A startup led by former GitHub CEO Thomas Dohmke is introducing its own platform for storing, reviewing and shipping code written by AI agents, competing with the company he used to run.

The startup, Entire, on Monday unveiled what it calls a Git forge — a service for hosting code, reviewing changes and running builds, the core of what GitHub does — designed for teams whose AI agents write code faster than people can review it. The company is pitching it as an alternative to GitHub.

The news comes two days before GitHub’s annual Universe conference in San Francisco. Microsoft acquired GitHub for $7.5 billion in 2018 and has since folded it into its CoreAI division.

The biggest change is Trails, Entire’s replacement for the pull request, the standard way developers propose and review code changes. The pull request “has become an audit log that only the agent reads,” Dohmke wrote in a blog post Monday. Instead of reading through every changed line, developers see a dashboard of scores from AI agents evaluating the code on each update.

According to Entire, multiple AI agents check the quality of each change, so developers can merge code without reading every line. Teams can also require a person’s sign-off, or passing automated tests, before any code goes in, the company said.

The launch comes as GitHub strains under the load from AI agents. The platform has had several significant outages in recent months, including one in August that lasted nearly eight hours. GitHub is moving from its own data centers to Microsoft’s Azure cloud to add capacity. GitHub said developers and agents made 7.38 billion commits, or saved code changes, on the platform in September alone.

Dohmke, who is based in Bellevue, Wash., left GitHub last year after nearly four years as CEO. Entire says it now has 50 employees. The company is fully remote.

Entire isn’t alone in rethinking code hosting for AI agents. AI coding company Cursor launched its own code-hosting service, Origin, in August. And Anthropic recently described building its own source control system as the number of code changes it merged each day grew from about 2,500 in May to more than 30,000 in September.

With the latest updates, Entire now lets developers host private code repositories directly on its service in the U.S., Europe, India and Australia, and says its service can handle up to 89 times as many code pushes per second as GitHub, based on its own benchmarks. It’s partnering with Buildkite and Depot on build and test pipelines, giving developers a way to move off GitHub Actions. Public repositories and self-hosting are on the roadmap.

Entire is also adding what it calls Continuous Memory, which lets AI agents search past coding sessions for the context behind earlier changes. The company says one part of it, a map of the codebase, cut token use by up to 71% in its testing.

That goes beyond the startup’s July launch, which mirrored existing GitHub repositories on a faster network rather than replacing them. Developers can still mirror from GitHub, or now move their code to Entire instead.

Microsoft’s venture arm M12 was among the backers of Entire’s $60 million seed round in February. Felicis led the round, with participation from Seattle-based Madrona and Basis Set Ventures, along with individual investors such as Yahoo co-founder Jerry Yang and Y Combinator CEO Garry Tan.

“Code review was always a bottleneck,” Dohmke said at a Madrona event last month. “I’ve never worked in a software company where I didn’t wait for somebody else to review my pull request, and you have to ping people in Slack three times. And there was this one guy that had to review everything because he knew what to look for. And now that bottleneck is 10 times bigger.”

Prime Intellect, a San Francisco-based AI cloud company valued at $1 billion, has acquired Moondream, a startup founded in Seattle by Amazon Web Services veterans that makes small, open-source AI models for understanding images.

Terms weren’t disclosed. Moondream has fewer than 10 employees, according to The Information, which first reported the deal Tuesday. Prime Intellect announced the acquisition later in the day in a blog post and on X.

Moondream emerged in 2024 with a $4.5 million pre-seed round from Felicis, Microsoft’s M12 GitHub Fund and Seattle-based Ascend. Its original model was small enough to run on phones and other devices.

The startup’s co-founders Jay Allen and Vik Korrapati are joining Prime Intellect to lead its inference work — running AI models after they’re trained — and to expand its efforts in vision and computer-using AI agents, the company said.

Prime Intellect is largely buying Moondream for its team’s skill at making models run faster, CEO Vincent Weisser said, according to The Information.

Moondream built Photon, an inference engine that compiles each AI model into a single program on the GPU, known as a megakernel, reducing the chip’s idle time between tasks. Engineers have historically written megakernels by hand, the report said.

“We share the conviction that specialized models can outperform much larger generalists on the tasks that matter to a business,” Weisser said in an email to GeekWire, adding that Prime Intellect will bring its evaluation, fine-tuning and reinforcement learning platform to Moondream.

Prime Intellect is one of a wave of “neoclouds” that provide computing power for AI developers and help them run open-source models. It raised $130 million in a Series A round led by Radical Ventures in July and said at the time that it had passed $100 million in annualized revenue.

Moondream was one of Prime Intellect’s earliest customers, Weisser said.

Prime Intellect will keep using the Moondream name for small, efficient AI models that handle images and other types of data, and the models will remain open, Weisser said. Its latest model, Moondream 3.1, is available through Prime Intellect’s inference service.

Moondream’s own cloud service will shut down Nov. 13, the startup said in a blog post. It will refund unspent credits and help customers move to Prime Intellect.

Korrapati, Moondream’s CTO, spent nine years at AWS. Allen, its CEO, spent seven years at AWS and previously worked at Porch, Zynga and Microsoft.

Moondream said earlier this year that it was moving to San Francisco. Korrapati announced the move in January in an online post calling AI “the biggest platform shift we’re going to see in our working lives.” Allen’s LinkedIn profile still lists him in Seattle.

Updated with comments from Prime Intellect CEO Vincent Weisser and details from Moondream’s announcement, including the planned shutdown of its cloud service.