Verify-gated CI & code review agent

Manual code review is quietly burning your best engineers.

Nits eat review time, real bugs still slip through, and a red build at midnight becomes someone's problem.

code-agent reviews every PR line-by-line and fixes failing CI automatically — proven green by your own tests, never the AI's opinion.

Get the binary — free View on GitHub

Available on

GitHub Marketplace VS Code Extension Azure DevOps Docker / any CI Try it locally

code-agent mascot — a calm, watchful reviewer

verify_cmd

exit 0 ✓

Your repo

never trains our models

The cost of doing this by hand

Manual review and manual CI triage don't scale with your team.

The 2am red build

A failing pipeline doesn't wait for business hours — and neither does the engineer who gets paged to fix it.

Every red build is someone's evening

Review theatre

Reviewers skim diffs they don't have time to truly read, so nits get comments while real bugs slide into main.

Senior hours spent on formatting, not risk

CI archaeology

Every failure means scrolling thousands of log lines just to find the one that actually matters.

Minutes lost to logs, every single run

The mechanism, not the magic

One sentence: it doesn't call itself done. Your tests do.

code-agent is a verify-gated harness, not a chat window guessing at fixes. Every run ends in one of two states: a draft PR backed by a passing test command, or an honest “could not verify.”

Read the failure

Parses your CI log, the diff, and run history — no LLM guessing at raw logs.

Plan a scoped fix

Localizes the exact file and symbol instead of rewriting half the repo.

Run your real verify command

pytest, go test, npm test — whatever already gates your CI. Retries until it passes, or stops.

Open a draft PR — only if green

No verify_cmd exit 0, no PR. Success is your test suite's opinion, never the model's.

Where it runs

Already living where your team ships code.

Customer-hosted by design — it runs on your runner, your machine, or your pipeline with your own LLM key. We never ask for your source code.

GitHub Marketplace

GitHub Action

Drop it into any workflow. It reviews PRs and fixes red CI on your own runners.

uses: kramlipi/code-agent-action@v0

View details VS Code Extension

VS Code / Cursor

Guided setup, inline fixes, and PR review right from the command palette.

ext install kramlipi.code-agent

View details Azure DevOps

Azure Pipelines extension

First-class task for organizations standardized on Azure DevOps.

task: CodeAgent@1

View details Any CI

Docker image

GitLab CI, CircleCI, Jenkins, Buildkite — anywhere a container runs.

docker pull ghcr.io/kramlipi/code-agent:latest

View details Try it locally

CLI binary

No install commitment. Point it at a repo, bring your own LLM key, see a draft PR.

code-agent run "fix failing tests" --verify-cmd "pytest -q"

View details

Why teams trust it in production

Guardrails first. Autonomy second.

Customer-hosted

Runs on your laptop, your runner, or your pipeline. We never ask you to upload your repository.

Bring your own LLM key

Gemini, Claude, OpenAI, or a local Ollama model — you control the provider and the spend.

No training on your code

Your repository contents are never used to train models. Ever.

Never edits CI workflows

GitHub Actions YAML is explicitly off-limits, so a green build always means a trustworthy build.

Design partner program: up to 10 teams get 90 days of the Business plan in exchange for one honest metric — time-to-green, CI minutes saved, or a short quote. No case-study theater, just receipts. Apply by email.

Pricing