unify parse api

2026-09-18 04:35:42 +02:00
parent adb529d14f
commit bbd285109d
5 changed files with 491 additions and 443 deletions
--- a/docs/api/routes/delivery/parse.md
+++ b/docs/api/routes/delivery/parse.md
@@ -0,0 +1,90 @@
+# `POST /delivery/parse`
+
+Protected. Requires a valid session cookie (`credentials: 'include'`).
+
+Parse document (SSE)
+
+Accepts an invoice or delivery note as PDF or text file
+(`multipart/form-data`, field `file`) and streams extraction results as
+Server-Sent Events until parsing is complete.
+
+PDFs are read via `pdftotext -layout`; text files are decoded as UTF-8.
+Pages are separated by the form feed character (`\f`). If no form feed
+is present, the input is treated as a single page.
+
+The extracted pages are fed to an LLM (OpenRouter) which fills a structured
+manifest of header data and line items. The OpenRouter API key is read from
+the authenticated user's company setting `keys.openrouter`; the endpoint
+returns `400` if the key is not configured.
+
+The stream ends after the final `result` event (or an `error` event on failure).
+
+
+## Body (`multipart/form-data`)
+
+| Field | Required | Type | Description |
+|-------|----------|------|-------------|
+| `file` | yes | file | Document to parse (max 50 MB, application/pdf or text file) |
+
+## SSE Events
+
+The response is a `text/event-stream`. Events are emitted in order until the stream closes.
+
+| Event | Description | Payload |
+|-------|-------------|---------|
+| `status` | Document parse confirmation and LLM status messages | `{ "message": "…" }` |
+| `llm_call` | Fired before each LLM request | `{ "model": "…", "promptSize": 0, "messages": 0 }` |
+| `tool_call` | Fired when the output_manifest tool is invoked | `{ "name": "output_manifest" }` |
+| `usage` | Token usage after each LLM response | `{ "input": 0, "cached": 0, "output": 0, "cost": 0 }` |
+| `progress` | Intermediate accumulated extraction state (basedata / notes / items) | `ManifestResult` |
+| `result` | Final accumulated extraction — emitted just before the stream closes | `ManifestResult` |
+| `error` | Emitted on document parse or LLM failure; stream closes immediately after | `{ "message": "…" }` |
+
+## Response `200`
+
+`text/event-stream` — emits `status`, `llm_call`, `tool_call`, `usage`,
+`progress`, `result` (and `error` on failure). See SSE Events table below.
+
+**ManifestResult shape** (used by `progress` and `result`):
+
+```json
+{
+  "basedata": {
+    "date": "2026-01-15",
+    "kLieferant": 42,
+    "lieferantName": "Muster GmbH",
+    "manifestNr": "LS-2026-001",
+    "manifestType": "lieferschein"
+  },
+  "notes": ["Pos. 3: VPE-Angabe fehlte, Standardwert 1 angenommen."],
+  "items": {
+    "pagesFrom": 1,
+    "pagesTo": 2,
+    "items": [
+      {
+        "position": 1,
+        "count": 10,
+        "vpe": 1,
+        "unit": "Stk.",
+        "articleName": "Einhängesieb Stahl",
+        "liefArtNr": "ES-100",
+        "mwst": "19",
+        "eknetto": 450,
+        "page": 1
+      }
+    ]
+  }
+}
+```
+
+
+## Errors
+
+| Status | Description |
+|--------|-------------|
+| 400 | No file uploaded, unsupported file type, or OpenRouter key not configured — `{"error":"…"}` |
+| 401 | Unauthorized — `{"error":"…"}` |
+
+---
+
+_Generated from `@openapi` JSDoc. Do not edit by hand — run `npm run docs:api`._