Skip to main content
MergeWatch is an open-source GitHub App that reviews pull requests using a multi-agent AI pipeline powered by Amazon Bedrock. It deploys entirely within your AWS account — your code never leaves your infrastructure. You pay your own Bedrock bill. There is no per-seat pricing. The project is licensed under AGPL v3.

​
How it compares

MergeWatch

  • Open source (AGPL v3)
  • Runs in your AWS account
  • No per-seat pricing
  • Multi-agent parallel pipeline
  • You control the models (Bedrock)
  • Self-hosted or SaaS

Greptile

  • Closed source
  • SaaS only — code sent to vendor
  • Per-seat pricing
  • Single-model pipeline
  • Vendor-managed infrastructure
  • No self-hosted option

CodeRabbit

  • Closed source
  • SaaS only — code sent to vendor
  • Per-seat pricing
  • Single-pass review
  • Vendor-managed infrastructure
  • No self-hosted option

​
Key differentiators

No per-seat pricing

Install on your entire org. Costs scale with Bedrock usage, not headcount. A 200-person team pays the same platform cost as a 5-person team.

Your AWS account, your Bedrock bill

MergeWatch runs as a Lambda function in your account. Code is read from GitHub, processed in your VPC, and never sent to a third party. You see every Bedrock API call in your own CloudWatch logs.

AGPL v3 — genuinely open source

Read the source, audit it, fork it, contribute back. No open-core bait-and-switch. The AGPL v3 license means improvements must be shared, which keeps the project honest.

Multi-agent parallel review pipeline

Five specialized agents — security, bugs, style, summary, and diagram — run in parallel. An orchestrator agent deduplicates overlapping findings and ranks them by severity before posting to GitHub.

​
The review pipeline at a glance

Each agent is a separate Bedrock invocation running in parallel via Promise.all(). The orchestrator runs after all agents complete.
PR opened → GitHub webhook → SQS FIFO → Lambda
  ├── Security agent
  ├── Bug detection agent
  ├── Style agent
  ├── Summary agent
  └── Diagram agent
       ↓
  Orchestrator (dedup + rank)
       ↓
  GitHub review comments posted

​
Deployment models

ModelCode leaves your infra?Who pays Bedrock?Setup effort
Fully Self-HostedNoYou~30 min (SAM deploy)
SaaS + Customer Bedrock (BYOC)NoYou~10 min
SaaS + MergeWatch BedrockYes (to MergeWatch)MergeWatch (included)~5 min

​
Infrastructure

MergeWatch runs on a standard serverless stack:
  • AWS Lambda — review pipeline execution
  • Amazon DynamoDB — review state, repo config, user data
  • API Gateway — GitHub webhook receiver, dashboard API
  • SQS FIFO — queues PR events for ordered processing
  • Amazon Bedrock — Claude Sonnet, Haiku, Llama, Mistral (configurable)
Configuration lives in a .mergewatch.yml file at the root of each repository. The dashboard (Next.js) provides a UI for monitoring reviews, managing repos, and adjusting settings.
Both tools review PRs with AI. The differences are structural:Where your code goes. CodeRabbit is SaaS-only. Your code is sent to their servers for processing. MergeWatch can run entirely in your AWS account — code never leaves your infrastructure.How you pay. CodeRabbit charges per seat. MergeWatch has no per-seat fee. You pay your own Bedrock bill, which scales with usage, not team size.Review architecture. CodeRabbit uses a single-pass review. MergeWatch runs five specialized agents in parallel and deduplicates their findings before posting. This tends to produce fewer duplicate comments and better coverage across security, correctness, and style.Source availability. CodeRabbit is closed source. MergeWatch is AGPL v3. You can read every line of the review logic, audit it for your compliance needs, and contribute improvements.Model choice. CodeRabbit decides which models to use. MergeWatch lets you configure which Bedrock models run for each agent. You can swap in newer models as they become available without waiting for a vendor update.

Quickstart

Install MergeWatch and get your first review in under 10 minutes.

How it works

Detailed walkthrough of the review pipeline, agent architecture, and orchestration.