AI Documentation
Two shipped AI capabilities — both read-only, grounded in real data, never inventing answers. Everything else is roadmap.
1 · Lint Explanation
The Designer linter flags problems (quiet zones, font sizes, overflow). The AI explains why each finding matters and how to fix it — in plain language.
2 · Log Q&A
Ask questions about your print logs: "Which jobs failed yesterday?", "Why did runner linz-01 go offline?". Answers are grounded exclusively in your actual audit trail — if the data isn’t there, the AI says so instead of inventing an answer.
Fundament: deklaratives JSON als Tool-Use-Ziel
label.json and flow.json are schema-versioned, validatable and patchable — exactly what an LLM tool call needs. No proprietary binary format for a model to guess at.
POST /api/v1/ai/lint-erklaerung
{
"finding": {
"object": "bc_sscc",
"rule": "quiet_zone.min",
"value": 1.2,
"required": 2.0
},
"label_id": "lbl_8f2k3a"
}
// Response:
{
"explanation": "The quiet zone of 1.2 mm is below the\nGS1 requirement of 2.0 mm (10 x module width).\nScanners cannot reliably read the barcode otherwise.",
"fix": "Increase quiet zone to 2.0 mm",
"grounded": true
}POST /api/v1/ai/log-qa
{
"question": "Which jobs failed yesterday?",
"time_range": "yesterday"
}
// Response:
{
"answer": "2 jobs failed: job 7f21 (PAL-1,\nribbon empty) and job 8a03 (VER-3, timeout).\nBoth went to the error path.",
"sources": ["audit-log/entry/9912", "audit-log/entry/9915"],
"grounded": true
}Beide Endpunkte verlangen Authentifizierung (Bearer Token) und respektieren Rollen: Operator sehen nur ihre eigenen Drucke, Admins alles.
grounded: true/false. When false, the response only states that the data is insufficient.