Code review your team can follow.

Null Review prepares AI findings for self-managed GitLab changes, with saved context and controlled actions. Your team checks the reasoning and decides what happens next.

Inspect the recorded runPlan an evaluation

Local model calls and saved drafts are recorded. The corporate pilot and public software release remain open.

One change. One risk to inspect.

Synthetic walkthrough. No model call, CI or merge.

payments/query.py

Before · illustrative SHA A

  def get_payment(cursor, user_input):      sql = "SELECT * FROM payments WHERE id = "−     cursor.execute(sql + user_input)      return cursor.fetchone()

Human fix · illustrative SHA B

  def get_payment(cursor, user_input):      sql = "SELECT * FROM payments WHERE id = "+     cursor.execute(sql + "%s", [user_input])      return cursor.fetchone()

Illustrative finding payments/query.py:3 · SHA A

SQL injection risk

A user-controlled value becomes SQL syntax. Input may change the query predicate and expose unintended rows. Pass the value separately as a bound parameter.

Input becomes SQL syntax

A user-controlled value becomes SQL syntax.

A developer binds the value

Check the input boundary, apply parameter binding and commit the fix.

New SHA, separate checks

SHA B needs a fresh review. Its model result and CI are not run here. Approval and merge remain separate team actions.

The %s placeholder assumes a compatible Python driver. Use your driver’s binding syntax. A and B are illustrative labels, not commit hashes.

New review / SHA B
Not run
CI / SHA B
Not run
Approval / merge
Not performed; team decision
Inspect the recorded example

A real run. Educational code.

Recorded on a MacBook, 16 November 2026. A local GitLab and an external OpenAI model.

The recorded outcome

SHA
8a87c4f9
Run
9f90f98e
State
draft_ready
Findings
3

The answer reached a saved draft

A real model reviewed a synthetic SQL-injection change. After a local credential-fingerprint patch, the run completed with three findings. The patch addressed a workflow conflict, not the SQL code.

This public record is a summary of the saved acceptance report. It does not contain the original MR, full finding text or cabinet footage. It is not a live demo or an independent rerun.

The patches are not an accepted release. Publication, fix, new-SHA review, CI and the team decision still need a complete acceptance run.

Read the local results and limitsPublic record · JSON

From a diff to a decision.

Null keeps the review tied to the code revision and rules used for it. A changed revision needs a new check before acting on the result.

  1. Prepare context

    Select the revision, permitted code and rules. Save the review context as a ContextPack.

  2. Analyse the change

    The configured review model prepares draft findings from that context.

  3. Check the finding

    A reviewer checks the code, reasoning and practical consequence.

  4. Fix and recheck

    Review the new revision. Read CI results separately. The team decides whether to merge.

Trace the review workflow

A finding for the developer. A trail for the team.

Start with the code question. Then inspect the context, access and actions behind the answer.

Developer
Inspect a finding beside the change. Check why it matters before choosing a fix.
Engineering lead
Follow review, recheck and CI as separate evidence. Evaluate useful findings and review effort in a pilot.
Platform and security
Inspect repository scopes, service data and external model routes before connecting code.

Your deployment. Explicit model routes.

Null services and their PostgreSQL data run in your environment. An external review model receives the selected ContextPack. If external embeddings are configured, that endpoint receives indexed code fragments.

Your environment

Null Review

Services + PostgreSQL
ContextPack

Selected review context

Indexed fragments

If external embeddings are configured

Configured endpoints

Review model

Embedding model

The seven-service quickstart creates one tenant. Qdrant, ContextForge, SSO and a CI runner require separate setup and acceptance.
Follow the dataInspect access boundaries

Inspect what has been checked.

The checks below have different scopes. They do not establish pilot readiness or model accuracy.

16 Nov 2026 · GitLab and model
Assignment, mention and CI through the coordinator reached real model calls and saved drafts in local checks. CI is a separate synthetic case with four findings, not a completed fix/re-review path.
16 Nov 2026 · Restart
Saved records and Worker attempts matched after a normal restart, without repeated model calls. Backup/restore and five overlapping reviews were not tested.
16 Nov 2026 · Search in shadow
A fixture prepared five support fragments. The review model received only the diff. Retrieved context has not demonstrated a quality or cost improvement.
Historical MR3 · Separate evidence
One SQL finding, then zero after a fix; operator-assisted CI. Overall task result INCONCLUSIVE. The MR remained open.
Read the dated local evidenceOpen MR3 and its limits

Already evaluating Open Code Review?

Null focuses on the workflow around a finding: saved context, scoped data, controlled GitLab actions and decision history. Its current review engine is PR-Agent; Null is not an OCR add-on.

Review quality and integration effort need a matched evaluation. Null has no published superiority result.

Compare the workflow scope

One product. Choose who runs it.

Self-hosted and Enterprise use the same Null product. The difference is responsibility for deployment and support.

Self-hosted

Your team operates the services, credentials, model route and updates. Start by reviewing the source and installation requirements.

Review source access

Enterprise

Agree the environment, roles, rollout and support scope together. A pilot needs explicit acceptance criteria.

Review the pilot path

The public software release is being prepared. Full MR/fix/re-review/CI acceptance and five concurrent reviews remain open.

Start with one repository and a clear question.

Bring your target OS, GitLab edition, model route and technical owner. Agree what a useful finding looks like and how the team will evaluate it.

Request evaluation sourcePrepare a pilot brief

Read the requirements before sharing code or credentials.