Most “free AI code review” lists are GitHub lists. They name a cloud app, link the GitHub marketplace listing, and stop. If your code lives on a self-managed GitLab, an Azure DevOps Server behind a firewall, or a Bitbucket Data Center, those lists do not apply and they do not say so. The vendor docs for the tools themselves do say so, for each edition. This page is the version of that page for teams who are not on GitHub. Everything here was read from a vendor or project page on 2026-09-21 and carries that date, because “free” and “self-hosted” move monthly.
The honest headline: there are two genuinely free, self-hosted, open-source routes that cover all three forges, and the cloud free tiers you hear about are narrower than the marketing suggests.
What “free” means for code review off-GitHub
For the teams this site serves, “free” usually resolves to one of three things, and they are not the same:
- An open-source project you run yourself, paying only for the LLM calls its your model makes. This works anywhere, including air-gapped.
- A hosted tool’s free tier. These are frequently GitHub-only or gated on the repository being public on GitHub.
- A vendor plugin that ships inside the forge’s own free edition.
The first category is the dependable one for GitLab self-managed, Azure DevOps Server, and Bitbucket Data Center, because it does not depend on a vendor choosing to support your edition. The second and third change by edition and by quarter.
The two open-source projects with real off-GitHub support today are the PR-Agent open-source project (community-owned, MIT) and Alibaba’s Open Code Review. Both run as a CLI against a git repo, both let you bring your own LLM endpoint, and neither requires the forge to have native AI features. That last point is exactly why they cover the self-managed editions the cloud tools ignore.
PR-Agent: the open-source route, now community-owned
As of the vendor docs read on 2026-09-21, PR-Agent is no longer a Qodo product. Qodo donated it to the open-source community and the project now lives in the PR-Agent org on GitHub. The docs moved to docs.pr-agent.ai. The README is explicit that this project is not the Qodo free tier, and that Qodo’s hosted free version for open-source projects sits at a separate page.
For our purposes the important part is the platform support matrix on the current docs. The core review tools, Describe, Review, Improve and Ask, are listed as available on GitHub, GitLab, Bitbucket, Azure DevOps and Gitea. CLI and app/webhook usage are supported across all of them. So the same open-source agent you would run on GitHub against a GitLab merge request, a Bitbucket pull request, or an Azure Repos pull request.
Edition detail matters and is where the docs are honest. The tagging bot is not listed for GitLab. “Ask on code lines” is listed only for GitHub. /help_docs is temporarily disabled since v0.36.1 pending a fix for a credential-exposure issue. So even the open-source route has features that only work on one provider.
Installation is pip install pr-agent, and you bring your own model key. The docs list OpenAI, Anthropic Claude, Google Gemini, DeepSeek, Mistral, and anything reachable through LiteLLM, which includes Azure OpenAI and Ollama. For a self-managed GitLab behind a firewall, that means the only external dependency is whichever model endpoint you already use.
One operational detail worth noting before you upgrade anything: the Docker namespace moved. Docker images from 0.34.2 are published under pragent/pr-agent; older releases up to v0.31 remain at the frozen codiumai/pr-agent namespace, per the project README dated 2026-09-21.
Open Code Review: the older Alibaba tool, now open source
Open Code Review originated as Alibaba’s internal AI code review assistant and was incubated into an open source project. It is a CLI, Apache-2.0, and forge-agnostic in the sense that it operates on a git diff rather than a forge integration. Configuration is an LLM endpoint.
The workflow, from the project readme dated 2026-09-21:
npm install -g @alibaba-group/open-code-review
ocr config provider
ocr config model
cd your-project
ocr reviewThe ocr review command reads staged, unstaged and untracked changes; ocr review --from main --to feature-branch covers a branch range; ocr scan reviews whole files with no git history, which matters for auditing unfamiliar code. A delegation mode lets a host AI agent supply the model so you do not need a separate key.
The design is a hybrid. Deterministic logic handles file selection, file bundling and rule matching, while an agent does the dynamic work of reading context and deciding where comments belong. The project publishes a benchmark, built from 50 popular repositories and 200 pull requests across 10 languages, comparing itself against general-purpose coding agents on the same underlying model. The figures are theirs, not independently verified here, and I would treat the precision-versus-recall trade-off it describes as a design claim rather than a settled result.
For a self-managed forge the appeal is the same as PR-Agent: it runs where the forge runs, needs no forge-side integration, and sends diffs to a model endpoint you control.
What the cloud “free” tiers do and do not cover
The hosted tools with recognizable free tiers are not the free-everywhere answer. CodeRabbit and Qodo both advertise open-source free versions, and both document which forge editions they support. My earlier on this site covers the self-hosted edition split in detail for CodeRabbit and Qodo, but the short version for the free tiers is the fundraising catch: most “free for open source” is GitHub-shaped because that is where open source is measured.
Azure DevOps surfaces its own distinction. Microsoft’s documented AI path for Azure DevOps is an MCP server, and the GitHub Copilot code review feature that does land in Azure Repos is documented on the Azure DevOps Services edition. There is no free native AI PR reviewer built into Azure DevOps Server; teams there use the open-source CLIs above or a webhook bridge.
Deciding for a self-managed forge
For a team on self-managed GitLab, Azure DevOps Server, or Bitbucket Data Center, the decision tree is short. If you want no third-party SaaS touching your code, run PR-Agent or Open Code Review as a CLI or webhook with a model endpoint you already have. Choose PR-Agent when you want the established tool with the richer per-forge documentation and the /review, /improve, /ask workflow. Choose Open Code Review when you want Alibaba’s precision-focused hybrid and the whole-file scan, or when an npm-installed CLI fits your pipeline better than a Python package.
If you are on a cloud forge you can compare the hosted free tiers, but read the vendor’s edition support page first. The free-tier wording almost never covers the self-managed editions, and that gap is exactly the one most summaries gloss over.
Every claim in this page was read from the linked project or vendor page on 2026-09-21. If you are reading this later, re-check the support matrix on docs.pr-agent.ai and the open-code-review readme, because free-tier scope and edition support are the two things that drift fastest.