AI Code Review for GitLab, Bitbucket and Azure DevOps (2026)

Direct answer

The field narrows sharply off GitHub. Of 16 reviewers we checked on 2026-10-18, 11 support GitLab, 9 support Bitbucket Cloud, and only 6 support Azure DevOps. CodeRabbit is the only commercial tool documenting all four hosts plus every self-managed edition. PR-Agent (MIT since 23 April 2026) is the most portable: GitHub, GitLab, Bitbucket, Azure DevOps and Gitea, free. Four well-reviewed tools - cubic, Graphite, Macroscope, Cursor Bugbot - are GitHub only.

Disclosure: Gitar is our AI code review category partner. Partners get a guaranteed entry in relevant lists, never a rating, a rank, or the removal of a criticism. Gitar has no published third-party review-quality benchmark, and its own platform matrix lists GitHub Enterprise Server as unsupported. Both facts are below.

If you are on GitHub, this is the wrong page: read AI code review tools that support GitHub instead, which covers GitHub.com, Enterprise Cloud, *.ghe.com data residency and GHES exhaustively. If your repository is large, AI code review for a large codebase covers monorepos and big diffs. This page is about everywhere else.

AI code review tools for GitLab merge requests

Direct answer: 11 of the 16 tools we checked support GitLab merge requests. GitLab’s own Duo Code Review is the native option: assign @GitLabDuo as a reviewer on Premium or Ultimate. Third parties with documented GitLab support: CodeRabbit, Greptile, Qodo, Gitar, Codacy, Korbit, Sourcery, Bito, PR-Agent, Kodus. Seven also document GitLab self-managed. Verified 2026-10-18.

Here is the matrix. Every cell comes from the vendor’s own documentation, checked on 2026-10-18. Where a vendor’s docs and a third-party comparison disagreed, we used the vendor’s docs and said so in How we verified this.

Tool	GitLab SaaS	GitLab Self-Managed	Bitbucket Cloud	Bitbucket Data Center	Azure DevOps Services	Azure DevOps Server	Gitea	Perforce
CodeRabbit	Yes	Yes	Yes	Yes	Yes	Referenced, guide not public	No	No
Greptile	Yes	Yes, Enterprise	Yes, all plans	Yes, Enterprise	No	No	Yes, Enterprise	Yes, P4 Code Review
Qodo	Yes	Yes, incl. air-gapped	Yes	On-prem / air-gapped only	Yes	Not documented	No	Gerrit instead
Gitar	Yes, incl. Free	Yes	Yes	Yes	Yes	Yes	No	No
PR-Agent	Yes	Yes, you host it	Yes	Yes, you host it	Yes	Yes, you host it	Yes	No
Kodus	Yes	Yes, you host it	Yes	Yes, you host it	Yes (Azure Repos)	Yes, you host it	Forgejo	No
GitLab Duo Code Review	Yes	Yes, incl. self-hosted models	n/a	n/a	n/a	n/a	n/a	n/a
GitHub Copilot code review	No	No	No	No	Yes, public preview	No	No	No
Claude Code Review	No	No	No	No	No	No	No	No
Codacy static analysis	Yes	Via Self-hosted	Yes	Yes (Server)	Not documented	No	No	No
Codacy AI Reviewer	No	No	No	No	No	No	No	No
Korbit	Yes	Not documented	Yes, Marketplace app	Not documented	No	No	No	No
Sourcery	Yes	Yes	No	No	No	No	No	No
Bito	Yes	Yes	Yes	Not documented	No	No	No	No
cubic	No	No	No	No	No	No	No	No
Graphite	No	No	No	No	No	No	No	No
Macroscope	No	No	No	No	No	No	No	No
Cursor Bugbot	No	No	No	No	No	No	No	No
SonarQube Server 2026.5	Yes	Yes	Yes	Yes	Yes	Not stated	No	No

Three things in that table are worth stopping on.

CodeRabbit is the only commercial reviewer documenting all four major hosts and every self-managed edition of each. Its platforms overview lists GitHub.com, GitHub Enterprise Server, GitLab.com, GitLab self-managed, Bitbucket Cloud, Bitbucket Data Center, Azure DevOps, and a private-network option that connects a locked-down GHES or self-managed GitLab via reverse tunnel without opening inbound access. That breadth, not review quality, is the reason it keeps winning non-GitHub procurement.

Greptile is broader than its reputation and narrower than you need on Azure. Its code providers page lists GitHub Cloud, GHES, GHEC with data residency, GitLab.com, GitLab Self-Managed, Bitbucket Cloud, Bitbucket Data Center, Cursor Origin (beta), Gitea and Perforce (P4) with P4 Code Review. Azure DevOps is absent. Greptile is the only tool in this matrix documenting Perforce, which makes it the default answer for game studios and hardware teams on Helix Core, and a non-starter for anyone on Azure Repos.

Four tools that comparison pages recommend freely are GitHub only. cubic’s FAQ says it directly: “Not yet. We currently only support GitHub.” Graphite’s docs state it “is built on top of GitHub’s APIs.” Macroscope documents pull requests in GitHub and a local CLI. Cursor Bugbot has no non-GitHub path. If a listicle recommends these to you as a GitLab or Azure DevOps team, it did not check.

AI pull request reviewer for GitLab

Direct answer: Use GitLab Duo Code Review first. It needs no integration: assign @GitLabDuo as a reviewer, or use /assign_reviewer @GitLabDuo. Duo Code Review on self-hosted models reached GA in GitLab 18.4. The newer agentic Code Review Flow reached GA in GitLab 18.8, works on Free, Premium and Ultimate, needs no add-on, and bills GitLab Credits. Verified 2026-10-18.

GitLab merge requests are not GitHub pull requests with a different name, and the differences change which tool fits.

Approval rules, not required reviewers. GitLab’s approval rules can require approvals from specific groups, code owners, or a minimum count, and can reset on new commits. An AI reviewer posting comments does not clear an approval rule. Neither Duo Code Review nor any third party in the matrix approves a merge request. Gitar is the only vendor that documents the constraint from its own side, listing block-merge as conditional on GitLab and Bitbucket approval rules rather than universal.

Pipelines are the enforcement surface. The reliable pattern on GitLab is a job in .gitlab-ci.yml that runs the reviewer or its gate, publishes a status, and is added to the merge request’s required pipeline. That is how SonarQube on GitLab CI has worked for years and it is still the most dependable way to make an AI finding block a merge.

Group-level installation matters more on GitLab than on GitHub. CodeRabbit’s docs note that a GitLab group installation extends automatically to newly added repositories in that group. GitLab’s own automatic Duo Code Review at group or application level is GitLab.com only, with self-managed support listed as a future release in the 18.4 notes. If you run self-managed and want automatic review across 200 projects, that gap is a project, not a toggle.

The native path versus the paid path. Duo Code Review requires the GitLab Duo Enterprise add-on and is available on Premium and Ultimate across GitLab.com, Self-Managed and Dedicated. Code Review Flow is the agentic successor, GA in 18.8, needs no add-on, and consumes GitLab Credits instead, with Free-tier access in 18.10 and later. Those are two different products behind one @GitLabDuo handle, and which one runs depends on whether the initiating user holds a Duo Enterprise seat. Read your own invoice before you assume which you are using.

On sovereignty, GitLab has the strongest documented story in this category. Duo Code Review on self-hosted models went GA in 18.4 with Mistral, Meta Llama, Anthropic Claude and OpenAI GPT families, served through vLLM, Azure OpenAI or AWS. Only SonarQube’s AI CodeFix and the open source pair (PR-Agent, Kodus) come close on bring-your-own-model. One honest caveat: no published benchmark measures self-hosted local-model review quality against a cloud frontier model. The known weak spot is cross-file reasoning, which is the part of review that matters most. We cover that evidence gap in AI code review benchmarks compared.

For the GitLab-specific setup walkthroughs we already have: GitLab code review tools, CodeRabbit GitLab integration and Codacy GitLab integration.

AI code review tools with Bitbucket integration

Direct answer: 9 of 16 tools support Bitbucket Cloud: CodeRabbit, Greptile, Qodo, Gitar, Codacy static analysis, Korbit, Bito, PR-Agent and Kodus. Only 4 document Bitbucket Data Center: CodeRabbit, Greptile (Enterprise), Qodo (on-prem or air-gapped only) and Gitar. Codacy’s AI Reviewer does not work on Bitbucket at all. Verified 2026-10-18.

Bitbucket is where the gap between “integrates” and “works properly” is widest, and Qodo is the only vendor honest enough to publish it.

Qodo’s git provider feature support page states that Bitbucket, Cloud and Data Center both, supports only standard markdown. The practical consequences, from Qodo’s own matrix: no collapsible sections for overflow findings, no expandable description, code, recommendation, evidence or prompt sections, and no severity thresholds for inline comments. The reviewer runs. The output is flatter and harder to skim. Qodo’s page says it “strives for feature parity across all providers” and then documents where it does not have it, which is more than any competitor does.

The deployment-model trap on Data Center. Qodo’s deployment model support matrix shows Bitbucket Data Center as the most restricted provider it serves: on-prem and air-gapped only, no multi-tenant and no single-tenant cloud. There is no quick cloud trial for a Bitbucket Data Center shop. Budget a deployment, not a 14-day pilot.

Bitbucket Pipelines is your gate. Bitbucket has no equivalent of GitHub’s check runs ruleset UI, so merge checks plus a Pipelines step is the enforcement path. Merge checks can require a passing build and a minimum number of approvals; an AI comment satisfies neither. Wire the reviewer into a Pipelines step that exits non-zero on the findings you actually care about, and keep the approval human.

Korbit arrives through the Atlassian Marketplace, which is a different trust and billing path from a direct OAuth install. Korbit’s Bitbucket app creates Bitbucket issues to track findings, grouped by category. It is listed for Bitbucket Cloud; Data Center is not mentioned. Korbit’s published rate card is Pro $12/author/mo annual or $15 monthly and Max $18/$24, billed per Git author rather than per seat, with Max free for all open source.

Codacy is the cautionary tale. Codacy documents Bitbucket Cloud and Server for static analysis across 40-plus languages. Its AI Reviewer is cloud-and-GitHub-only. A Bitbucket team buying Codacy gets the deterministic engine and not the AI layer, and nothing on the pricing page tells you that. This is why the matrix above splits Codacy into two rows.

See also our Bitbucket-specific pages: Bitbucket code review tools, AI pull request review for Bitbucket and Codacy Bitbucket.

AI code review tools for Bitbucket repositories

Direct answer: For Bitbucket Cloud, pick on cost shape: CodeRabbit at $24/dev/mo (Essentials, 1 linked repo) or Greptile at $30/seat where a seat is any developer who received a review. For Bitbucket Data Center, you have four options and all four need a sales conversation or a deployment. PR-Agent is the only free path and it runs anywhere. Verified 2026-10-18.

The repository-count question decides more Bitbucket deals than the review quality question, because Bitbucket workspaces tend to be repository-sprawling where GitHub orgs tend to be monorepo-ish.

Constraint	CodeRabbit	Greptile	Qodo	Gitar	PR-Agent
Entry price	$24/dev/mo annual ($30 monthly)	$30/seat/mo	~$30/mo flat up to 30 users	$20/user/mo annual ($25 monthly)	$0 licence, you pay model tokens
Linked repositories	1 on Essentials, 5 Team, 10 Advanced, 20 Enterprise	Unmetered, billed by author	Not capped by repo	Not capped by repo	Unlimited
Seat definition	Per developer	Any dev who received a review	Flat team fee plus credits	Per user, capped at 50 users	None
Metered overage	$0.25 per reviewed file past hourly limit	$1 per credit past 50/author/mo, credits do not pool	$0.012 per credit, pooled, no rollover	None published	Your own model bill
Files per review	150 Free/Essentials, 300 Team and above	Apex depth costs 10 credits	Not published	Not published	Your config
Free tier	Public repos only. Private is 1 summary-only review/hour	1 active dev, 50 credits/mo	None permanent, 14-day trial	Free for OSI-licensed public repos	Free at any size
Bitbucket Data Center	Yes	Yes, Enterprise	On-prem or air-gapped only	Yes	Yes, you host it

Two readings of that table.

CodeRabbit’s 1-repository Essentials cap is the most consequential constraint in this market and almost every comparison page omits it. A Bitbucket workspace with 12 active repositories does not fit on Essentials at any seat count. You are on Team at $48/dev/mo or Advanced at $72 before you have compared a single finding. Add the $0.25 per reviewed file overage past the hourly limit and the 5/8/10/12 hourly limits that degrade to 1 per hour under fair usage, and the monthly bill is genuinely hard to forecast. We model it in the AI code review price book.

Greptile’s per-author credit model fails in the opposite direction. Credits are per author and do not pool, so one very active developer blows through 50 credits while four quiet ones waste theirs. On a Bitbucket workspace with uneven contribution, that is a worse shape than CodeRabbit’s. Pick the model that matches your contribution curve, not the one with the lower sticker. CodeRabbit vs Greptile works through the arithmetic.

AI code review tool with Azure DevOps support

Direct answer: Six, and that is the whole list: CodeRabbit, Qodo, Gitar, PR-Agent, Kodus and CodeAnt, plus GitHub Copilot code review, which reached public preview for all Azure Repos customers in the sprint 279 update dated 4 September 2026. SonarQube Server 2026.5 integrates for quality gates. Greptile, Korbit, Sourcery, Bito, cubic, Graphite, Macroscope, Cursor Bugbot and Anthropic’s Code Review have no Azure DevOps path. Verified 2026-10-18.

This is where the field collapses, and it is the single most useful fact on this page.

Copilot code review on Azure Repos is now the lowest-friction native option, and it has real limits. From Microsoft’s documentation, last updated 29 September 2026:

• Enablement is three-tiered: a Project Collection Administrator turns it on for the organization, a Project Administrator sets project defaults and whether repositories may override, a repository administrator enables individual repositories.
• Billing goes to the Azure subscription linked to your Azure DevOps organization, metered in GitHub AI credits at 1 credit equals $0.01, visible in Azure Cost Management under meter category GitHub and meter subcategory GitHub Copilot for AzDO. Charges take 48 hours to appear.
• Review effort is Low, Medium or High, set as a project default with optional repository override, and selectable per review. Higher effort costs more tokens. Configurable Lite and Balanced levels are listed as coming soon.
• Hard limits during preview: pull request must be Active with Merge Succeeded, repository 10 GB or less, 100 changed files or fewer, 100 changes or fewer, 1 completed review per merge commit, 1 concurrent review per PR, 5 concurrent reviews per organization, 2 per user.
• Copilot always leaves a Comment review. It never approves and never requests changes, so it does not satisfy required-reviewer policies and does not block merging. It also does not re-review after new commits; you request again.
• TFVC is not supported. Git repositories in Azure Repos only.
• Reviews run on an Azure Pipelines agent pool. Self-hosted agent pools are not supported. Managed DevOps Pools are, with the latest Ubuntu Server image; Windows images are not supported.
• Automatic review is a separate branch policy: enable “Automatically request Copilot code review” per target branch. It can include draft pull requests.

Note the 100-file ceiling against the GitHub side of the same product: GitHub removed the 300-file / 20,000-line cap for Copilot code review on 27 August 2026. Azure Repos still has a tighter cap than GitHub did before that change. Same vendor, same feature name, different product.

CodeRabbit’s Azure DevOps integration is the most thoroughly documented third-party path. Two auth methods, with Microsoft Entra service principal recommended for new organizations and PAT for existing setups. The PAT scopes are specific: Code read/write/manage, Work Items read/write/manage, Identity read (vso.identity), and optionally Build read for pipeline failure remediation. CodeRabbit’s docs are explicit that Azure DevOps write operations require vso.identity on top of resource-specific writes, which is the kind of detail that saves a day of debugging. Two gotchas from the same page: PRs authored by CodeRabbit’s own service account are not reviewed, and the Edit Subscriptions webhook permission cannot be granted from the Azure DevOps UI and needs Azure CLI.

Qodo covers Azure DevOps Services but not Server. Its install guide documents an OAuth app registration via Azure AD, or a PAT with Code, Project and Team, Pull Request Threads and Wiki at read/write. The warning in its own docs: “PAT scopes alone are not sufficient. The user who creates the PAT must also have project-level permissions (typically Project Admin).” Qodo’s deployment matrix lists Azure DevOps Cloud as the only provider with no on-prem option, so a self-hosted Qodo on Azure DevOps Server is not a documented configuration.

AI code review tools for teams using Azure DevOps

Direct answer: On Azure DevOps Services, start with Copilot code review (no new seat, Azure-metered) and bake off CodeRabbit or Qodo against it. On Azure DevOps Server on-premises, the list is down to three documented options: Gitar, PR-Agent and Kodus. CodeAnt claims Server support that we could not confirm at its docs. Verified 2026-10-18.

Azure DevOps Server is the hardest shortlist in this entire category, and the honest answer is uncomfortable.

Gitar is the only commercial tool whose own platform matrix names Azure DevOps Server with a dedicated setup guide. Its platform support page lists GitHub cloud, GitLab SaaS, GitLab self-managed including Free, Bitbucket Cloud, Bitbucket Data Center, Azure DevOps Services, Azure DevOps Server, and AWS CodeCommit in preview. Published rate card: Core $20/user/mo annual or $25 monthly, Pro $40/$50, both capped at 50 users, free for OSI-licensed public repos, 14-day trial, now hosted at sonarsource.com/plans-and-pricing/gitar/ after Sonar acquired Gitar in May 2026.

Two things we have to say about our own partner. First, Gitar’s matrix lists GitHub Enterprise Server as not supported, which is the inverse of almost every other vendor in this category and contradicts secondary sources that list GHES as partially supported. We are reporting the vendor’s own page. Second, Gitar still has no published third-party review-quality benchmark. Sonar’s trust centre added a Gitar SOC 2 Type 2 report and a signed ISO 27001:2022 certificate on 2026-09-07, which is a compliance fact, not a quality one. Also from Gitar’s matrix: one-click apply fixes work only on GitHub and GitLab, verified commit badges are missing on Azure DevOps and Bitbucket Cloud, and block-merge is conditional on approval rules on GitLab and Bitbucket.

PR-Agent and Kodus are the free self-hosted answers, and on Azure DevOps Server they are often the right ones. PR-Agent documents GitHub, GitLab, Bitbucket, Azure DevOps and Gitea, runs as CLI, Docker, GitHub Action or webhook server, and takes OpenAI, Anthropic, Gemini or local models via Ollama. MIT since 23 April 2026, community-governed under The-PR-Agent organisation with Qodo as a sponsor, 13.4k stars, v0.48.0 on 2026-10-16. Kodus is AGPLv3 since September 2026, free to self-host at any team size, documents GitHub, GitLab, Bitbucket, Azure Repos and Forgejo, and accepts any OpenAI-compatible endpoint via vLLM, Ollama, TGI or LiteLLM.

The trade is explicit. c-CRAB scored PR-Agent at 23.1%, third of four agents tested, behind Claude Code at 32.1% and Devin at 24.8% and ahead of Codex at 20.1%. Disclosure: c-CRAB is not conflict-free. Five of six authors are academic (National University of Singapore, Zhejiang University), but Ridwan Shariffdeen lists SonarSource as his affiliation, and the paper carries a disclaimer that its conclusions do not represent SonarSource policy. Sonar sells SonarQube and acquired Gitar in May 2026, and Gitar is a paid sponsor of this site. SonarQube is not among the four tools c-CRAB evaluates, which blunts the conflict considerably, but we flag vendor-adjacent benchmarks everywhere else so we flag this one too.

Azure Boards integration is the one thing Azure DevOps teams get that GitHub teams do not. Qodo validates changes against linked work items and supports Microsoft Entra ID; Kodus documents work-item validation too. If your definition of a good review includes “does this diff actually do what the linked work item asked,” that capability is only available on Azure DevOps and it is genuinely under-covered.

Our existing Azure pages: Azure DevOps code review, AI pull request review on Azure DevOps, CodeRabbit on Azure DevOps and the Azure DevOps pipeline guide.

Code review tools comparison for GitHub, GitLab, and Bitbucket

Direct answer: Across the three hosts, 5 tools work on all three: CodeRabbit, Greptile, Qodo, Gitar and PR-Agent, plus Kodus. Add Azure DevOps and it drops to CodeRabbit, Qodo, Gitar, PR-Agent and Kodus. Codacy, Korbit and Bito cover GitHub, GitLab and Bitbucket but stop at Azure. Sourcery covers GitHub and GitLab only. Verified 2026-10-18.

The terminology differs and so does the mechanism, and conflating them is how teams buy the wrong tool.

Concept	GitHub	GitLab	Bitbucket	Azure DevOps
Change unit	Pull request	Merge request	Pull request	Pull request
Native gate	Branch protection rules / rulesets, required status checks	Approval rules, required pipeline	Merge checks, minimum approvals	Branch policies, required reviewers and build validation
CI surface	GitHub Actions	GitLab CI/CD (.gitlab-ci.yml)	Bitbucket Pipelines	Azure Pipelines
Review-comment API	Review comments plus check runs	Discussions on diffs	Inline comments, no check-run analogue	Pull Request Threads
Self-managed edition	GitHub Enterprise Server	GitLab Self-Managed	Bitbucket Data Center	Azure DevOps Server
Native AI reviewer	Copilot code review (Pro and up)	Duo Code Review / Code Review Flow	None	Copilot code review (public preview)
Can the native AI approve?	Capability shipped, verify in your org	No	n/a	No, documented
Third-party tools in our matrix	16 of 16	11	9 Cloud, 4 Data Center	6

Read that bottom row as a procurement constraint. A GitHub team choosing an AI reviewer is picking from sixteen. An Azure DevOps Server team is picking from three. That is not a difference in degree.

The second-order effect is on evidence. Every benchmark in this market runs on GitHub pull requests: GitHub’s ReviewBench (219 PRs, 187 repos), the Martian Code Review Bench, Entelligence’s benchmark, Augment’s re-run, c-CRAB’s 184 PR instances from 67 repositories. No published benchmark measures review quality on GitLab merge requests, Bitbucket pull requests or Azure DevOps pull requests. Not one. So the honest position for a non-GitHub team is that the benchmark literature tells you about model and product quality in general and nothing about how your host’s integration actually behaves. Pick on documented capability and cost shape. We keep the full provenance ledger in AI code review benchmarks compared.

And the line we will keep repeating until someone falsifies it: no party without a commercial stake has published a false-positive rate, precision figure or noise measurement for any shipping commercial AI code reviewer. SWE-PRBench and CR-Bench publish false-positive data, but for bare models in fixed harnesses, not for products you can buy.

What “supports GitLab” actually means, and the three vendors who tell you

Three vendors publish a per-platform capability grid. Everyone else publishes a logo row. The difference is worth money.

Qodo publishes git provider feature support and names what degrades: Bitbucket and Gerrit lose collapsible sections and expandable detail; Azure DevOps loses CI feedback features and chat in code suggestions; GitLab loses self-review features (demand_self_review, approve_pr_on_self_review); Gerrit loses folder-specific and language or framework filtering, and cannot give automatic feedback on work-in-progress changes because Gerrit changes have no source branch separate from the change ref.

Gitar publishes a per-platform grid: one-click apply fixes on GitHub and GitLab only, no auto-merge on Bitbucket Cloud or AWS CodeCommit, no verified commit badges on Azure DevOps, Bitbucket Cloud or CodeCommit, and CodeCommit drops PR comment threads from the Changes tab after a fresh commit.

CodeRabbit publishes a separate docs page per platform with platform-specific auth and scope detail rather than a single grid.

If a vendor cannot give you that, assume the non-GitHub integration is younger, ask for a reference customer on your exact host and edition, and run the trial on a repository where a flat, un-collapsible review comment will not get the pilot cancelled.

How we verified this

Everything here was checked on 2026-10-18 against vendor documentation and primary announcements, never against another comparison page.

What we checked and where:

• CodeRabbit platform list: docs.coderabbit.ai/platforms/overview for the eight-entry platform list including Bitbucket Data Center, GitLab self-managed and the private-network reverse tunnel; the Azure DevOps page for Entra service principal versus PAT, the exact PAT scopes, the vso.identity requirement, the service-account PR exclusion and the Azure CLI webhook caveat.
• Greptile providers: greptile.com/docs/integrations/code-providers, which carries an explicit plan-requirement column and lists Gitea and Perforce (P4) with P4 Code Review. Azure DevOps is absent from that page.
• Qodo: deployment model support for the multi-tenant / single-tenant / on-prem / air-gapped grid, git provider feature support for the per-provider degradation list, and the Azure DevOps install guide for auth methods and PAT scopes.
• Gitar: docs.gitar.ai/connecting-code/platform-support for the full matrix including Azure DevOps Server, GitLab self-managed including Free, AWS CodeCommit in preview, and GitHub Enterprise Server listed as not supported.
• GitLab Duo: GitLab 18.4 release notes for the Duo Code Review on GitLab Duo Self-Hosted GA entry (Premium and Ultimate, Self-Managed, Duo Enterprise add-on) and for automatic Duo Code Review at group level being GitLab.com only; the Duo in merge requests docs for the @GitLabDuo reviewer behaviour and add-on requirement; the Code Review Flow docs for GA in 18.8, Free tier with GitLab Credits in 18.10 and later, and self-hosted model support.
• Copilot on Azure Repos: Microsoft Learn, updated 29 September 2026 for prerequisites, the three enablement scopes, effort levels, the full requirements-and-limits tables, the never-approves statement, agent pool constraints and Azure Cost Management billing; the Azure DevOps sprint 279 update dated 4 September 2026 for the public preview announcement and the feature list; the GitHub Changelog of 2 June 2026 for the technical preview start and the statement that usage does not draw down included Copilot AI credits.
• PR-Agent: The-PR-Agent/pr-agent for the five-platform list, MIT licence and 13.4k stars.
• Others: cubic’s docs FAQ (“Not yet. We currently only support GitHub.”), Graphite’s docs (“built on top of GitHub’s APIs”), Bito’s AI code review overview for GitHub, GitLab, Bitbucket plus GitHub and GitLab self-managed, Sourcery’s code review docs for GitHub.com, GitLab.com, GHES and self-hosted GitLab quickstarts with no Bitbucket or Azure entry, Codacy’s supported languages and tools for 40-plus languages and the GitHub / GitLab / Bitbucket provider set, and SonarQube Server DevOps platform integration which states that SonarQube Server 2026.5 integrates with GitHub, Bitbucket, GitLab and Azure DevOps.

What we could not verify:

• CodeRabbit’s Azure DevOps Server guide. CodeRabbit’s Azure DevOps page tells self-hosted Azure DevOps Server users to follow a separate Azure DevOps Server setup guide, but we could not open that page at any URL path we tried on 2026-10-18. Separately, CodeRabbit’s self-hosted docs list Azure DevOps among the platforms a self-hosted CodeRabbit supports and gate the whole deployment behind an Enterprise plan with 500 or more seats, with configuration provided during onboarding rather than published. So the capability is claimed and the mechanism is not public. We have marked the cell “Referenced, guide not public” rather than yes or no.
• CodeAnt’s Azure DevOps Server support. Multiple secondary sources, including CodeAnt’s own blog, state that CodeAnt supports both Azure DevOps Services and Azure DevOps Server. We could not confirm it from docs.codeant.ai, which only documented a GitHub connect flow on the pages we could read. Reported, not confirmed. CodeAnt has no tool page on this site for the same reason.
• Korbit’s self-managed support. Korbit’s own site states GitHub, GitLab and Bitbucket; the Atlassian Marketplace listing covers Bitbucket Cloud at version 1.1.8-AC. We found no Korbit documentation site to confirm GitLab self-managed or Bitbucket Data Center. Marked “Not documented”, not “No”.
• Macroscope’s non-GitHub support. Macroscope’s docs reference pull requests in GitHub and a local CLI and do not enumerate providers. We could not find a GitLab, Bitbucket or Azure path and have marked all four as No, but this is inference from absence.
• Semgrep and DeepSource per-platform PR decoration. Both are relevant to these hosts and we could not open an authoritative per-platform page for either on 2026-10-18, so neither is in the matrix. DeepSource was acquired by Harness on 9 September 2026, so its roadmap is in flux regardless.
• Any review-quality benchmark on a non-GitHub host. None exists that we can find, for any tool. Every published benchmark runs on GitHub pull requests. The absence is the finding.
• Whether the tighter Azure Repos caps (100 files, 100 changes) are preview-only. Microsoft’s page says the requirements and limits “apply during the preview and can change.” GitHub removed the equivalent GitHub-side caps on 27 August 2026, so the direction is clear, but the Azure timeline is not published.

One correction to our own prior coverage. Our GitHub platform page records Gitar as listing GHES with a partial-support marker. Gitar’s current platform support page lists GitHub Enterprise Server as not supported. We are reporting the vendor’s live page and flagging the change rather than silently reconciling the two.

Recommendations

Criteria stated: documented support for your exact host and edition, a cost model you can forecast, and a gate that actually blocks a merge. Review quality is deliberately not the first filter, because no benchmark has ever been run on a non-GitHub host, so any ranking you read is GitHub evidence applied to a different integration.

If you are on GitLab SaaS: enable Duo Code Review first, then bake off CodeRabbit. Duo needs no integration and no third-party data path, and if you hold Duo Enterprise seats you are already paying for it. Then run CodeRabbit on one repository for a fortnight and compare. The decision usually comes down to whether Duo’s single-pass, diff-focused review is enough; if it is not, the Code Review Flow at GA in 18.8 is the cheaper next step before a third-party seat line.

If you are on GitLab Self-Managed with data-sovereignty constraints: Duo Code Review with self-hosted models. GA in 18.4, Premium or Ultimate, Duo Enterprise add-on, served via vLLM, Azure OpenAI or AWS. It is the strongest native sovereignty story in this category. Know the two gaps going in: group-level automatic review is GitLab.com only, and no published benchmark compares local-model review quality against a cloud frontier model, with cross-file reasoning the known weak spot. Read AI code review for regulated teams before you design the deployment.

If you are on Bitbucket Cloud: CodeRabbit or Greptile, and pick on your contribution curve. CodeRabbit if your repository count is low and your contribution is even, because the $24 seat is predictable but Essentials covers 1 repository. Greptile if you have many repositories and a few heavy committers, remembering that credits are per author and do not pool. If your workspace is repository-sprawling and your contribution is uneven, both models hurt and you should price PR-Agent against them honestly.

If you are on Bitbucket Data Center: CodeRabbit, Greptile Enterprise, Qodo or Gitar, and budget four to six weeks. Qodo is on-prem or air-gapped only on Data Center, so there is no quick trial. Greptile needs Enterprise. CodeRabbit’s fully self-hosted agent needs 500 or more seats. Gitar has the simplest published rate card of the four at $20/user/mo annual, capped at 50 users. Start the sales conversation before you build the business case.

If you are on Azure DevOps Services: start with Copilot code review, specifically because it is not a new seat. It is metered to your existing Azure subscription at $0.01 per AI credit with per-project cost tags and budget alerts, it enables at organization, project or repository level, and a branch policy makes it automatic. Then measure for a month against its real limits: 100 files, 100 changes, 5 concurrent reviews per org, no automatic re-review after a push, and it will never approve or block. If those limits bite, CodeRabbit has the best-documented third-party Azure DevOps path and Qodo adds Azure Boards work-item validation, which nothing on GitHub offers.

If you are on Azure DevOps Server on-premises: Gitar, PR-Agent or Kodus, and that is the honest list. Gitar if you want a commercial product with a published rate card and a documented Server setup guide; note the 50-user cap, the lack of a third-party quality benchmark, and that its own matrix says no to GHES if you also run GitHub Enterprise Server somewhere. PR-Agent or Kodus if you want zero licence cost and no code leaving your network, accepting c-CRAB’s 23.1% for PR-Agent as the measured trade. We cover the self-hosting ladder in self-hosted AI code review tools and the licence detail in open source AI code review tools.

If you run two or more hosts: standardise on PR-Agent as the floor and buy one commercial tool for your primary host. Nobody sells a multi-host discount, repository caps are what actually bite, and PR-Agent is the only reviewer in this market with the same capability on five hosts at zero licence cost. Run it everywhere as the baseline, then spend your budget where the pull request volume actually is.

For the category-wide shortlist across all hosts, best AI code review tools is the parent page to this one.

Whichever you pick, make the gate a status check and keep approvals human. On GitLab that is a required pipeline, on Bitbucket a merge check plus a Pipelines step, on Azure DevOps a build validation branch policy. No AI reviewer in this matrix approves a change on any host, and the two native ones (Duo, Copilot) document that they never will. Then read the language question, because it changes the answer more than the host does: AI code review by language.