Competitor intelligence for SaaSCase studies · updated Nov 3, 2026
AdRadarStart free

AdRadar / Cases / OpenClaw

Open-source agent · 2025–26

An agent prototype "prompted into existence" became GitHub's most-starred project in months

Peter Steinberger's OpenClaw, first Clawdbot, passed 145,000 GitHub stars by early February 2026 and led to a job offer from OpenAI.

Evidence: MediumFortune is a credible outlet; its key numbers appear without a named source.
145K+GitHub stars by early Feb 2026Fortune
2Mvisitors in the peak weekFortune
~1 hto the first prototypeFortune lede
Up to $10K/molost on server costsSteinberger, per Fortune

At a glance

Builder
Peter Steinberger, an Austrian developer who founded and bootstrapped PSPDFKit.
Built with
Prompting AI agents. Fortune names no tools or models and gives no share of AI-written code.
Product
A local-first autonomous agent acting as a "digital employee" via WhatsApp or Telegram, with memory in Markdown files.
Business model
Open source. After joining OpenAI, the project moves to a foundation OpenAI supports.
Names
Clawdbot, then Moltbot, then OpenClaw, after a naming disagreement with Anthropic.

Timeline

  1. April 2025A simple Twitter-analysis tool convinces him AI can handle "the plumbing of code".
  2. Late 2025 (per secondary sources)Prototype released as Clawdbot; Fortune says it took about an hour to build.
  3. UndatedRenamed Moltbot, then OpenClaw, after a disagreement with Anthropic; his accounts are briefly hijacked by crypto-linked actors.
  4. Early Feb 2026Passes 145,000 GitHub stars.
  5. 15 Feb 2026Steinberger announces he is joining OpenAI.
  6. 19 Feb 2026Fortune profile published.

How it was built

  1. 1He was annoyed the product did not exist, so he "just prompted it into existence" (Lex Fridman podcast).
  2. 2A local-first design keeps the agent on the user's own hardware.
  3. 3Steinberger says he nearly deleted the project at one point.

Why it took off

  1. 1A concrete promise: an always-on assistant that runs on your machine.
  2. 2Open source with a personal backstory attracted press and developers at once.
  3. 3Controversy and rebranding kept it in the news.
  4. 4A creator with 13 years running PSPDFKit and a reported exit behind him.
Be careful with

Read the numbers with these caveats.

  • Security: autonomous agents with access to email and browsers carry risk. Fortune links a separate piece on this.
  • Account hijacking during the rename shows what can go wrong on a fast path to fame.
  • Fortune's "six-figure offer" is speculation and the "44th project" claim is not supported by its own link.
  • Server costs were a real drain: up to $10K a month.
§ EvidenceWhat the sources say
Confirmed by a source we read
  • Fortune article exists (19 Feb 2026, by Eva Roytburg) with the quotes and claims above.
Reported figures
  • 145K+ stars, 2M weekly visitors, one-hour prototype.
  • Multiple job offers including outreach from Mark Zuckerberg.
Sources disagree
  • Other sites cite higher star counts for later dates.
§ Check it yourselfWith AdRadar

Don’t take the headline on faith. Watch OpenClaw yourself.

See what OpenClaw is doing in public: its ads, the pages they lead to, how its site changes and what people say about it.

Partial fit. An open-source project rarely buys ads. Mentions and site pages are the useful parts.

Ads

Archives what a company runs on Meta, Google and LinkedIn: new, stopped and long-running ads.

Check whether companies building on or competing with OpenClaw advertise against it.

Landing pages

Reads the pages its active ads link to and flags headline, offer and price changes.

Landing tracking only fills when active ads link to pages.

Site pages

Reads the public sitemap: new and updated pages, page types, publishing rhythm.

Follow the docs and site sitemap for new pages after the move to a foundation.

Reviews and mentions

Collects Trustpilot reviews and Reddit mentions, and can summarise the complaints.

Read Reddit mentions on setup problems, security worries and what people run.

Track OpenClaw in AdRadar

You can repeat this.

The parts that do not depend on a famous founder.

  1. 1Open-source the thing you wish existed, with a one-line promise.
  2. 2Tell the origin story publicly and keep it honest.
  3. 3Budget for infrastructure costs before a viral spike.
  4. 4Plan names, handles and trademarks before you go viral.
Watch your competitors in AdRadarRead another case

Sources

← Previous caseDualShot RecorderNext case →Moltbook