StarSling raises $3M and brings bring-your-own-model agents to code review
Bessemer and Y Combinator backed the GitHub Actions startup, which charges for runner time while customers pay their own model providers.
By Ryan Merket · Published
Primary source: X - Yonas Beshawred
Why it matters
StarSling is turning code review into programmable CI infrastructure: customers keep control of models and review logic, while the startup earns revenue from the compute beneath every automated pass.
Yonas Beshawred (@yonasbe) and Daniel Worku (@dbworku) have raised a $3 million pre-seed round for StarSling, their San Francisco developer infrastructure startup, and launched code review agents that run inside GitHub Actions using a customer's own model account.
https://x.com/yonasbe/status/2102443015583224224
The round, announced on September 22nd, includes Bessemer Venture Partners and Y Combinator, along with Vermilion Cliffs Ventures, Precursor Ventures, Cervin Ventures, Outset Capital and Transpose Platform. StarSling said it will spend the capital on compute and additional founding engineers in the San Francisco Bay Area.
The financing also includes angel checks from Sentry co-founder David Cramer, former GitHub engineer Zach Holman, Rocket Money co-founder and CTO Idris Mokhtarzada and Phosphor Capital founder Kulveer Taggar. StarSling said engineering and product leaders from OpenAI, Anthropic, ClickHouse, Google, Microsoft, Meta, Coinbase and Shopify also participated.
Beshawred previously founded StackShare, the developer community and enterprise software business acquired by FOSSA in 2024. Worku led the 12-person engineering group behind Netflix Console, an internal developer portal used across Netflix, after working on Facebook's AI Camera team. The founders have known each other for over a decade and started StarSling in 2025 before joining Y Combinator's Spring 2025 batch.
Their latest product, StarSling Review Runners, extends the startup beyond the continuous integration infrastructure it introduced in April. The product lets engineering teams define separate reviewers for areas such as security, APIs, tests, databases or frontend architecture. Each reviewer can use repository-specific instructions, scripts and skills.
Customers bring an API key for their chosen model provider. StarSling supplies the compute that runs the review job, starting at $0.004 per minute for a two-vCPU runner, while the model provider bills the customer directly for tokens. That structure lets StarSling sell infrastructure without reselling model usage or adding another per-seat AI subscription.
The team's review configuration stays in the repository and is versioned with the code. StarSling says instructions and skills are loaded from the trusted base branch before the model runs, preventing code submitted in a pull request from rewriting the reviewer evaluating it. The model receives a read-only token, and a separate step publishes the resulting review.
Each agent produces one structured comment containing a verdict, findings organized by severity, file and line references, and proposed fixes. If a developer pushes another commit while a review is running, StarSling cancels the old pass so the next completed review covers the latest version.
Moving from faster CI to automated verification
Review Runners broaden the pitch Beshawred and Worku have been developing around StarSling's original product. StarSling Runners replace GitHub Actions' default Ubuntu runners and use agents to inspect workflows, logs and machine telemetry. Those agents can then open pull requests that change caching, dependency installation, test sharding and build parallelization.
StarSling says its infrastructure has processed more than 2 million CI jobs. It also claims its runners and optimization agents have saved customers including Mastra and Better Auth more than 28,000 hours of compute. Those figures are company-reported, although StarSling has published examples of agent-generated pull requests and customer-specific measurements.
Among those examples, StarSling says Mastra cut one test suite's wall-clock time from 29 minutes and 56 seconds to 5 minutes and 6 seconds. Better Auth reduced an end-to-end suite from 2 minutes and 22 seconds to 1 minute and 4 seconds, while Partcl reduced its typical runner queue time from 9 minutes and 30 seconds to 35 seconds.
The code review launch gives StarSling a second place to charge within the pull request workflow. Coding agents can produce larger volumes of changes, but those changes still have to clear tests and human review. StarSling is betting that engineering teams will want automated reviewers built from their existing scripts and internal standards, while retaining control of the underlying model and its token bill.
Bessemer partner Elliott Robinson tied the investment to the founders' experience operating developer systems at scale. In StarSling's funding announcement, Robinson said the firm backed Beshawred and Worku's plan to bring AI agents into DevOps infrastructure after watching coding tools advance faster than the systems used to verify and ship their output.
Review Runners opened in early access on September 22nd. StarSling's existing CI runners remain available through its GitHub App, with the new review product billed through the same runner account.