Aica

Motivation

There are already excellent code review tools such as pr-agent and cursor. However, pr-agent relies on code hosting services like GitHub, which limits its usability, and cursor is not open source, meaning it cannot be fully customized or integrated with other tools.

So, I decided to create a new tool with the following characteristics:

• Customizable
• Open Source
• Platform Independent

Features

• AI Coding Agent
• AI Code Review
• MCP(Model Context Protocol) supported. stdio and SSE transports are supported.
• Automatic knowledge retrieving for code review
• Symbol based code search for retrieving knowledge
• Vector based document search for retrieving knowledge
• Generate summary of changes
• Generate commit message
• Create pull request with AI-generated title and body
• Prompt customization
• Slack notification
• Single binary executable by bun build --compile
• GitHub Actions integration. (See wiki page to setup actions.)

Install

Build and install a binary:

# Install bun before build aica.
#
# Official Install Document:
# https://bun.sh/docs/installation#installing

git clone https://github.com/dotneet/aica.git
cd aica

bun install
bun run build
cp ./dist/aica path-to-your-bin-directory

Setup GitHub Token:

# if GITHUB_TOKEN is not set, aica try to get token from `gh auth token`.
export GITHUB_TOKEN=your_github_token

Setup LLM:

# You can set the following items in your environment variables or aica.toml file.
# Configure at least one of the following providers:

# Anthropic (If you use agent, strongly recommend to use claude sonnet model)
export AICA_LLM_PROVIDER=anthropic
export ANTHROPIC_API_KEY=your_api_key
export ANTHROPIC_MODEL=claude-3-5-sonnet-20241022

# OpenAI
export AICA_LLM_PROVIDER=openai
export OPENAI_API_KEY=your_api_key
export OPENAI_MODEL=o3-mini

# Gemini
export AICA_LLM_PROVIDER=google
export GOOGLE_API_KEY=your_api_key
export GOOGLE_MODEL=gemini-2.0-flash

Configuration

You can customize the configuration by creating a aica.toml file.

See aica.example.toml.

aica.toml must be placed in one of the following directories.

• root directory of the repository
• ${HOME}/.config/aica/aica.toml
• ${GITHUB_WORKSPACE}/aica.toml

Language

You can specify the language for AI output. The language can be specified either through AICA_LANGUAGE or in the language section of aica.toml. By default, the language is automatically detected from the LANG environment variable.

export AICA_LANGUAGE=Japanese

Priority:

1. AICA_LANGUAGE
2. language.language in aica.toml
3. Automatic detection from LANG

Usage

Review

# review the diff from HEAD
aica review [options] [pattern]

# review specific files
aica review src/main.ts

# review the files matching the specific glob pattern
aica review "src/**/*.ts"

Options:

• --dir: Target directory path
• --slack: Send notification to Slack

Agent

# execute AI agent with a prompt
aica agent "your prompt here"

# execute AI agent with a instruction file
aica agent -f instruction.txt

# conversation mode. You can interact with AI agent by typing your prompt.
aica agent

This command executes a task using an AI agent. The agent automatically determines and executes the necessary actions based on the given prompt.

Recommend to use anthropic claude 3.5 sonnet for agent.

NOTE: This command has potential to break your file system. Please be careful.

Aica Agent as an MCP Server

# start a MCP server
aica mcp server

# start a MCP server with a config file
aica mcp server --config aica.toml

Reindex

# reindex the code and document databases
aica reindex

Summary

# generate a summary of the diff from HEAD
aica summary [options]

Options:

• --dir: Target directory path

Commit Changes

This command commits changes with an AI-generated commit message.

# commit all changes(including untracked and unstaged changes) with an AI-generated commit message
aica commit [options] # commits all changes (including untracked and unstaged changes) with an AI-generated commit message

aica commit --staged # commits only staged changes with an AI-generated commit message

aica commit --push # commits all changes and pushes to remote repository

Options:

• --staged: commit only staged changes.
• --dryRun: Show result without execution
• --push: Push to remote repository after committing

Create Pull Request

This command creates a pull request on GitHub.

# Creates a pull request. If there are changes, they will be committed automatically.
aica create-pr [options]

# Creates a pull request with only staged changes
aica create-pr --staged

Options:

• --withSummary: Generate summary of diff from HEAD (default: true)
• --body: Pull request body
• --dryRun: Show result without execution
• --staged: Only include staged changes

Generate Commit Message

# generate a one-line commit message based on the diff from HEAD
aica commit-message [options]

Options:

• --dir: Target directory path

Show Configuration

# show current configuration
aica show-config [options]

Options:

• --default: Show default configuration

Other Commands

• aica --version: Show version information
• aica --help [command]: Show help information for a specific command or general help

Add Context

.cursorrules and .clinerules are automatically added to the context. To customize the context, configure the [rules] section in aica.toml.

Additionally, .cursor/rules/*.mdc files are supported by default. This function can be disabled through the settings.

MCP(Model Context Protocol)

create a mcp.json file to define the MCP server.

example:

[
  {
    "name": "example-server",
    "type": "stdio",
    "command": "node",
    "args": ["./server.js"]
  },
  {
    "name": "example-server",
    "type": "sse",
    "url": "http://localhost:3001/sse"
  }
]

set setupFile in aica.toml like below.

[mcp]
setupFile = "mcp.json"

GitHub Actions

Setup

See wiki page for details.

Commands

You can interact with AI features by leaving comments with the following commands.

# edit the code with AI
/aica edit "your prompt here"

# update summary in the pull request body
/aica summary

# review the latest diff
/aica review