One product, over ten powerful tools!
What each one does, in a line, and where it runs: on your desktop, on GitHub pull requests and CI, or both. One download, one licence, every product below.
Desktop = the macOS, Windows or Linux app and its command line · GitHub PRs / CI = the GitHub Action, the Docker image, GitLab and Jenkins
| Product | What it does | Desktop | GitHub PRs / CI |
|---|---|---|---|
| Churn measurement | Exactly what changed between two snapshots of the code: changed, deleted and added logical statements per file and per language, at any scale. The instrument everything else rides on. 62 languages. | ✓every compare mode | ✓posted on every pull request |
| TRUE_CHURN & data metrics | Generated files subtotalled out of the headline number, rule by named rule; working code split from data; REWORK and REP_CHURN, the share of a change that was wholesale replacement. | ✓same scan | ✓same scan, in the PR comment |
| Code Browser | One self-contained page that reads the code: the side-by-side diff with every statement classified and coloured, the files, the classes, and the include map drawn as a picture. | ✓opens under every run | ✓as an artifact of a scheduled run |
| AI Agent Scan | Every place your code calls AI: agent SDKs, model endpoints, exec-on-model-output, committed credentials, agent infrastructure; file and line for each, 22 languages, drawn as the Agent Map. | ✓its own mode, and with churn | ✓in the Action’s default mode |
| AI Bill of Materials & policy gate | An inventory of every AI provider, endpoint and jurisdiction the code reaches, native or CycloneDX, and a gate that fails the build on egress to CN/RU/KP/IR or an unapproved provider. | ✓command line only | ✓Action inputs bom, gate |
| Baselines & merge gate | Commit a baseline of today’s findings and fail a build only on what is new against it, so an old codebase can be gated from the first day without a clean-up. | ✓command line only | ✓Action inputs baseline, fail-on-new |
| Build-File Alert | Every build, CI, packaging and dependency file that changed between two versions, with install hooks flagged at the top, and a standing inventory of the build surface. The route the xz-utils backdoor took. | ✓every compare mode | ✓in the PR comment |
| AI Code Scan | Files scored for generated-code characteristics, HIGH, Elevated or Normal, as a review aid that says where to look first. A pointer, never a verdict on who wrote a line. | ✓AI Audit modes | ✓opt-in, mode both |
| Trend database & summary | Every run accumulates in a local database, never overwritten; a one-page summary charts churn and AI findings across releases for the people who approve the work. | ✓records every run | ✓scheduled runs, not PRs |
| AI Agent Tracer | What your coding agents did, read from the session logs they leave on the machine: what each ran, deleted, pushed and sent, to which hosts, under which model, what the human refused. Six agents’ logs. New in 2.2.0. | ✓its own mode; the logs live here | —not a pull-request tool |
| MCP server | Lets Claude Code, Cursor and other assistants run CodeDelta directly, so a question about churn since the last release returns the tool’s reproducible numbers rather than a guess. | ✓local, beside the app | — |
One download, two places to run it
Desktop. The macOS, Windows or Linux bundle. Pick a mode, point it at two snapshots, one project or a folder of agent logs, and read the report in a local window. The same bundle runs headless from a terminal or cron, which is where the AI Bill of Materials, the gate and the baseline are reached on a desktop. The command line.
GitHub pull requests and CI. The GitHub Action runs inside your own runner on every pull request: churn and agent findings posted as a comment, SARIF into the Security tab, and a merge gate when you turn it on. The same engine runs as a Docker image from GHCR, in GitLab, in Jenkins, or on a schedule for the nightly trend. Nothing leaves your repository. CLI / CI reference.
Every product above ships in one bundle under one licence, free to evaluate until 31 October 2026. Where a product is marked command line only on the desktop, the desktop bundle’s own command line runs it; the app window does not yet have a control for it.