Architecture Overview

MergeWatch can run as a fully managed SaaS service or as a self-hosted Docker deployment. Both architectures follow the same review pipeline — the difference is where the components run and which LLM provider handles inference.

​

SaaS architecture

The managed SaaS deployment runs entirely on AWS in MergeWatch’s account.

​

SaaS component table

​

SaaS latency breakdown

---

​

Self-hosted architecture

The self-hosted deployment runs as a Docker container on your own infrastructure.

​

Self-hosted component table

​

Self-hosted latency breakdown

---

​

Concurrency model

Both architectures run the eight review agents — security, bugs, style, error handling, test coverage, comment accuracy, summary, and diagram — in parallel via Promise.all().

• SaaS: WebhookHandler invokes ReviewAgent asynchronously via the AWS Lambda Invoke API (InvocationType.Event). There is no queue between the two functions and no per-PR serialization. Two rapid pushes to the same PR each trigger a separate ReviewAgent invocation that runs concurrently.
• Self-hosted: The Express server handles each webhook request in a single async function. Concurrent PRs are handled by Node.js’s event loop. For high-volume organizations, scale horizontally by running multiple container replicas behind a load balancer.

In both modes, if two invocations overlap on the same PR, the last one to finish overwrites the summary comment (last write wins). Inline review comments are additive and not affected by concurrency.

​

Next steps