Chainguard Checks moves CI into the AI agent development loop

CI as a tool call, not a queue

Chainguard, which describes itself as the trusted source for open source, says traditional CI has become a bottleneck in the age of AI agents. CI was built for developers who push code a few times a day, the company argues. Agents iterate constantly, so every change lands in the same queue, and the faster they write code, the longer that queue gets.

By the time results return, according to Chainguard, the agent has lost the context it was working in, and teams pay for the same setup work repeatedly. The company traces the root problem to where CI sits: outside the development loop, after the work is done.

Chainguard Checks turns CI into something that can be used as a tool call and that runs identically everywhere. Paired with the newly launched Chainguard Workspaces, an agent session suspends while checks run to halt compute consumption. It then resumes in seconds, with structured failure logs injected directly into its context so it can iterate until tests pass.

What Checks includes

Chainguard lists several components of the product:

• Warm snapshots: every CI run starts from a snapshot with dependencies already installed. Snapshots are rebuilt when the lockfile changes, so setup cost is paid once in the background rather than on every commit.
• Hardened microVM environments: each workflow step runs in its own hardware-isolated microVM built from zero-CVE Chainguard OS Packages, including FIPS-compliant toolchains, so untrusted code cannot reach other workloads.
• No credentials to steal: runners hold no long-lived credentials. A host-side proxy adds short-lived credentials to outbound requests in flight, and tools that need a credential on disk get an encrypted, short-lived stand-in.
• Network access by approval only: declarative egress policies block connections to unapproved endpoints and can automatically deny requests or destroy the microVM to prevent data exfiltration.
• Signed attestations: every run produces a signed attestation tying together the environment, test result and commit.
• Merge without re-running: merge gates can verify a passing attestation and merge without running tests again.

The security argument

Chainguard also frames the product as a security fix. Runners typically combine long-lived secrets with untrusted code, the company says, so a single prompt-injected agent or malicious install script can steal static keys straight from memory. Removing standing credentials and restricting network egress is meant to close that gap.

Every passing run generates a signed attestation that travels with the code. Downstream merge gates can verify it and merge immediately, which Chainguard says eliminates queue time and redundant compute spend.

"Agents shouldn't just write code faster. They should be able to prove it works," said Dan Lorenc, CEO and Co-founder of Chainguard. He said the company built Checks because its own engineers work alongside thousands of AI agents, and traditional CI was "the slowest and least secure part of how we build."

Partner and analyst views, and availability

Brad McCoy, Senior Manager, Cloud Platforms at Cuscal, an early design partner, said agents are exposing flaws in traditional CI systems, from security models that give runners ambient access to secrets to inefficiencies such as cold setups and delayed post-commit validation.

James Governor, Analyst and Co-founder of RedMonk, said existing CI platforms are struggling under the load from agentic software development, leading to downtime, lost developer productivity and unshipped code. He said Chainguard Checks is designed to enable true shift left security and assurance while keeping CI from being a bottleneck.

Chainguard Checks is available now in Beta. More information is at chainguard.dev/checks. For teams adopting coding agents, the release reflects a push to treat verification and supply chain trust as part of the agent's own workflow rather than a step after submission. Pricing was not given in the announcement.