🐶 Codedog

Review your Github/Gitlab PR with ChatGPT

Codedog is update to langchain v0.2

What is codedog?

Codedog is a code review automation tool benefit the power of LLM (Large Language Model) to help developers review code faster and more accurately.

Codedog is based on OpenAI API and Langchain.

Quickstart

Review your pull request via Github App

Install our github app codedog-assistant

Start with your own code

As a example, we will use codedog to review a pull request on Github.

0. Install codedog

pip install codedog

codedog currently only supports python 3.10.

1. Get a github pull request

from github import Github

github_token="YOUR GITHUB TOKEN"
repository = "codedog-ai/codedog"
pull_request_number = 2

github = Github(github_token)
retriever = GithubRetriever(github, repository, pull_requeest_number)

2. Summarize the pull request

Since PRSummaryChain uses langchain's output parser, we suggest to use GPT-4 to improve formatting accuracy.

from codedog.chains import PRSummaryChain

openai_api_key = "YOUR OPENAI API KEY WITH GPT4"

# PR Summary uses output parser
llm35 = ChatOpenAI(openai_api_key=openai_api_key, model="gpt-3.5-turbo")

llm4 = ChatOpenAI(openai_api_key=openai_api_key, model="gpt-4")

summary_chain = PRSummaryChain.from_llm(code_summary_llm=llm35, pr_summary_llm=llm4, verbose=True)

summary = summary_chain({"pull_request": retriever.pull_request}, include_run_info=True)

print(summary)

3. Review each code file changes in the pull request

review_chain = CodeReviewChain.from_llm(llm=llm35, verbose=True)

reviews = review_chain({"pull_request": retriever.pull_request}, include_run_info=True)

print(reviews)

4. Format review result

Format review result to a markdown report.

from codedog.actors.reporters.pull_request import PullRequestReporter

reporter = PullRequestReporter(
    pr_summary=summary["pr_summary"],
    code_summaries=summary["code_summaries"],
    pull_request=retriever.pull_request,
    code_reviews=reviews["code_reviews"],
)

md_report = reporter.report()

print(md_report)

Deployment

We have a simple server demo to deploy codedog as a service with fastapi and handle Github webhook. Basicly you can also use it with workflow or Github Application.

see examples/server.py

Note that codedog don't have fastapi and unicorn as dependency, you need to install them manually.

Configuration

Codedog currently load config from environment variables.

settings:

How to release