MergeWatch vs CodeRabbit

​

At a glance

​

CodeRabbit

• What it is. A commercial AI code review platform that positions itself as the most-installed AI app on GitHub. (coderabbit.ai)
• Trigger model. PR webhook is primary. Also supports IDE integration, CLI, and an “Agentic Chat” bot interface. (coderabbit.ai)
• Where it runs. SaaS by default. Self-hosted is available only to Enterprise customers with 500+ seats. (docs.coderabbit.ai)
• LLM flexibility. On default SaaS the model is not publicly documented. The self-hosted build lets customers plug in OpenAI, Azure OpenAI, AWS Bedrock (Claude 3/3.5/4 family), or Anthropic directly. (docs.coderabbit.ai)
• Open source. Closed-source commercial product. Runs a “free for OSS projects” program and has distributed $600K+ to maintainers. (coderabbit.ai/oss)
• Agent architecture. Described as “agentic reviews” with codegraph/AST context, 40+ integrated linters and security scanners, and a “Fix with AI” remediation step. Agent count or single-vs-multi-pass composition not publicly documented. (coderabbit.ai)
• Pricing. Free $0 (14-day Pro trial, PR summarization, IDE reviews). Pro$24/user/month annual (linters/SAST, Jira/Linear, 5 MCP connections, 5 reviews/hour). Pro Plus $48/user/month annual (custom pre-merge checks, 15 MCP connections, 10 reviews/hour). Enterprise custom (RBAC, SSO, self-hosting, SLA). (coderabbit.ai/pricing)
• Data handling. “SSL encrypted data,” “zero data retention post-review,” SOC 2 Type II certified. (coderabbit.ai)

​

Where MergeWatch differs

• Self-host with a single docker-compose up, no seat minimum. MergeWatch runs as three Docker containers (Express server, Next.js dashboard, Postgres 16) plus your chosen LLM provider. Any team size can self-host.
• Fully open source under AGPL-3.0. Every agent prompt, the orchestrator, and the comment templates live in the public repository. You can read exactly what runs on your code.
• LLM choice beyond the big four. MergeWatch ships first-party providers for Amazon Bedrock, Anthropic direct, LiteLLM (100+ providers via one proxy), and Ollama (local/air-gapped). You can switch at any time with a single LLM_PROVIDER env change.
• Merge-readiness score. MergeWatch produces a 1–5 score per PR based on severity + confidence across all findings. CodeRabbit does not document an equivalent composite signal.
• Agent-authored PR detection. MergeWatch classifies each PR as source='agent' vs source='human' using commit trailers, branch prefixes, and PR labels configured in .mergewatch.yml. Agent-authored PRs get a stricter prompt suffix that flags hallucinated imports, tests without assertions, and stale APIs.
• Conventions auto-discovery. Drop an AGENTS.md, CONVENTIONS.md, or .mergewatch/conventions.md in the repo root and the content (up to 16 KB) is injected into every agent’s prompt automatically — no YAML rules DSL to learn.
• No per-seat pricing. Managed SaaS bills usage directly via a prepaid Stripe balance; self-hosted bills only your LLM provider.

​

When CodeRabbit might be the better fit

• You want the most widely-installed GitHub Marketplace brand and already have budget for per-seat SaaS.
• You want 40+ pre-integrated third-party linters and SAST scanners wired into the review with no configuration.
• You are a 500+ seat enterprise that wants self-hosting and an established SOC 2 Type II vendor with a sales motion around it.
• You actively use CodeRabbit’s Jira / Linear integrations and chat surface in ways MergeWatch does not replicate.

​

Cross-links

• Full quick matrix — all eight tools side by side.
• Feature matrix — boolean checkmark grid.
• Self-hosting MergeWatch — the docker-compose up path.