Get started¶

KramLipi Code agent¶

Increase your code coverage and review automatically.

Developers ship code faster — but unit tests and failed builds pile up. This agent fixes that in the CI pipeline (and locally).

This is the only start page (Quick Start merged here).

How kramlipi helps (collage)¶

kramlipi three jobs: fix the build, more coverage, smart on the go

1/4 Sound familiar — broken build, chatbot says fixed, still broken, babysitting

2/4 How the agent helps — failing build, tries fixes, stops when your test passes

3/4 What you get — build MR, coverage MR, pipeline or laptop

4/4 No more babysitting — try free at get-started

Story in four panels: pain → how it helps → what you get → no babysitting.
Share the same images on LinkedIn / X (captions in the product repo docs/SOCIAL-COLLAGE-SERIES.md).

1. Binary first¶

Google Drive — linux/ · macos/ · windows/

Also: GitHub Releases

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

2. ENV (pick a model + API key)¶

Set CODE_AGENT_MODEL (LiteLLM string) and the matching provider key.

Gemini (default / recommended)¶

Key: Google AI Studio

export CODE_AGENT_MODEL=gemini/gemini-3.1-flash-lite
# also fine: gemini/gemini-2.0-flash
export GEMINI_API_KEY=YOUR_SECRET_KEY

PowerShell (Windows):

$env:CODE_AGENT_MODEL="gemini/gemini-3.1-flash-lite"
$env:GEMINI_API_KEY="YOUR_SECRET_KEY"

Claude¶

export CODE_AGENT_MODEL=anthropic/claude-sonnet-4-20250514
export ANTHROPIC_API_KEY=YOUR_SECRET_KEY

Key: Anthropic Console

OpenAI¶

export CODE_AGENT_MODEL=openai/gpt-4o
export OPENAI_API_KEY=YOUR_SECRET_KEY

Release binaries: Free activates on the first metered run. Set CODE_AGENT_LICENSE_KEY for Team/Business.

Cursor?¶

No public Cursor model API for code-agent — use Gemini / Claude / OpenAI. Full list: Quick Start.

Other providers (optional)¶

export CODE_AGENT_MODEL=deepseek/deepseek-chat
export DEEPSEEK_API_KEY=YOUR_SECRET_KEY

export CODE_AGENT_MODEL=openrouter/anthropic/claude-3.5-sonnet
export OPENROUTER_API_KEY=YOUR_SECRET_KEY

export CODE_AGENT_MODEL=openai/my-model
export CODE_AGENT_API_BASE=https://your-proxy.example/v1
export CODE_AGENT_API_KEY=YOUR_PROXY_KEY

Variable	Required when
`CODE_AGENT_MODEL`	Always (or config default)
`GEMINI_API_KEY`	`gemini/…`
`ANTHROPIC_API_KEY`	`anthropic/…`
`OPENAI_API_KEY`	`openai/…`
`CODE_AGENT_ECONOMY_MODE`	Optional — `true` to save cost (default off)

3. What do you want to do?¶

Increase code coverage¶

Pain: Coverage / missing tests block merge.

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

Auto-detect picks the verify command from your repo (Auto verify). Override only if needed: --verify-cmd "go test ./...".

→ Use cases → Coverage · Features → Coverage

Fix a broken build¶

Pain: CI red; log archaeology at midnight.

go test ./... 2>&1 | tee /tmp/ci.log

code-agent experts run bug-fix \
  --log /tmp/ci.log \
  -w /path/to/your-repo

Verify is auto-detected from the project (Auto verify). Add --verify-cmd "…" only to pin the same command as CI.

→ Use cases → Fix CI · Python · Go · Java

Review a PR (inline comments)¶

Pain: No first-pass review on every PR.

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

CI copy-paste (GitHub / GitLab / Azure): Code review CI

Economy mode is off by default. Opt in: CODE_AGENT_ECONOMY_MODE=true.

git diff main...HEAD > /tmp/mr.diff
code-agent experts run code-review --diff-file /tmp/mr.diff -w .

→ Use cases · Features → code-review

Other features¶

I want to…	Command	Detail
Impacted tests only	`experts run test-intel --pr N`	Use cases → test-intel
Flaky tests — memory, skip, admin	`web serve` → flake panel ⚙	Flaky test admin
Babysit PR	`experts watch --pr N --verify-cmd "…"`	Use cases → watch
Missing metrics	`experts run monitoring-expert --dry-run`	Features → experts
Alert → fix	`experts run sre-expert --log alert.json`	Use cases → SRE

Flag	Meaning
`-w`	Repo path
`--verify-cmd`	Must exit `0`

Rule: success = verify exits 0, not model opinion.

4. Docker (optional)¶

Quick `docker run`¶

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

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

Browser UI (~60 seconds)¶

export GEMINI_API_KEY=your-key
curl -fsSL -o docker-ui.sh \
  https://gist.githubusercontent.com/kramlipi/d31f4f454cd127cfb552e5ed5e854af3/raw
chmod +x docker-ui.sh
bash docker-ui.sh
# → http://127.0.0.1:8080

Windows gist: docker-ui.ps1

Pricing (adoption + revenue)¶

Plan	Price	You get
Community (Free)	$0	Prove green verify locally; capped runs; 14-day commercial eval
Team — CI Repair	$49/mo	Commercial CI rights, higher caps, ~5 seats
Business — PR Babysit	$199/mo	+ `experts watch`, more seats/volume

Contact / keys: cluevion@gmail.com

How to run (three modes)¶

Mode	When
Local binary	Laptop fix / coverage / dry-run review
CI (Actions / GHCR)	Every PR or failed workflow — `ghcr.io/kramlipi/code-agent` or binary on runner
Local UI	Browser demo — Docker UI gist above

Enterprise trust (short)¶

Runs on your machine or your CI with your LLM key. We don’t host your monorepo.

Exit codes¶

Code	Meaning
`0`	Success
`1`	Config / startup / doctor failure
`2`	Agent ran but task or verify failed

Next¶


Auto verify (no `--verify-cmd` guesswork)	Auto verify
Flaky tests (storage + skip)	Flaky test admin
PR review + linters	Code review CI · Linter sandbox
Pain → command	Use cases · Pains catalog
All flags & experts	Features
Home	index