AI & automation · read from upstream 2026-10-19
What the Craft suite can actually automate today
“AI-powered” is the easiest claim to make and the hardest to verify in a search result dominated by mirror sites. This page lists only what is visible in the storytold organisation, states how mature each piece is, and names what we could not confirm.
In the open
Repositories that exist right now
| Repository | What it is | How to treat it |
|---|---|---|
| craft-ai-integrations ↗ | “Interim AI layer for the Craft apps: connections, keys, providers, client, egui widgets (first draft)” | Early stage — the project labels it a first draft |
| craft-common-actions ↗ | Shared actions used across the Craft apps — the closest thing to a common automation surface. | Read the source before depending on it |
| craft-libs ↗ | Rust libraries shared by the apps; where headless or batch reuse would live. | Active |
| pdfcraft (release assets) ↗ | The PdfCraft release ships pdfcraft-cli builds for Linux x86_64 and aarch64, alongside the app. | Ships in the release |
| photocraft-corpus ↗ | Test data for PhotoCraft — the honest way to judge PSD compatibility claims. | Test data |
| craft-repo-status-app ↗ | Public view of PR and build status across the suite. | Active |
Descriptions are quoted or summarised from the repository's own GitHub description. The AI layer is explicitly interim — plan against its current state, not against a roadmap.
Practical automation that needs no vendor service
- Batch PDF work from a script with pdfcraft-cli — Linux x86_64 and aarch64 tarballs, plus a macOS universal build.
- Every app's release carries its own command-line build (macOS universal), so a pipeline can call the app instead of driving its UI.
- Pin a specific release in CI and verify it against SHA256SUMS.txt before running it.
- Build from source for the exact crate versions your pipeline depends on.
- Run document work fully offline where confidentiality — legal, medical, client NDA — rules out a cloud endpoint.
- Self-host FigCraft's companion server build if you need the collaborative side on your own infrastructure.
Why local-first matters for AI work
The suite's apps keep files on your disk, which changes what an assistant integration can be: instead of uploading a client's source document to somebody's hosted feature, the AI layer upstream is built around providers and keys you configure yourself. That is a different cost-and-control trade-off from a subscription suite's built-in assistant, and it is the main reason an open-source stack is worth evaluating for confidential material.
Next steps