AI Code Review + CI Repair | Kramlipi — Try Free

Verify-gated AI agents

AI code review that also
turns red builds green.

Kramlipi reviews pull requests, repairs broken CI with a verify gate, adds missing tests, catches flaky builds, and proves every change with the commands your team already trusts.

✓ Free local tier ✓ No signup ✓ No credit card ✓ Your runner, your code

Get started today See how it works

Bring your own Gemini, Claude, OpenAI, or local model key.

pull request #184 · review

Checkout validation AI review

71if amount > balance:

72 return approve(order)

72 return decline(order)

73return approve(order)

K kramlipi-code-review[bot]

This branch approves an order when funds are insufficient. Reverse the result and add the boundary case before merge.

✓ verify_cmd: exit 0

↑ coverage: +4.8%

Proof before “done”

Use it as

GitHub Action Binary Web Chat CLI Chat VS Code Azure DevOps GitLab CI CircleCI Docker AWS / GCP

Pain in. Verified outcome out.

One agent loop for the work slowing every release.

Each feature starts with a developer pain and ends with evidence your team can inspect.

PR #1842 · checkout boundary first-pass review

▲ Potential issue · kramlipi

Validate non-positive amounts before the gateway call so declined edges stay deterministic in CI.

See a sample review →

CI run #2481 verify_cmd exit 0

test_checkout_boundary AssertionError: expected declined

00:42

→ localized src/checkout.py:72
→ added boundary regression test
→ running pytest -q

pytest -q 214 passed · 0 failed · draft PR ready

01:18

See the repair loop →

coverage-growth · unit tests +3 focused tests

test_charge_rejects_zero generated from missed branch · checkout.py:69

new

test_charge_rejects_negative generated from missed branch · checkout.py:69

new

pytest -q tests/test_checkout.py real suite run after write

green

See test generation →

Coverage risk on this diff risk branches

Map changed symbols → missing branches
Target meaningful paths, not vanity %

See coverage targeting →

Diff-aware test selection 12 / 214 selected

Changed files + coverage hints → impacted tests
Safe full-suite fallback when confidence is low

See selection logic →

Flake evidence · same SHA unstable

ci.yml · run #2479 commit a1c3e92 · failed once

fail

ci.yml · run #2480 commit a1c3e92 · passed on retry

pass

Same workflow + same commit = flake signal
Surfaced in UI — not silently rerun forever

See flake evidence →

Pain: reviews queue for hours

Ship faster with first-pass PR review.

Grounded scanners and an AI critic add focused inline comments before senior engineers arrive.

Pain: red CI blocks the team

Move broken builds from red to green.

The bug-fix agent reads the failure, changes code, and opens a draft only after your verify command passes.

Pain: fast code lowers confidence

Protect quality with tests added automatically.

Coverage-growth and test-intel agents identify missed behavior, generate focused tests, and run the real suite.

Pain: coverage quietly falls

Increase coverage where the diff creates risk.

Map changed symbols to tests and target meaningful branches instead of inflating a vanity percentage.

Pain: every PR runs everything

Run tests for the code that actually changed.

Diff-aware selection narrows work safely, reducing avoidable queue time and compute while keeping a fallback.

Pain: flaky tests waste reruns

Detect flakes instead of normalizing retries.

Same-workflow, same-commit pass/fail evidence flags unstable builds and surfaces them in the UI.

Target: up to 30% fewer wasted CI minutes through diff-aware selection and flake reduction. Actual savings depend on repository shape and must be measured from runner data.

The mechanism, not magic

The AI does not decide that it succeeded. Your tools do.

Kramlipi keeps working until the repository’s objective command passes—or stops honestly without publishing a false win.

Read the evidenceCI logs, changed files, repository rules

grounded

Make a scoped changeSmall edit inside the workspace jail

focused

Run your real checkpytest, npm test, go test, Maven, or your command

exit 0

Share the receiptInline review or verified draft pull request

reviewable

Broken build agent

Wake up to a verified draft—not another wall of logs.

Point the agent at the failed run. It finds the relevant failure, edits only what is needed, and reruns your trusted command before it raises a pull request.

✓ Parses known CI failures before asking the model.

✓ Never turns off tests or edits CI to manufacture green.

✓ Publishes only after objective verification succeeds.

CI run #2481 Draft PR ready

test_checkout_boundaryAssertionError: expected declined

00:42

→ localized src/checkout.py:72
→ added boundary regression test
→ running pytest -q

✓

pytest -q214 passed · 0 failed

exit 0

↗

fix/checkout-boundaryDraft pull request opened

#185

Test intelligence

Grow useful coverage and spend CI where the change creates risk.

The agent connects changed files to tests, highlights untested behavior, and keeps flaky results separate from real regressions.

✓ Selects impacted tests with a safe full-suite fallback.

✓ Adds regression tests around the behavior being changed.

✓ Flags same-commit pass/fail evidence as a likely flake.

Diff intelligence 5 files changed

82%+4.8% this PR

tests/test_checkout.pyselected ✓

tests/test_discounts.pyselected ✓

tests/test_admin.pyskipped safely

Meet the agent where you work

Binary, browser, terminal, editor, or CI marketplace.

Start locally in minutes, then move the same verify-gated loop into your team’s delivery system.

terminal · code-agent setup ok

$ code-agent setup

↓ downloading linux binary…

✓ binary installed → ~/.local/bin/code-agent

✓ provider key detected · free local tier

✓ workspace ready · no signup

$ code-agent doctor

all checks passed

Available

Local binary

Download one file and run on your repository. The free local tier needs no account or card.

code-agent setup Get started →

localhost:8787 / chat verifying

Fix the red checkout test and open a draft.

Kramlipi · web Reading CI log → editing checkout.py → running pytest -q

✓

verify_cmd 214 passed · draft PR ready

exit 0

Available

Web Chat

Pick a repository, chat with the agent, watch verification, and open a draft from the browser.

code-agent web serve -w . Open web chat →

zsh · agent loop inspect → edit → verify

$ code-agent run “fix failing tests”

observe parsed AssertionError in test_checkout

edit src/checkout.py:72 · +boundary guard

verify pytest -q

✓ green · writing draft PR summary

Available

CLI Chat

Describe the goal in plain language and keep the whole inspect-edit-verify loop in your terminal.

code-agent run "fix failing tests" Run in terminal →

GitHub Actions · PR #1842 checks

ci.yml / unit failed · named workflow trigger

red

→

kramlipi/code-agent-action bug-fix expert · verify-gated

run

✓

Draft PR opened only after verify_cmd exit 0

green

Available

GitHub Action

Review every pull request or start a verified bug-fix job when your named CI workflow fails.

uses: kramlipi/code-agent-action@v0 See Action →

VS Code · Kramlipi extension

> Kramlipi: Repair CI

Kramlipi: Repair CI

Kramlipi: Review Diff

Kramlipi: Grow Coverage

Kramlipi: Doctor

Package

VS Code extension

Run doctor, repair CI, review a diff, and grow coverage from the command palette.

ext install kramlipi.code-agent Install extension →

CI marketplaces customer-hosted

Azure DevOps GitLab CI CircleCI AWS GCP GKE Docker any shell

✓

Same verify-gated loop your runner · your LLM key · no code host

BYO

Packages

Every major CI marketplace

Azure DevOps, GitLab, CircleCI, AWS containers, Google Cloud GKE, Docker, and any shell runner.

customer-hosted · BYO LLM See deployment modes →

“Available” means directly usable today. “Package” means a distribution package exists; public marketplace approval may still depend on the marketplace operator.

Resources

Watch it. Read it. Run it.

Choose the fastest route for the way you learn.

Safe enough for real repositories.

Guardrails first. Autonomy second. Objective verification always.

✓ Customer-hosted

Runs on your machine or runner.

✓ No model training

Your repository is not our training set.

✓ Workspace jail

File access remains inside the selected repo.

✓ Fail closed

No verified pass means no success claim.

Make your next review faster—and your next green build real.

Free local tier · no signup · no credit card · bring your own model key.

Get started today →

AI code review that also
turns red builds green.

One agent loop for the work slowing every release.

Ship faster with first-pass PR review.

Move broken builds from red to green.

Protect quality with tests added automatically.

Increase coverage where the diff creates risk.

Run tests for the code that actually changed.

Detect flakes instead of normalizing retries.

The AI does not decide that it succeeded. Your tools do.

Wake up to a verified draft—not another wall of logs.

Grow useful coverage and spend CI where the change creates risk.

Binary, browser, terminal, editor, or CI marketplace.

Local binary

Web Chat

CLI Chat

GitHub Action

VS Code extension

Every major CI marketplace

Watch it. Read it. Run it.

How-to videos

Documentation

Get started

Safe enough for real repositories.

Make your next review faster—and your next green build real.

AI code review that also turns red builds green.

One agent loop for the work slowing every release.

Ship faster with first-pass PR review.

Move broken builds from red to green.

Protect quality with tests added automatically.

Increase coverage where the diff creates risk.

Run tests for the code that actually changed.

Detect flakes instead of normalizing retries.

The AI does not decide that it succeeded. Your tools do.

Wake up to a verified draft—not another wall of logs.

Grow useful coverage and spend CI where the change creates risk.

Binary, browser, terminal, editor, or CI marketplace.

Local binary

Web Chat

CLI Chat

GitHub Action

VS Code extension

Every major CI marketplace

Watch it. Read it. Run it.

How-to videos

Documentation

Get started

Safe enough for real repositories.

Make your next review faster—and your next green build real.

AI code review that also
turns red builds green.