ReviewAI Gerrit Plugin

Note

TL;DR - ReviewAI is a Gerrit plugin by Amarula Solutions that brings AI-powered code review to the Gerrit change page through a sidebar chat interface, supporting automatic Patch Set reviews, follow-up conversations, and multiple AI provider backends. - Licensed under Apache 2.0, supporting OpenAI, DeepSeek, Gemini, Ollama, and MoonShot via LangChain, with specialized review agents for correctness, testability, code quality, documentation, and security.

Contact Us

What is ReviewAI?

ReviewAI is an Apache 2.0-licensed Gerrit plugin that integrates an AI-powered code review assistant directly into the Gerrit UI. It provides a sidebar chat interface on the change page where developers can ask questions about Patch Sets, trigger automatic reviews, and maintain per-review conversation history.

Originally forked from xielong/chatgpt-code-review-gerrit-plugin, Amarula Solutions has evolved it into a multi-provider, multi-agent review platform with 587+ commits of development.

Multi-Provider Backend

ReviewAI supports multiple AI providers through LangChain: OpenAI, DeepSeek, Google Gemini, MoonShot, and Ollama (local, self-hosted). Users select provider and model combinations from a sidebar dropdown. Tokens are configured per-provider and can be stored in Gerrit’s secure.config.

Specialized Review Agents

Three agent modes provide increasing sophistication:

• Single Agent — one AI reviewer evaluates the entire Patch Set

• Scoped Agents — separate agents for code changes and commit messages

• Specialized Agents — dedicated agents for correctness, testability, code quality, documentation, and security (from the Sashiko project)

On-Demand Code Context

When configured, the AI can request repository context via tool calls: file listing, reference search, and file reading — providing deeper understanding beyond just the patch diff.

How It Works

1. A Patch Set is submitted to Gerrit.

2. ReviewAI formats the changed code with configurable context lines and sends it to the selected AI provider.

3. The AI returns a structured JSON response parsed into Gerrit comments (patch-set-level and inline).

4. A Code-Review vote is optionally cast.

5. Developers can continue the conversation via Gerrit comments by mentioning the AI user (@{gerritUserName}) or through the sidebar chat.

ReviewAI also supports topic-aware review — grouping related Patch Sets by topic for a unified AI review across a series.

Features at a Glance

Sidebar Chat	Full conversation interface embedded in the Gerrit change page
Automatic Review	AI reviews every Patch Set on submission, posting comments and optionally casting votes
Slash Commands	/review, /help, /suggest (generates Gerrit suggested edits), /forget_thread, /configure, /show, /directives
Multi-Provider	OpenAI, DeepSeek, Gemini, MoonShot, Ollama via LangChain
Model Selection	Select provider/model from the sidebar dropdown
Specialized Agents	Correctness, testability, code quality, documentation, security agents from the Sashiko project
Directives System	Mandatory plain-English review instructions, configurable globally, per-project, or dynamically
Comment Filtering	Relevance-based filtering of AI comments with configurable threshold (default 0.6)
Code Context	ON_DEMAND mode lets the AI request repository context via tool calls (file listing, reference search, file reading)
ZDR Mode	Zero Data Retention for OpenAI — uses local plugin memory instead of server-side conversation storage
Secure Config	API tokens stored in secure.config with optional encryption

ReviewAI sidebar chat interface embedded in the Gerrit change page

AI-generated Patch Set review results with inline comments

Provider and model selection from the sidebar dropdown

Follow-up conversation through the ReviewAI sidebar

AI comments and Code-Review vote visible in the Gerrit change log

Getting Started

Requirements

• JDK 21 and Bazel (Bazelisk recommended)

• Gerrit compatible with the plugin’s Bazel external deps

• API tokens for at least one AI provider

Build from Source

git clone https://github.com/amarula/reviewai-gerrit-plugin.git
cd reviewai-gerrit-plugin

# Link into a Gerrit checkout
cd gerrit/plugins
ln -s /path/to/reviewai-gerrit-plugin .

# Build
cd ../..
bazelisk build plugins/reviewai-gerrit-plugin:reviewai-gerrit-plugin

# Development build (includes admin/debug features)
bazelisk build plugins/reviewai-gerrit-plugin:reviewai-gerrit-plugin-dev

Installation

Upload the built JAR to $gerrit_site/plugins/ and restart Gerrit.

Configuration

Set up in $gerrit_site/etc/gerrit.config under [plugin "reviewai-gerrit-plugin"]:

Parameter	Description
gerritUserName	Gerrit username of the AI bot user (required)
aiTokens	Provider tokens, e.g. OpenAI/{key}, Gemini/{key}
aiProviders	Provider routes to expose (default: OpenAI)
aiModels	Specific model routes by provider
aiReviewPatchSet	Enable automatic review (default: true)
enabledVoting	Allow AI to cast Code-Review votes (default: false)
agentSpecializationLevel	SINGLE_AGENT, SCOPED_AGENTS, or SPECIALIZED_AGENTS
maxReviewLines	Max lines included in review (default: 1000)
codeContextPolicy	NONE or ON_DEMAND (default: NONE)
enabledFileExtensions	File types to include in reviews
directive	Mandatory review instructions (repeatable)

For the complete configuration reference, see the README.

Key Configuration Parameters

Setting	Purpose
aiDomain	Custom API endpoint (overrides provider defaults)
aiReviewTemperature	Temperature for Patch Set reviews (default: 0.2)
aiReviewCommitMessages	Also review commit messages (default: true)
patchContextLines	Context lines around changes (default: 3)
aiMaxMemoryTokens	Max tokens in conversation memory (default: 16K)
ollamaDomain	Ollama server endpoint (default: http://localhost:11434)
commentRelevanceThreshold	Minimum relevance score for AI comments (default: 0.6)

Testing

bazelisk test plugins/reviewai-gerrit-plugin:reviewai_tests
bazelisk test plugins/reviewai-gerrit-plugin:reviewai_tests \
  plugins/reviewai-gerrit-plugin:reviewai_dev_tests

Why Amarula Maintains This

Amarula Solutions runs Gerrit-based code review across all development workflows — from embedded Linux BSPs and kernel drivers to Jenkins pipelines and Flutter applications. ReviewAI reduces review latency, catches issues early, and provides a consistent first-pass analysis before human reviewers engage.

The plugin has been extensively tested in production environments and evolved through real-world feedback into a mature, multi-provider AI review platform.

License and Attribution

• License: Apache License, Version 2.0

• Repository: amarula/reviewai-gerrit-plugin

• Upstream: Forked from xielong/chatgpt-code-review-gerrit-plugin

Tip

Want AI-powered code review in your Gerrit workflow? Amarula Solutions provides ReviewAI deployment, custom directive configuration, and multi-provider AI integration for on-premise Gerrit instances. Contact us