writes files, runs commands, and plugs into MCP and Skills. It supports Codex OAuth and API key login, and stores data locally. Advanced features include Goal (set acceptance criteria and it iterates until tests pass), Long Project (keeps a project progress doc across sessions), distilling candidate skills from execution logs, and Computer Use. The README states: "No benchmarks are published yet".
  • AXCrew (same organization, created 2026-09-27, 102 stars, 0 forks): AX's desktop console, written in Rust, with an Android client (approve and dispatch tasks over the LAN). It can connect to SSH workspaces, orchestrate AX instances across multiple machines as distributed tasks, and resume from checkpoints after a disconnect.
  • Website: axium.ax-labs.workers.dev, tagline "One agent system across every machine you own."
  • Authors: of 103 commits, 98+4 came from HuangZeLinCute (email [email protected]), and 1 from Keenable's Ilya Bogin. The HuangZeLinCute GitHub account can no longer be found (the users API and noreply id 122612645 both return 404). The account was deleted or banned; the reason is unverified. The README is bilingual Chinese/English, and a mirror was opened on AtomGit/GitCode at gitcode.com/axlab/AX (72 stars, with a G-Star incubator badge). Judgment: an individual or small team in China.
  • Iterates quickly: from 9/22 to 11/03 it shipped 20 releases (v0.1.0 through v0.3.17).

2. Competitors

Terminal coding agents are one of the most crowded open-source tracks right now. Star counts are 2026-11-09, pulled live from the GitHub API:

ProductFormLaunchedScaleWhat sets AX apart
Claude Code (anthropics/claude-code)Big-lab CLI2025-02155k starsAX is model-agnostic and local-first, with built-in multi-machine orchestration
OpenAI Codex CLI (openai/codex)Big-company CLI, also in Rust2025-04134k starsBoth are single Rust binaries. AX logs in directly via Codex OAuth, so it is close to a new shell around a Codex account.
opencode (anomalyco/opencode)Open source CLI2025-04226k starsTop multi-model open-source CLI; AX adds Goal, Long Project, and desktop/Android apps
Gemini CLIBig-lab CLI2025-04109k starsSame as above
Goose (aaif-goose/goose)Open-source Rust agent with a desktop app2024-0856k starsThe Rust-plus-desktop combination already exists in Goose. AX's added value is multi-machine orchestration.
Crush / Qwen Code / Kimi CLI / AiderOpen source CLI2023–202511k–50k starsSame product category
Happy (slopus/happy)Remote-control Claude Code/Codex from your phone2025-0725k starsAXCrew's Android approvals are exactly Happy's core selling point
Vibe Kanban / claude-squad / Crystal / OmnaraParallel multi-agent runs, remote control20253k–29k starsAXCrew's multi-machine, multi-task orchestration also falls in this category
google/ax(Agent Executor)Big-company orchestration runtime2026-0315k starsName collision: searching "AX agent" returns mostly Google AX, so AX is hard to find through search

Conclusion: every AX feature has a more mature version elsewhere. Its distinctiveness is packaging a single CLI, desktop app, Android app and multi-machine orchestration into one Rust stack, and no competitor offers that combination. But it has no benchmarks, no demo videos, and no user feedback showing the combination is appealing. Its stars did not come from it (see below).

3. Star curve and growth tactics

3.1 curve: two steps with a flat stretch in between

repo_day data from our GitHub database (10-09 to 10-27 is deep history; daily snapshots started 10-28):

Datestarfork
10-09201
10-15521
10-16874
10-171065
10-181156
10-19 → 11-04115 → 112 → 113 (17 days barely moved, 10-25 even dropped 3)6–7
11-051167
11-061299
11-0717110
11-0819812
11-09244 (live API check 247)18 (checked live 20)

AXCrew 11-06 was first spotted at 52 stars, 11-07 was 81,11-08 was 101, and then it stopped moving. Its fork count has stayed at 0.

Repos that genuinely take off show a decaying long tail after the peak. AX's curve instead reads "+4 in 95 days, flat for 17 days, then +4 in 131 days." Between the two steps, releases kept coming (10-19, 10-23, 10-26, 11-03), but stars did not move at all.

3.2 Fork list: 19/21 are the same batch of fake accounts

A live GitHub API query of AX's 21 forks shows that apart from keenableai (10-31) and Treppeo (9-30, now 404), 19 more have usernames made of a random English word plus a alphanumeric suffix, such as ravenjudges8e, shindig-callous8j, decimal-lantern7f, 29daresimmer, integerintense-5t. These accounts:

  • were all registered between 2026-08-26 and 10-26, with 0 followers and 0 following;
  • use generic Western names (Bernard Greco, Henry Taylor, Liam Evans…);
  • have bios drawn from a few fixed templates: "Just a chef who enjoys astronomy and painting." "Working as Photographer. Love coffee & kayaking." "Teacher — reading, woodworking";
  • timing lines up exactly with the two star waves: 10-14 between 10-16 and 6, and 11-05 between 11-09 and 13.

Looking at what else these accounts forked: almost all are small Chinese developer repos created in the last month or two, such as hujjun/mcd-saver-strategist (1329 stars), liuda30201/AI-Coding (605), spideymooon/McOracle (609), oleyyu/OpenClaudeDesktop (430), piperhex/remoteai (1451), 7sen1ot/zcode2api, wusuoweiju48/google-anti-cn-doctor, Dushiqi1987/codex-remote-Skill and others. In our database these repos show the same pattern: a one-day spike, then flat or falling. mcd-saver-strategist lost 7 stars over the last 214 days, which looks like GitHub purging a batch of fake accounts. The likely explanation is a service network that buys stars and forks for small Chinese repos, with AX as one of its clients.

(The stargazer list could not be pulled: our token is fine-grained, and neither the REST nor the GraphQL stargazers endpoints return it. The fork list is enough to make the point.)

3.3 Real usage: about 120 downloads

GitHub Release asset download counts. Across all 20 AX releases, about 120 downloads (v0.3.16 has the most, at 20; v0.3.17 has only 6). AXCrew's two releases total 23. The GitCode mirror page shows 0 downloads. The official install script pulls from GitHub Releases, so the real install count behind 247 stars is likely in the double digits. There are also 0 open issues, and only one external PR, from Keenable.

3.4 Checked channel by channel

ChannelResult
Xx-radar search over the last 60 days for "Axium-Labs", "AXCrew", "axium.ax-labs" and "Axium AX agent": 0 results
HN / Reddit / V2EX / Zhihu / BilibiliJina, Exa and Brave found no posts specifically about AX. The Reddit "Rust local coding agent" hit was a different project, xencode.
GitHub TrendingNot posted anywhere (the repo is under 60 days old, so it could not qualify; this proves nothing)
Directory sitesskillsllm.com listed it 11/7 (with 211 stars at the time), and ossradar.net and gitofanton.com have pages for it too. All of these pull from GitHub star growth automatically, so they show the result of starring, not the source. skillsllm sells Featured slots for $49/99 for 30 days, and the page doesn't show whether AX bought one.
KeenableA Keenable (search API startup) employee 10-29 opened a PR adding the Keenable MCP as a search fallback, and keenableai 10-31 forked it. Keenable is batch-submitting integration PRs to agent libraries. Ilya Bogin has 368 PRs in total, and AX is just one of the repos it hit. Stars didn't move in the 7 days after, so this isn't a real growth source.
GitCode G-StarThe mirror carries a G-Star incubator badge and has 72 stars. It is one domestic distribution channel, but a small one.
AdsOpen-source CLI; no ad campaigns found.

4. Why it "broke out" (it didn't really)

  1. Stars came from a boosting service: both star waves line up exactly with the fork timestamps of 19 template fake accounts. The repo sat still for 17 days in between and even lost stars. The same batch of fake accounts was also boosting a dozen or so small Chinese repos over the same period.
  2. No genuine spread: no discussion on X, HN, Reddit or Chinese communities, and total downloads come to about 120.
  3. The product really does have a lot of features: 48 days, 20 releases, complete docs, bilingual Chinese/English, plus desktop and Android apps. That is what makes the boosted repo look like a legitimate project on the surface and lets it slip past recall rules that rely on "solid code + rising stars."
  4. What the boosting bought: entry into auto-curated directories like skillsllm, ossradar and gitofanton that index by star growth, plus our own recall. That is probably the point: to get directory sites and repo-scanning bots like Keenable's to come looking.

5. Differentiation

On paper the differentiation is "one Rust stack: terminal agent + desktop console + Android approvals + distributed multi-machine orchestration with checkpoint resume + Goal acceptance iteration + skill distillation from execution logs." Taken individually, Claude Code, Codex, opencode, Goose, Happy and Vibe Kanban each have more mature versions, but packaging them together is rare. However, this differentiation has nothing to do with the stars: nobody is discussing it, downloads are tiny, and the stars were bought. The name AX also collides with Google's Agent Executor (google/ax, 15k stars), so it is nearly impossible to find through search.

6. What to borrow

  • A cautionary tale: buying a few hundred stars for an agent open-source project can get it indexed by auto-directories like ossradar and skillsllm, and picked up by integrators' bots and data-recall systems. But it does not bring real users, and the download numbers give it away.
  • How to spot it (reusable): check whether fork account names follow the "word + two-digit suffix" pattern, whether bios are template sentences, whether the star curve is stepped, and the ratio of Release downloads to stars. This is especially useful for repos under 60 days old that are not on trending.
  • At the product level, the "multi-machine agent orchestration + phone approvals" direction clearly has demand (Happy at 25k stars and Vibe Kanban at 29k stars prove it), but the leaders already have it.

7. Uncertain

  • The stargazer list could not be pulled (insufficient token permissions), so the share of boosted stars is inferred from the fork list and was not verified account by account.
  • Whether the author's account 404 was removed on purpose or banned by GitHub is unverified.
  • We can't tell if the star-farming was bought by the author or came from a mutual-star group; either way the conclusion is the same.

Sources

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