CodeRabbit vs Graphite (2026)

Integration is deep on GitHub across repositories, pull requests, CI and merge, with surfaces in the CLI, VS Code, MCP, the web pull request page and the PR inbox. Cursor Cloud Agents now run inside the Graphite pull request, a post acquisition integration and the clearest product evidence that the two products are converging.

The orchestration is real and layered: a stack aware merge queue with basic and advanced tiers, an automations engine, a CI optimizer, and automatic rebase of dependent branches as changes land. Sequencing dependent pull requests to keep trunk green while keeping the chain current is multi step workflow execution over a dependency graph, which is what this axis measures. It orchestrates the merge pipeline rather than coordinating multiple agents, so it earns Full on the multi step half rather than the multi agent half.

Every pull request opened or updated starts a review with no person involved, and automations fire on pull request events. Trigger coverage spans pull request events, merge queue events and CI events, plus the CLI, VS Code, MCP, web and inbox surfaces, with Slack as a notification output. Everything is anchored to GitHub and the local development loop: there is no chat initiated or ticket initiated invocation and no scheduled runs.

Whole codebase context and code indexing are confirmed first party, with admin controls over code indexing. Indexing is optional and described as indexing code for short periods of time to speed Graphite Chat's tool calls and the PR page, while reviews that reference related files through the index are marked coming soon. By default Graphite stores no code and reads it from GitHub. No index architecture, retrieval method or scope statement is documented, so a maintained retrieval structure grounding the agent's reviews is not yet shipped, which keeps this at Partial.

Learnings are opt in and stored per organisation, which is durable customer scoped state rather than session memory; incremental review state persists across pushes within a PR.

The AI reviewer tunes itself from accepted and rejected comments, which is real but is feedback loop adaptation of a model's behavior rather than a documented memory layer. Imported style guides and rules are static configuration the customer maintains. Nothing is named as a memory feature with a described mechanism; the line is whether the vendor documents persistence as a capability rather than as a training characteristic.

Guardrails are strong on the input side (rules, path instructions, gating checks) but the product auto publishes review comments without an approval step, and its actions are advisory rather than merging code, so the oversight question is narrower here than for an agent that acts on systems.

Oversight is largely structural: the AI reviewer's output is advisory findings a person accepts or rejects. What lifts it toward the gate this axis wants is the merge queue, which sequences and merges approved pull requests under rules the team configures, with advanced settings on Enterprise, and ACLs that bound who can do what. It stays below Full because no per action approval gate on the agent's own actions is documented, and there is no shadow or preview mode for automatic approval.

Full. SOC 2 Type II is stated on the FAQ, landing and security pages, and SSO, custom RBAC and audit logging are documented on the Enterprise plan, which is a commercial gate rather than an absent capability. The trust center itself does not render for automated reading, so the report details were not re-read.

The pricing page's Admin section enumerates the identity and governance matrix in full, and the bar is met twice over: SOC 2 Type II plus continuous penetration testing on the attestation side, and SAML and SSO, ACLs, SIEM audit log export, code indexing controls, AI privacy controls and GitHub Enterprise Server support on the named control side.

Analytics, learnings and audit logging are documented, but they report on review activity and configuration rather than tracing why the agent reached a given conclusion. Downgraded on the same reading applied to abnormal-ai, actively-ai and adonis in this session; the PR walkthrough does surface reasoning per finding, which is why this is P rather than N.

Team insights and custom analytics are genuine reporting, and the Enterprise SIEM audit log is a real governance artifact, but both measure people and access rather than agent behavior. Reporting on throughput and cycle time is not a retained per action trace of what the AI reviewer did and why. A documented per review execution record would move this to Full.

Self hosting is gated to Enterprise customers at 500 or more seats, a significant commercial threshold, but the capability including air gapped operation is documented.

Graphite is hosted software. GitHub Enterprise Server support is listed on Enterprise, but that is support for a customer hosted code host, not a deployment option for Graphite itself; portability of the code host does not confer residency on the platform. No region selection, VPC, private instance or self hosting of Graphite is documented.

Recipes and checks are customer authored templates rather than a vendor library of prebuilt agents, which is why this stays at P rather than moving to F.

AI review customization covering automations, filters and rules is reusable configuration, but this axis rewards prebuilt assets the customer installs or adopts, and every one of these is authored by the customer. Graphite ships a single reviewer agent with no library, gallery, template set or marketplace on the pricing matrix, the features navigation or the homepage.

Model choice is real but gated to self hosted Enterprise at 500 plus seats; cloud customers get no selection surface. Graded F on the documented buyer facing choice, with the gating recorded here.

Graphite names Anthropic and OpenAI as the model providers behind its AI features and keeps the choice itself. AI privacy controls and code indexing controls are admin capabilities over data handling, not model selection. The model layer may change materially: Cursor's stated plan on acquisition was to use its coding models to make Graphite's AI features more intelligent and to merge Graphite's reviewer with Cursor's Bugbot.

MCP here is the inbound direction, CodeRabbit consuming external tools for context, which is the opposite arrow from a vendor exposing its own MCP server; credited on the API plus integration surface rather than on the MCP servers alone.

MCP is listed on the pricing page as a feature on every plan, including the free Hobby tier, grouped under Stacking alongside the CLI and VS Code extension. The grouping indicates an MCP server exposing Graphite's stacking operations to other agents rather than a general platform API. No REST API, SDK or webhook surface is documented, and Full would need a documented API or SDK for building on Graphite.

Downgraded per the axis rule that this measures what the customer can test of the agent, not what the agent tests of the code. Unit test generation is the product's output; there is no harness for evaluating review quality or regression testing agent behaviour.

The axis and the product coincide: the customer points a shipped review product at their own code and it finds defects, with customer configurable rules and filters. This is a defect detection product rather than an agent running a project's existing tests. No harness for evaluating the agent's own behavior exists, so the grade rests on the product being a test of the customer's work.

Every documented action runs through git and the GitHub API. Reviewing and navigating code, applying commits, rebase operations through the CLI and self healing CI are all programmatic, which is exactly what this axis excludes. No browser, screenshot or GUI automation appears on the pricing matrix, the features navigation or the homepage. Cursor Cloud Agents now run inside the pull request, and if those agents carry browser tooling that could change, but that capability belongs to Cursor's record and is not documented on any Graphite surface.