Follow callers, imports, and surrounding code. Deep review investigates beyond the changed lines, then checks the candidate findings in a separate verification phase.
DEEP REVIEW / how it works
Illustrative investigation
Context plan / example paths
READING ORDER3 PRIORITY PATHS
01
Changed filebilling/invoices.py
Tenant filter removed from the lookup.
Start here
02
Incoming callerapi/invoices.py
Trace the organization supplied by the request.
Follow next
03
Outgoing importbilling/models.py
Check whether the model scopes the query.
Inspect boundary
Suggested starting points. Full snapshots remain available.
The context plan prioritizes changed files, outgoing imports, and incoming callers for investigation.
ADVISORY BY DESIGN
More leads to investigate. A separate report. Deep-review findings never block your merge.
See how a finding becomes a proposed fix →Opt-in beta for eligible repositories. Broader discovery can produce more false positives; standard review remains the merge-gating review.
03 / Auto-fix, with a clear handoff
Found it? Move straight to the fix.
Scope the change. Inspect the proposal. See what was checked. Keep the decision to ship in your hands.
01 Scope the fix→02 Generate a patch→03 Inspect checks→04 You decide
billing-service / proposal #12
Illustrative workflow
Proposed fix
Restore the tenant boundary.
Limit the invoice lookup to the current organization. Keep the change focused on the finding.
Scope
1 lookup · 1 regression test
Decision
Awaiting your review
Example proposal and checks. No code is executed in this demo.
[cv]Review platformCode Voucher hosted platformDashboard · orchestration · review data
↔Separate data boundaries
Your infrastructure / planned
>_Review agentsYour infrastructure · your runnerExecution in your environment
Your infrastructure / self-hosting concept
[cv]Review platformYour dashboard & orchestrationDashboard · orchestration · review data
↔
>_Review agentsYour infrastructure · your agentsIsolated review execution
↕
Your configured model connection
MODEL PROVIDERYour key or supported subscriptionModel requests follow the selected provider’s data boundary.
Get started with the hosted platform.
Connect a repository and use hosted execution. Bring your own API key or supported subscription.
Bring execution into your environment.
Planned: register your own runners while keeping the hosted dashboard. The platform and runner have separate data boundaries; this does not make the whole service self-hosted.
Run the platform on your terms.
Concept: host the platform, review data, and execution yourself. External model requests still follow your selected provider’s data policy.
06 / Bring the AI you already use
A new review workflow. Your existing AI account.
Bring your own key or supported subscription. Keep the model relationship you already have, and make it useful across your reviews.
CONNECTION WALKTHROUGH
Connection preview / illustrative
>_Provider API keyCONFIGURATION PREVIEW
provider
your model provider
credential
•••• •••• ••••
purpose
review execution
billing
your provider account
EXAMPLE ONLY / NO CREDENTIALS COLLECTED
Your model usage is billed through your provider. This example does not collect credentials.
Supported subscription
01
Choose an eligible provider
Check support for your subscription.
02
Account authorization
Authorize through the provider’s supported flow.
03
Return to your reviews
Review execution uses the authorized connection.
Use an eligible subscription connection. Provider support and subscription limits apply. This is an illustrative connection preview.
Further ahead / product direction
Choose the depth. Set the budget.
Per-repository spending, model choices, and review limits—so more automation comes with clearer costs.
07 / Open-source release planned
Look under the hood. Make it yours.
Review infrastructure deserves to be inspectable. We want teams to understand the system, extend it, and keep their options open.
Open-source release and self-hosting are being shaped.
Inside Code Voucher / architecture sketch
[/]Review orchestrationHow work moves from stage to stage
[/]Agent executionWhere investigation happens
[/]Findings & evidenceWhat the reviewer found, and why
[/]Patch proposalsHow the next change takes shape
Back to the work that needs you
Give your next change a second look.
Connect a GitLab repository. Open a merge request. Get evidence and a focused place to start your review.