CHUNK SIDECARS — BY CIRCLECI
Stop wasting tokens waiting for feedback on your agent's code.
CI is too expensive for the frequent small code mistakes that agents make. Chunk sidecars spin up and run a CI-like microbuild in seconds.
- Two commands to get started
- Works directly in your agent's inner loop, so they can fix errors using warm context
- Mirrors your CI stack for accurate results
brew install CircleCI-Public/circleci/chunk && chunk init
SEE IT IN ACTION
27 seconds. Real validation. No commit required.
27s
Average microbuild run time
78%
Faster than a full CI pipeline run
5x
Fewer tokens per agent cycle
375h
Saved per 5K builds
Early results from internal testing and early access customers.
HOW IT WORKS
Three commands to a faster feedback loop
01
Install the Chunk CLI
Install via Homebrew in seconds. Chunk auto-detects your stack, test commands, and build system; no manual configuration.
brew install CircleCI-Public/circleci/chunk
02
Initialize your project
Run chunk init in your project root. It configures your validation hooks, wires up your agent's settings, and connects to CircleCI.
03
Let your agent validate
Run the chunk-sidecar skill in Claude Code, Cursor, or any agent. It syncs, validates, fixes failures, and hands back clean code: no developer intervention needed.
THE PROBLEM
AI speeds up the inner loop. Issues get caught after agents have moved on.
AI development broke the balance between local iteration and remote validation. Agents commit fast; CI feedback arrives but the context window has closed. The result: flooded pipelines, wasted tokens, and a growing gap between activity and delivery.
Chunk sidecars restore the balance. Lightweight checks run in the inner loop, where agents are actively working, so feedback arrives while the change is still in motion.
Start validating in 60 seconds
Free for all CircleCI users. Agent-agnostic: works with Claude Code, Codex, Cursor, or anything you've built.
brew install CircleCI-Public/circleci/chunk && chunk init