A hackathon with 7,615 signups bought 518 stars in a single day: how Switch's stars were bought

Switch, open-sourced by SandboxAQ, brings agents like Claude Code and Codex into Slack, Teams, and Discord channels to work alongside people. It grew by 113 stars over 4,000 days, more than half of them in the last 10 days. Yet it has barely been discussed on HN or X. These stars were bought one by one by a well-funded company: a press release first, then a Product Hunt launch, then sponsored newsletters, LinkedIn influencers, and finally a prize hackathon. Below: how many stars each move bought, which of these tactics individual developers can use too, and why I don't recommend going head-on into this category.

Nov 4 at midnight: hackathon kicks off

On Nov 4 2026 at 0:01 (India time), the "Flip the Switch" hackathon, co-hosted by the Indian developer community WeMakeDevs and DevTools Academy, kicked off online and runs until Nov 7. The signup page says "7,615 registered," with a total prize pool of $10K:

• Best business workflow and best personal/creative workflow, $2,000 each;
• "Best Core Contribution": PRs to Switch's open-source core repo, the best 20 get $100 each;
• Best project built with the voice tool Wispr Flow, $500; best blog gets a pair of AirPods; post on social media with #HowISwitch, and the top 5 each get a Logitech mechanical keyboard; 20 randomly picked entries get merch, and every submission gets a certificate.

The signup page spells out the contribution rules in detail: first, pick an issue tagged hackathon in the repo, comment @-mentioning a Switch employee with your plan, wait for assignment before starting, each person can have at most 5 open PRs at once, and your first PR requires signing the contributor agreement. Two days before kickoff (11-02), WeMakeDevs founder Kunal Kushwaha posted a livestream promo for the kickoff.

That day, the repo's stars went from 2,954 to 3,472, +518, its biggest single-day gain ever. Forks went from 190 to 215, then to 278 over the next two days (GitHub daily stars, 11-03 to 11-06). When thousands install Switch Console, claim issues in the repo, and fork the code, clicking star along the way is natural.

What it is

Switch calls itself an "open-source AI coworker": a self-hosted middle layer that connects on one end to agents (Claude Code, Codex, OpenCode, or agents written with LangChain) and on the other to chat apps (Slack, Microsoft Teams, Discord, Telegram, Mattermost). Each agent has a role, and a "Manager" opens rooms, adds people, and passes context between agents. The website and copy repeat two points: it is not a chat app, and it is not an agent vendor. Swap agents or models and the context stays; the data stays in your hands.

Getting started means a desktop app (Switch Console, mainly macOS for now) plus Docker. Running locally includes a bundled Mattermost, and within 10 minutes agents can reply to you in a channel. The docs also offer an MCP server so your own agent can read the docs and teach you how to use it.

Behind it is SandboxAQ, a company spun out of Alphabet that works on AI and quantum security, and it has raised a large amount of funding. The top commit authors are all company employees (amaudruz alone has 513 commits), and Console ships several releases a day. The license is Apache 2.0 plus Commons Clause, which accurately means "source visible" but no commercial resale. The money isn't in the repo: in Oct 19 the company launched two paid enterprise products, Switch Trust (risk and guardrails) and Switch ROI (cost and return), backed by a reference customer, New York's Mount Sinai West hospital. The open-source part is the top half of the sales funnel, and the stars prop up the "developers are using it" facade for enterprise sales. (Inference.)

The stars came in three waves, and each one matches a specific expense

(GitHub daily stars; there is no day-by-day data from 10-18 to 10-27, so these 11 days' ~+1,100 can't be broken down further. By later, the repo page showed 11-06 stars and 4,028 forks, 290)

The first wave is the level of a big company's normal open-source launch: a press release plus the #1 PH spot, and 750 in two months.

The second wave is the most worth examining. In Daily Dose of Data Science's Oct 19 issue, Switch was one of three pieces: first, it notes that Anthropic, OpenAI, and xAI are all racing to build an "AI coworker," but each keeps the model, runtime, and memory inside its own walls; then it demos a scenario: the Manager is given two production incidents (after a morning deploy the inference API starts reporting 500, and a training job stalls for six hours), it opens two linked rooms, brings in software engineer, architect, and QA agents to investigate in parallel, and finds that a single config push caused it, then posts a merged incident postmortem in the original channel. The body links three times to the GitHub repo, the end has "(don't forget to star 🌟)", and the line below it reads "_Thanks to Switch for partnering today!_" In other words, this was clearly labeled sponsored content that directly asked readers to star the repo.

Both were paid placements, but The Neuron's link went to the website and barely brought in any stars in a week. Daily Dose linked to the repo with "don't forget to star," and the count then rose about 1,100 in 11 days (the 10-28 influencer wave contributed too, so the two can't be separated). Stars grow where the ad links.

In Oct 28, Indian LinkedIn users Hina Arora (careerwithhina, 143 comments) and Brij Pandey posted the same day, marked as Brand partnership, with nearly identical copy: first the "copy-paste tax" of pasting chat logs into AI and pasting the answers back into chat, then the question "what if the agent just joined the channel?" The same copy sent to a batch of influencers is classic influencer marketing.

The third wave is the hackathon from the opening. What makes a hackathon better value than buying posts is that signups don't just glance at it: they actually install, use, and go claim work in the repo. The "$100 per PR, top 20 winners" line cost only $2,000, and bought a wave of forks, issue discussion, and PRs. The small prizes for best blog and #HowISwitch then produced a batch of "I built X with Switch" posts on dev.to (several appeared between 11-03 and 11-06), which also created search-engine content as a side effect.

On the free channels, it was nearly invisible: no HN post; on X, the only thing you can find is one HPCwire reposting of the press release from Aug 27, with 588 views and 0 likes; on Reddit, just one news repost in r/artificial.

Who it's competing with

(GitHub,11-06)

The table shows two things. First, the need to "plug agents into chat apps" is real: cc-connect and nanoclaw, both built by individuals, have far more stars than Switch, and they grew organically. Second, this path is already crowded, with Slack itself and every agent vendor pushing into it. Switch's product differences (five chat platforms, works with any agent, self-hosted) are real, but its stars didn't come from those differences. Those differences only make the paid promotion make sense.

How many people know about it

The circle is small and mostly bought. English industry media ran the press release, and the real buzz is on LinkedIn and in Indian developer communities (WeMakeDevs, DevTools Academy, hackathon posts on dev.to). I haven't seen anyone copying it. Similar projects either came before it or appeared at the same time.

If it were you

Verdict: C. You can learn a whole playbook for buying stars, and several of its moves are affordable for individuals. But the repo itself isn't worth following. Its 4,000 stars came from a big company paying for distribution continuously, which followers can't match. The organically grown projects in the category, like cc-connect and nanoclaw, already have over ten thousand stars, and Slack's official features are pressing down on top of them.

Who it's for: developers or small companies that already have an open-source project that works reasonably well but nobody knows about, especially developer tools willing to spend a few hundred to a few thousand dollars on promotion. Also useful for people doing open-source operations (DevRel) at a company, since this is a timeline you can copy.

Who it's not for: individuals who want to build a "plug agents into Slack / Teams" project from scratch. You'd be competing with Slack's official offering, with every agent vendor, and with a pile of individual projects already at ten thousand stars. In this category, a new repo is almost impossible to notice.

How to do it, ranked by how worthwhile:

1. Copy its promotion, not its category. The cheapest and most effective move is the second step: find a developer-focused newsletter or a big LinkedIn tech account and sponsor one issue. Ask for the body to link directly to the GitHub repo, and make a demo video for one specific scenario (Switch's demo runs "two production incidents, one message, investigated in parallel"), ending with "don't forget to star it." Don't copy The Neuron's ad that links to the website. That one brought almost no stars.
2. With a bit more money, run a small hackathon that rewards PRs to the core repo. It doesn't need 10k dollars. Co-host it with a developer community (WeMakeDevs in India, or similar communities in Chinese-speaking regions, Latin America, or Southeast Asia). First, tag a batch of issues in the repo, set "the best N PRs get $50–100 each," and add a best-blog prize. Stars, forks, issue discussions, and tutorial posts will come along.
3. If you really want to work in this direction, take the part big companies won't. For example, build shared multi-person agent rooms only for Chinese enterprise IM like Feishu, WeCom, and DingTalk (cc-connect is single-user remote control, and Switch doesn't support Chinese IM). Or only build the narrow piece where multiple agents hand off work within one channel. The star ceiling is lower, but you don't have to go head-on against Slack.

How to get users: rank the channels the way Switch validated them. Developer newsletters and LinkedIn tech accounts that link to the repo come first, then co-hosted developer-community events, then tutorial posts on dev.to and similar platforms, then press releases and Product Hunt. Switch never tapped HN or X. If you can catch fire there organically, you won't need to spend this money.

Before when?: No fixed deadline on learning the playbook. Item 3 is best acted on before the first half of 2027, since Slack, Teams and Feishu are all adding features to put agents in channels, and the open space will keep shrinking.

What would show you got it wrong: ① Two or three weeks after the Nov 7 hackathon ends, Switch is still gaining three-digit stars a day, or it hits the HN front page. That would mean the product has pull of its own, and this category is wider than estimated here. ② Conversely, if by late November it drops back to single-digit stars a day, the stars were mostly from event participants clicking along, so the effect of item 2 should be discounted, and the numbers only count as one-time. ③ If an individual project gets a thousand stars from one or two newsletter sponsorships, then item 1 is worth even more than estimated here.

Data and sources

GitHub: the sandbox-quantum/switch repo page, commit history, daily stars and forks (8-27, 10-09 to 11-06), and star counts for similar repos (11-06). WeMakeDevs hackathon page (wemakedevs.org/hackathons/switch, read via 11-06). Daily Dose of Data Science, "Agents 2.0 is here!" (2026-10-19). The The Neuron sponsorship record on sponsorgap.com (10-13). PR Newswire press release (8-26) and SandboxAQ news page (10-19). Product Hunt launch page (9-08). LinkedIn: posts by Akshay Pachaar, Hina Arora, Brij Pandey, Kunal Kushwaha, and others. Anything marked "inference" is inference.