AI NewsDev toolsAnnouncement

StarSling launches CI that opens its own speed-up pull requests

StarSling is a Y Combinator R7B company. It launched on Hacker News with a GitHub Actions replacement that opens its own pipeline speed-up pull requests.

AI News

Editorial2 min read

LinkedInX
Reveneau illustration of a continuous integration pipeline

Why it mattersCoding agents are now the heaviest users of CI, not people. A pipeline that watches itself is a different product from GitHub Actions.

A team using Claude Code or Cursor does not open five pull requests a day. It opens fifty. Standard CI was not built for that volume. The queue is now the slowest step of the week for many engineering teams. StarSling is a Y Combinator R7B company. It launched on Hacker News with a direct answer to that volume problem.

The product has two parts. The first is a faster runner. A team changes one line in a workflow file. That line goes from runs-on: ubuntu-latest to runs-on: starsling-ubuntu-26.04. Jobs then run on 5th Gen AMD EPYC hardware. StarSling says the swap is up to ten times faster than GitHub's default Ubuntu runner.

What the agent part does

The second part is where the "self-improving" claim comes from. A repository connects to StarSling. An AI agent reads workflow files, logs and telemetry. It opens pull requests that cache dependencies, split tests across machines or rework job structure. The team reviews each pull request like any other. StarSling says the pipeline gets faster the longer a repository uses it, because every merged suggestion stays merged.

Three named customers carry specific numbers on the launch page. Better Auth's end-to-end tests went from 2 minutes 22 seconds to 1 minute 4 seconds. StarSling calls this two times faster. Mastra's test suite went from 29 minutes 56 seconds to 5 minutes 6 seconds. That is six times faster. Partcl's queue time went from 9.5 minutes to 35 seconds, which the page reports as sixteen times shorter. These are the company's numbers from its own customers. A reader takes them as such.

Billing and limits

Pay-as-you-go pricing is $0.008 per minute, billed per second. StarSling says that is 33 percent cheaper than GitHub Actions at $0.012 per minute. Queue time is not billed at all. A free trial gives 2,000 minutes a month with no card. The company also offers NVIDIA RTX 4090, 5090 and PRO 6000 GPU runners for machine learning work. It offers a persistent disk option it claims is 4.5 times faster than GitHub's own caching. Total CI jobs run on the platform are over three million.

What a buyer should check

StarSling sits between GitHub Actions and a team's own control plane. A workflow change is cheap. An agent that opens pull requests against the pipeline is not. It needs access to the repository and permission to edit workflow files. A team should treat the first few suggestions like a new contractor. Read each pull request. Reject what is wrong. Watch for a run that passes the test suite and fails in production. That is a known failure for every CI optimiser, agent or not.

The pitch is the shape of a real change in how software teams work. Agents write the code now. The queue that reviewed and released a person's output every two weeks has to carry fifty merges a day. A pipeline that watches itself and reopens for speed is a different product from the one GitHub released in 2019. The Hacker News discussion and the named customers are a first read on whether it holds up.

Source

StarSling, launched on Hacker News.

SourceStarSling

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX
Start a project