Before you connect a repository.

Sign in at app.corbulo.dev, create a project, and connect your repos. Corbulo starts building the code graph and analysis from there. Free to start; paid plans from $60/mo.

Not yet. Ninety percent precision is the product target, not a current benchmark claim. We are building and testing against a reviewed corpus to raise precision without narrowing the analysis until it misses important bugs. We will publish the methodology and results once the benchmark is stable.

Read access to the repos you connect and write access only for posting review comments. Permissions are scoped per repo, fully auditable, and revocable from your GitHub settings at any time.

GitHub only at launch. GitLab and Bitbucket are on the roadmap. If you need them, tell us. It helps us prioritize.

LLM Powered reviews only: unique PRs per billing period, deduplicated by repo and PR number. Analysis Engine reviews — the deterministic engines, static analyzers and graph impact — are unlimited on every plan, and failed analyses never count.

A project is a workspace that groups related repositories, team members, and settings. For example, a microservices backend and its frontend could share one project. Free includes 1 project with 1 repository, Indie 3 projects with up to 3 repositories each; Pro and Enterprise are unlimited.

Pro is $175/mo flat and includes up to 5 seats and 500 LLM Powered PR reviews per month, shared across the team. Need more seats? Talk to us about Enterprise.

Every member of your billing team. The team owner's subscription covers all members.

Project and repository creation are blocked at the plan's limit. When a paid plan reaches its LLM Powered review count, new PRs still get an Analysis Engine review until the period resets. Upgrade anytime via the billing portal.

Yes. Add an Anthropic API key, choose the model, and optionally set a custom Anthropic-compatible base URL. Corbulo uses that project configuration for model-based analysis instead of the platform default.

Yes. Corbulo performs a shallow git clone to analyze your code. Clones are temporary — created in an isolated directory, used for analysis, and deleted immediately after. The bulk of what's stored is a structured knowledge graph of symbols, dependencies, and relationships. Small code snippets (a few lines each) are kept alongside detected issues to show you exactly what was flagged. Code diffs for PR reviews are fetched on-demand from GitHub and not stored.

We use Anthropic-compatible open source models hosted on Ollama Cloud. The specific models may change as we evaluate performance. If you have a preference, you can use the BYOK feature to bring any Anthropic-compatible model.

On the cloud plan, code snippets are sent to open source models hosted on Ollama Cloud for analysis. Your code is not sent to OpenAI, Anthropic, or any other large AI provider. We plan to move to self-hosted models on infrastructure we fully control in the future. With BYOK, you choose exactly where your code goes.

Corbulo's cloud infrastructure runs on Hetzner Cloud in Germany. All data — analysis results, health scores, and metadata — is stored in the EU. Source code is never persisted; only structured analysis results are kept.

Analysis results (health scores, detected issues, conventions, architecture dimensions) are stored for the lifetime of your project. Repository clones are ephemeral and deleted immediately after analysis. Deleting a project permanently removes all associated data. Removing a repository from a project does not delete its analysis history.

No. Your code is not used to train or fine-tune any model, and no training pipeline exists in Corbulo.

Each analysis runs in a sandboxed Docker container with read-only access to the repository and no network access. Containers are destroyed after every run. On Kubernetes deployments, each analysis runs as an isolated Job with its own pod. No analysis job can access another project's data or reach external services.

Data is encrypted in transit with TLS and at rest. GitHub tokens and BYOK provider credentials are encrypted with AES-256-GCM. Corbulo-issued API keys are stored as SHA-256 hashes, not plaintext.

Only the repos you explicitly connect. You choose which repos to analyze and you can revoke access from GitHub at any time. No blanket org-level access.

Not yet. We're an early-stage product and SOC 2 is on our roadmap. If compliance certification is a requirement, our Enterprise self-hosted option lets you run Corbulo inside your own audited infrastructure.

Yes. Enterprise plans include self-hosted options via Docker Compose or Kubernetes with Helm charts. Run Corbulo inside your own infrastructure — your code and data never leave your network.