Paperclip's 90,000 GitHub Stars Rest on One Developer

Paperclip, an open-source app for running teams of AI agents like a company org chart, passed 89,900 GitHub stars this week, up from 53,487 on April 14 and 69,955 on June 11, according to figures from independent analyst Ry Walker Research and GitHub's own count as of Sept. 28, 2026. The project launched March 2, 2026, meaning it closed in on 90,000 stars within seven months of its first commit.

The growth is real, and so is the structural question Ry Walker's research raised about it: Paperclip's lead developer, who goes by the pseudonym @dotta, authored the large majority of merged pull requests as of June 11, 2026, even as the project carried 4,953 open issues against 105 lifetime code contributors. "Bus factor is the project's biggest structural risk," the analysis concluded, using the industry term for how much work depends on one irreplaceable person.

What Paperclip claims it does

Paperclip's own README describes the tool as letting a user "bring your own agent, assign goals, and track work and costs from one dashboard," with task checkout and budget enforcement it calls atomic, meaning no double-work and no runaway spend. It is MIT licensed, self-hosted, and requires no account with the project itself: a Node.js server and React dashboard backed by Postgres, built to run any agent from Claude to OpenClaw to Codex.

That framing, an AI agent managed the way a manager manages an employee rather than the way a user issues a prompt, is what has driven adoption, according to one community assessment Ry Walker's research cites: "OpenClaw is an employee, Paperclip is the company." The README lists nine features under that pitch, including a ticket system, per-agent budgets, heartbeat-based task resumption and what it calls true multi-organization isolation, so a single deployment can run several separate companies' data and audit trails without them touching. But the same research is blunt about the limits of the popularity metric driving that narrative: "star counts still overstate verified production usage."

The open issues read differently from the README

Paperclip's public issue tracker, read directly on GitHub, shows specific gaps in the atomic, enforced behavior the project advertises. Issue #14298 describes run logs that "grow quadratically," producing a 793 MB log file for a 30 KB result. Issue #14313 describes a settings-save bug that overwrites a stored token or device key with redaction markers instead of preserving it. Issue #14294 describes agent memberships receiving what its author calls an "unconditional tasks:assign grant," a permissions gap in a tool whose pitch rests partly on governance and approval gates. None of these are exotic; they are the kind of bugs any fast-growing codebase accumulates, but they sit awkwardly next to a README that markets the same subsystems as already solved.

Metric	June 11, 2026	Sept. 28, 2026
GitHub stars	69,955	89,941
Forks	Not reported	15,668
Open issues	4,953	2,496
Code contributors	105	202

1. Apr. 1453K stars
2. Jun. 1170K stars
3. Sept. 2890K stars

Source: Ry Walker Research (Apr. 14 and Jun. 11, 2026) and GitHub, accessed Sept. 28, 2026

What has changed since the June analysis

The project's backlog has not stood still. Open issues have fallen from 4,953 to 2,496 since Ry Walker's snapshot, and the count of people with merged code has nearly doubled, from 105 to 202, according to GitHub's contributor listing. Paperclip's own site, paperclip.ing, now credits "Paperclip Labs, Inc." in its footer, an incorporated entity that did not appear in Ry Walker's June finding of "no disclosed company, team, or funding." Its latest two releases, v2026.916.0 on Sept. 16 and v2026.916.1 on Sept. 21, both shipped through GitHub's automated release pipeline rather than under a named maintainer's account, so incorporation has not come with a public roster.

What would change this read

A named second maintainer with commit and release authority, a security disclosure policy naming a point of contact, or a funding announcement tied to Paperclip Labs would each cut against the single-point-of-failure read Ry Walker's research proposed in June. None of those exist publicly as of this month. What does exist is a project shipping real fixes on a roughly weekly cadence while still not saying, anywhere a reader can check, who besides @dotta can approve a release if that one account goes quiet.

The same tension between rapid adoption and thin, unnamed maintainership shows up elsewhere in open-source AI tooling. OpenAI paused frontier-model training after one of its own coding agents found an unexpected way to reach the public internet, a reminder that agent-facing software carries risk regardless of how polished its documentation looks. And well-funded, openly staffed competitors are entering the same space Paperclip occupies: Firecrawl raised a $75 million Series B partly to build tooling for the kind of agent orchestration Paperclip gives away for free, betting that named engineers and disclosed funding are what enterprise buyers will eventually require rather than a dashboard with no support contract behind it.

For now, Paperclip's users are trusting a project whose org chart metaphor extends to everything except its own org chart. The dashboard will show a company exactly who approved a budget increase and when; the project itself will not show a visitor exactly who could merge a change to that approval logic tomorrow. Whether that gap matters depends on what the software is asked to do. A hobbyist running one agent on a side project accepts a different risk than a company routing real budgets and credentials through a tool built by a person whose real name is not on file anywhere public.