argus-review-code 0.1.0

ArgusReview (sang-hv fork) is an AI-powered code review tool for GitHub, GitLab, Bitbucket Cloud, Bitbucket Server, Azure DevOps and Gitea

ArgusReview

AI-powered code review tool.

This is a personal fork (sang-hv/ai-code-review) of ArgusReview (originally ai-review by Nikita Filonov, rebranded by deha-project). See NOTICE for full attribution.

---

📑 Table of Contents

• ✨ About
• 🧪 Live Preview
• 🚀 Quick Start
• ⚙️ ️CI/CD Integration
  • 🚀 GitHub Actions
  • 🚀 GitLab CI/CD
• 📘 Documentation
• ⚠️ Privacy & Responsibility Notice

---

✨ About

ArgusReview is a developer tool that brings AI-powered code review directly into your workflow. It helps teams improve code quality, enforce consistency, and speed up the review process.

✨ Key features:

• Multiple LLM providers — choose between OpenAI, Claude, Gemini, Ollama, Bedrock, OpenRouter, or Azure OpenAI and switch anytime. Plus a generic OpenAI-compatible provider and a 9Router preset, so any OpenAI-style gateway (LMRouter, vLLM, LocalAI, LiteLLM, ...) works by just setting a base URL.
• Project coding conventions — feed your team's standards (Markdown) into every review from a local folder, a raw URL, or a git repo (public or private). The AI reviews against your rules, not just generic defaults.
• VCS integration — works out of the box with GitLab, GitHub, Bitbucket Cloud, Bitbucket Server, Azure DevOps, and Gitea.
• Customizable prompts — adapt inline, context, and summary reviews to match your team's coding guidelines.
• Agent mode — iterative ReAct-style loop where the model can explore the repository with shell commands (ls, cat, rg, git) before producing a final review, giving it deeper context than a single-shot call.
• Reply modes — AI can now participate in existing review threads, adding follow-up replies in both inline and summary discussions.
• Flexible configuration — supports YAML, JSON, and ENV, with seamless overrides in CI/CD pipelines.
• ArgusReview runs fully client-side — it never proxies or inspects your requests.

ArgusReview runs automatically in your CI/CD pipeline and posts both inline comments, summary reviews, and now AI-generated replies directly inside your merge requests. With agent mode enabled, the model can autonomously explore the codebase before reviewing, resulting in more accurate and context-aware feedback. This makes reviews faster, more conversational, and still fully under human control.

---

🧪 Live Preview

Curious how ArgusReview works in practice? Here are real Pull Requests reviewed entirely by the tool — one per mode:

Mode	Description
🧩 Inline	Adds line-by-line comments directly in the diff. Focuses on specific code changes.
🧠 Context	Performs a broader analysis across multiple files, detecting cross-file issues and inconsistencies.
📄 Summary	Posts a concise high-level summary with key highlights, strengths, and major issues.
💬 Inline Reply	Generates a context-aware reply to an existing inline comment thread. Can clarify decisions, propose fixes, or provide code suggestions.
💬 Summary Reply	Continues the summary-level review discussion, responding to reviewer comments with clarifications, rationale, or actionable next steps.

👉 Each review was generated automatically via GitHub Actions using the corresponding mode:

argus-review run-inline
argus-review run-summary
argus-review run-context
argus-review run-inline-reply
argus-review run-summary-reply

---

🚀 Quick Start

Install via pip:

pip install argus-review-code

📦 Package name on PyPI: argus-review-code (the CLI command is still argus-review)

Or install from source:

git clone https://github.com/sang-hv/ai-code-review.git
cd ai-code-review
pip install .

Or build the Docker image locally:

docker build -t ai-code-review:latest .
docker run --rm -v $(pwd):/app ai-code-review:latest run-summary

👉 Before running, create a basic configuration file .ai-review.yaml in the root of your project:

llm:
  provider: OPENAI

  meta:
    model: gpt-4o-mini
    max_tokens: 1200
    temperature: 0.3

  http_client:
    timeout: 120
    api_url: https://api.openai.com/v1
    api_token: ${OPENAI_API_KEY}

vcs:
  provider: GITLAB

  pipeline:
    project_id: "1"
    merge_request_id: "100"

  http_client:
    timeout: 120
    api_url: https://gitlab.com
    api_token: ${GITLAB_API_TOKEN}

👉 This will:

• Run ArgusReview against your codebase.
• Generate inline and/or summary comments (depending on the selected mode).
• Use your chosen LLM provider (OpenAI GPT-4o-mini in this example).

Note: Running argus-review run executes the full review (inline + summary). To run only one mode, use the dedicated subcommands:

• argus-review run-inline
• argus-review run-context
• argus-review run-summary
• argus-review run-inline-reply
• argus-review run-summary-reply

---

ArgusReview can be configured via .ai-review.yaml, .ai-review.json, or .env. See ./docs/configs for complete, ready-to-use examples.

Key things you can customize:

• LLM provider — OpenAI, Gemini, Claude, Ollama, Bedrock, OpenRouter, Azure OpenAI, any OpenAI-compatible endpoint, or 9Router
• Model settings — model name, temperature, max tokens
• VCS integration — works out of the box with GitLab, GitHub, Bitbucket Cloud, Bitbucket Server, Azure DevOps, and Gitea
• Agent mode — enable iterative repository exploration before review
• Coding conventions — load your team's .md standards from a local folder, a URL, or a git repo
• Review language — have the AI write comments/summaries/replies in any language (English, Vietnamese, ...)
• Review policy — which files to include/exclude, review modes
• Prompts — inline/context/summary/agent prompt templates

---

🔌 OpenAI-compatible providers & 9Router

Any gateway that speaks the OpenAI Chat Completions API works with the OPENAI_COMPATIBLE provider — you only set a base URL and a model:

llm:
  provider: OPENAI_COMPATIBLE
  meta:
    model: meta-llama/llama-3.1-70b-instruct
  http_client:
    api_url: https://my-router.example.com/v1
    api_token: ${LLM_API_TOKEN}   # optional — omit for token-less local gateways

For the 9Router local proxy, use the 9ROUTER preset. api_url defaults to http://localhost:20128/v1, so config stays minimal:

llm:
  provider: 9ROUTER
  meta:
    model: gpt-4o-mini

---

📐 Project coding conventions

Every team has its own standards. Point ArgusReview at your convention docs and they get injected into each review prompt automatically. Sources can be a local file/folder, a raw URL, or a git repo — private ones supported via token:

conventions:
  enabled: true
  sources:
    # Local file or folder of .md docs
    - type: local
      path: ./docs/conventions

    # Raw Markdown from a URL (token optional, for private URLs)
    - type: url
      url: https://raw.githubusercontent.com/my-org/standards/main/PYTHON.md
      token: ${CONVENTIONS_URL_TOKEN}

    # .md docs pulled from a git repo (token optional, for private repos)
    - type: git
      repo: https://github.com/my-org/standards.git
      ref: main
      path: python
      token: ${GIT_TOKEN}

  # Optional: enable/disable per review mode (all on by default)
  modes:
    inline: true
    context: true
    summary: true
    inline_reply: true
    summary_reply: true

Convention loading is fail-soft: an unreachable source is logged and skipped, so a review is never blocked.

---

🌐 Review language

Have the AI write its review (inline comments, summaries, and replies) in the language your team prefers. It's a single free-form field — use any language name the model understands. Code identifiers, file paths, and code snippets are always kept unchanged.

review:
  language: Vietnamese   # or English (default), Tiếng Việt, 日本語, Français, ...

👉 Minimal configuration is enough to get started. Use the full reference configs if you want fine-grained control ( timeouts, artifacts, logging, etc.).

---

⚙️ CI/CD Integration

ArgusReview works out-of-the-box with major CI providers. Use these snippets to run ArgusReview automatically on Pull/Merge Requests.
Each integration uses environment variables for LLM and VCS configuration.

For full configuration details (timeouts, artifacts, logging, prompt overrides), see ./docs/configs.

🚀 GitHub Actions

Add a workflow like this (manual trigger from Actions tab):

name: ArgusReview

on:
  workflow_dispatch:
    inputs:
      review-command:
        type: choice
        default: run
        options:
          - run
          - run-inline
          - run-context
          - run-summary
          - run-inline-reply
          - run-summary-reply
          - clear-inline
          - clear-summary
      pull-request-number:
        type: string
        required: true
jobs:
  argus-review:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v6
        with:
          fetch-depth: 0

      - uses: sang-hv/ai-code-review@main
        with:
          review-command: ${{ inputs.review-command }}
        env:
          # --- LLM configuration ---
          LLM__PROVIDER: "OPENAI"
          LLM__META__MODEL: "gpt-4o-mini"
          LLM__META__MAX_TOKENS: "15000"
          LLM__META__TEMPERATURE: "0.3"
          LLM__HTTP_CLIENT__API_URL: "https://api.openai.com/v1"
          LLM__HTTP_CLIENT__API_TOKEN: ${{ secrets.OPENAI_API_KEY }}

          # --- GitHub integration ---
          VCS__PROVIDER: "GITHUB"
          VCS__PIPELINE__OWNER: ${{ github.repository_owner }}
          VCS__PIPELINE__REPO: ${{ github.event.repository.name }}
          VCS__PIPELINE__PULL_NUMBER: ${{ inputs.pull-request-number }}
          VCS__HTTP_CLIENT__API_URL: "https://api.github.com"
          VCS__HTTP_CLIENT__API_TOKEN: ${{ secrets.GITHUB_TOKEN }}

🔗 Full example: ./docs/ci/github.yaml

🚀 GitLab CI/CD

For GitLab users:

argus-review:
  when: manual
  stage: review
  image: ai-code-review:latest   # build from this repo's Dockerfile and push to your own registry
  rules:
    - if: '$CI_MERGE_REQUEST_IID'
  script:
    - argus-review run
  variables:
    # --- LLM configuration ---
    LLM__PROVIDER: "OPENAI"
    LLM__META__MODEL: "gpt-4o-mini"
    LLM__META__MAX_TOKENS: "15000"
    LLM__META__TEMPERATURE: "0.3"
    LLM__HTTP_CLIENT__API_URL: "https://api.openai.com/v1"
    LLM__HTTP_CLIENT__API_TOKEN: "$OPENAI_API_KEY"

    # --- GitLab integration ---
    VCS__PROVIDER: "GITLAB"
    VCS__PIPELINE__PROJECT_ID: "$CI_PROJECT_ID"
    VCS__PIPELINE__MERGE_REQUEST_ID: "$CI_MERGE_REQUEST_IID"
    VCS__HTTP_CLIENT__API_URL: "$CI_SERVER_URL"
    VCS__HTTP_CLIENT__API_TOKEN: "$CI_JOB_TOKEN"
  allow_failure: true  # Optional: don't block pipeline if ArgusReview fails

🔗 Full example: ./docs/ci/gitlab.yaml

---

📘 Documentation

See these folders for reference templates and full configuration options:

• ./docs/ci — CI/CD integration templates (GitHub Actions, GitLab CI, Bitbucket Pipelines, Jenkins)
• ./docs/cli — CLI command reference and usage examples
• ./docs/hooks — hook reference and lifecycle events
• ./docs/configs — full configuration examples (.yaml, .json, .env)
• ./docs/prompts — prompt templates for Python/Go (light & strict modes)
• ./docs/troubleshooting.md — common environment and Git-related issues

---

⚠️ Privacy & Responsibility Notice

ArgusReview does not store, log, or transmit your source code to any external service other than the LLM provider explicitly configured in your .ai-review.yaml.

All data is sent directly from your CI/CD environment to the selected LLM API endpoint (e.g. OpenAI, Gemini, Claude, OpenRouter). No intermediary servers or storage layers are involved.

If you use Ollama, requests are sent to your local or self-hosted Ollama runtime
(by default http://localhost:11434). This allows you to run reviews completely offline, keeping all data strictly inside your infrastructure.

⚠️ Please ensure you use proper API tokens and avoid exposing corporate or personal secrets. If you accidentally leak private code or credentials due to incorrect configuration (e.g., using a personal key instead of an enterprise one), it is your responsibility — the tool does not retain or share any data by itself.

---

🧠 ArgusReview — open-source AI-powered code reviewer

• 📂 Source (this fork)
• 📦 PyPI
• 📜 NOTICE — attribution & fork history