Skip to content

KramLipi Code agent

Symptom → solution.

Symptom: your PR is red, the CI log is thousands of lines, coverage gates stall merges, and review still sits on humans—while chat AI claims “fixed” with no proof. You lose nights to archaeology instead of shipping. Solution: Kramlipi AI Code Agent. It reads your git repo and the failure signal, edits with real tools, and only stops when your verify command exits 0. Coverage, bug-fix, and PR line review become measured outcomes—not guesses.

Quick start

Download Linux / macOS / Windows:

Google Drive · GitHub Releases · Release notes

chmod +x code-agent   # Linux / macOS
# Windows: code-agent.exe

2. ENV (Gemini)

Key from Google AI Studio:

export CODE_AGENT_MODEL=gemini/gemini-2.0-flash
export GEMINI_API_KEY=YOUR_SECRET_KEY

3. What do you want to do? (commands)

Increase code coverage

Pain: Coverage gate blocks merge.
Fix: Agent adds tests; your verify command must still exit 0.

code-agent run "increase unit test coverage" \
  -w /path/to/your-repo \
  --verify-cmd "go test ./..."

Python: --verify-cmd "pytest -q --cov=PACKAGE --cov-fail-under=80" → Coverage

Fix a broken build / failing tests

Pain: CI failed; log is unreadable; you need a scoped green fix.
Fix: Parse log → edit code → re-run the same CI command.

go test ./... 2>&1 | tee /tmp/ci.log
# or: pytest -q 2>&1 | tee /tmp/ci.log

code-agent experts run bug-fix \
  --log /tmp/ci.log \
  --verify-cmd "go test ./..." \
  -w /path/to/your-repo

Language walkthroughs: Python · Go · Java

Review a PR (inline line comments)

Pain: PRs merge without a first-pass review for bugs / nits / security smells.
Fix: LLM reviews the PR diff and posts GitHub inline comments (comment-only).

export GH_TOKEN=...   # or GITHUB_TOKEN in Actions

code-agent experts run code-review --pr 42 -w /path/to/your-repo
code-agent experts run code-review --pr 42 --dry-run -w /path/to/your-repo

More: Experts · Use cases

Other features

I want to… Do this Detail
Run fewer tests on a PR code-agent experts run test-intel --pr N -w . Experts
Babysit PR until green code-agent experts watch --pr N --verify-cmd "…" -w . Recipes
Find missing metrics code-agent experts run monitoring-expert -w . --dry-run Experts
Alert → reliability fix code-agent experts run sre-expert --log alert.json -w . Use cases
Canary / deploy gate code-agent experts run deploy-guard --metrics-file m.json Use cases
Flag Meaning
-w Repo folder the agent may edit
--verify-cmd Must exit 0 — success is not model opinion

Full walkthrough: Quick Start · 1-minute: Get started

Pricing

Free Team $49/mo Business $199/mo
Prove one green verify (capped) Commercial CI + PR babysit / higher volume

→ Get started — Pricing

4. Docker (second choice)

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

docker run --rm -it \
  -e CODE_AGENT_MODEL \
  -e GEMINI_API_KEY \
  -v "$PWD:/workspace" \
  ghcr.io/kramlipi/code-agent:latest \
  run "increase unit test coverage" \
  --verify-cmd "go test ./..." \
  -w /workspace

Image: ghcr.io/kramlipi/code-agent:latest · GHCR packages

After the quick commands

Next Why
Get started Binary first, use cases, pricing, how to run
Quick Start Install paths, flags, container mapping
Use cases Pain → command → benefit
Pains Full pain catalog
Commands Every CLI flag
Experts bug-fix, code-review, test-intel, …
Coverage Raise coverage with pytest-cov
Security overview Enterprise trust / data flow
Deployment modes Local binary · CI · local UI
config All config knobs

How verify works

--verify-cmd "pytest -q"
Exit Meaning
0 Success
1 Config / doctor problem
2 Agent ran but verify (or post) failed

The agent refuses to edit .github/workflows/** to cheat CI.

Site