From 3d71824295d71248b58d3134f7124c499bdce87c Mon Sep 17 00:00:00 2001
From: "fkrebs (via Claude)" <fkrebs@nucli.de>
Date: Tue, 26 May 2026 08:56:00 +0200
Subject: [PATCH] auto: doc-ingestion: Qdrant+TEI deployed; OWUI RAG config;
 Bifrost embed models

---
 services/doc-ingestion.md | 85 +++++++++++++++++++++++++++++----------
 1 file changed, 64 insertions(+), 21 deletions(-)

diff --git a/services/doc-ingestion.md b/services/doc-ingestion.md
index f31f128..763aa7c 100644
--- a/services/doc-ingestion.md
+++ b/services/doc-ingestion.md
@@ -1,27 +1,61 @@
-# Document Ingestion Pipeline (planned)
+# Document Ingestion Pipeline
 
-Ingest PDFs, Word, PPTX, XLSX from Nextcloud/Paperless into LobeChat's built-in
+Ingest PDFs, Word, PPTX, XLSX from Nextcloud/Paperless into Open WebUI's
 knowledge base and make them searchable via MCP.
 
 ## Architecture
 
 ```
 Nextcloud folder / Paperless webhook
-  → n8n trigger (already running on CT 104)
+  → n8n trigger (CT 104)
   → Docling MCP (port 18005) — PDF/DOCX/PPTX/XLSX → Markdown + structure
-  → LiteLLM /v1/embeddings  — Mistral-embed or text-embedding-3-large
-  → LobeChat knowledge base API  ← natively searchable in chat
-  → (optional) Qdrant sidecar — if multi-app vector search needed
+  → TEI /v1/embeddings  — multilingual-e5-base (local, 768d)
+  → Qdrant (shared vector store, http://qdrant:6333)
+  → Open WebUI knowledge base API  ← searchable in chat
 ```
 
-LobeChat already has `knowledge_base_files`, `chunks`, `embeddings` tables in its
-Postgres instance. Docling is already deployed as an MCP server on CT 104 — the missing
-piece is the n8n orchestration workflow.
-
 **Paperless shortcut**: Paperless-ngx already OCRs documents. Its full-text content is
 available at `/api/documents/?added__gt=<last_run>`. An n8n workflow can re-embed
 directly from Paperless's REST API without re-running Docling for already-OCR'd PDFs.
 
+## Deployed components (as of 2026-05-26)
+
+| Component | Location | Endpoint |
+|-----------|----------|---------|
+| **Qdrant** | CT 104, `ai-internal` net | `http://qdrant:6333` |
+| **TEI** (Text Embeddings Inference) | CT 104, `ai-internal` net | `http://tei:80` |
+| **Open WebUI** | CT 104, port 14002 | Uses Qdrant + TEI natively |
+| **Docling MCP** | CT 104, port 18005 | MCP server in gateway |
+
+### Embedding stack
+
+- **Model**: `intfloat/multilingual-e5-base` (768d, multilingual DE+EN, ~278 MB ONNX)
+- **Server**: HuggingFace TEI `cpu-1.6` — OpenAI-compatible at `http://tei:80/v1/embeddings`
+- **Auth**: none (internal network only)
+- **Upgrade path**: switch to `mistral/mistral-embed` via Bifrost (1024d, better quality) — requires re-index (drop + recreate Qdrant collections)
+
+### OWUI RAG config (env-driven)
+
+```
+VECTOR_DB=qdrant
+QDRANT_URI=http://qdrant:6333
+RAG_EMBEDDING_ENGINE=openai
+RAG_OPENAI_API_BASE_URL=http://tei:80
+RAG_OPENAI_API_KEY=none
+RAG_EMBEDDING_MODEL=intfloat/multilingual-e5-base
+CONTENT_EXTRACTION_ENGINE=docling
+CHUNK_SIZE=1200
+CHUNK_OVERLAP=150
+ENABLE_RAG_HYBRID_SEARCH=true
+```
+
+### Bifrost embedding models (available for external services / upgrade)
+
+Three embedding models tested and working via `http://bifrost:8080/v1/embeddings`:
+- `mistral/mistral-embed` (1024d) — ✓ production-ready
+- `mistral/codestral-embed` (1024d) — ✓
+- `gemini/gemini-embedding-001` (768d/1536d) — ✓
+
 ## Converter comparison
 
 | Tool | Image | Formats | Notes |
@@ -32,19 +66,28 @@ directly from Paperless's REST API without re-running Docling for already-OCR'd
 
 Start with Docling — already deployed. Add MinerU if academic paper OCR quality is needed.
 
-## MCP knowledge search
+## Sharing Qdrant with other services
 
-Add a `knowledge-search` MCP tool to the gateway that calls LobeChat's knowledge base
-search API. No new infra — LobeChat is already on `shared_backend`.
+Qdrant is on `ai-internal` network — any service on that network can use it:
 
-## Embedding options
+```python
+from qdrant_client import QdrantClient
+client = QdrantClient(url="http://qdrant:6333")
+```
 
-- **text-embedding-3-large** — already in LiteLLM, used by Coder workspaces
-- **IBM Granite embedding 30M** — tiny, fully local, CPU-only; add to LiteLLM as custom
-  provider pointing at an Ollama/IPEX-LLM instance if fully local embeddings are desired
-- **Mistral-embed / codestral-embed** — already available via LiteLLM
+n8n, MCP tools, and custom pipelines should use `http://tei:80/v1/embeddings` for
+consistent 768d vectors. Mixing models/dimensions in the same collection will fail.
 
-## Status
+## Intel Arc GPU passthrough (planned)
 
-Planned. Docling MCP is the only deployed component. n8n workflow and LobeChat API
-integration are the remaining work.
+CT 104 has Intel Core Ultra 7 155H iGPU but no `/dev/dri/render*` device is
+passed through. When passthrough is enabled, switch TEI to the Intel image:
+```
+image: ghcr.io/huggingface/text-embeddings-inference:intel-1.6
+```
+This uses IPEX and runs ~5–10× faster for embedding batches.
+
+## Pending
+
+- n8n workflow: Nextcloud/Paperless → Docling → TEI → Qdrant
+- MCP knowledge-search tool pointing at Qdrant (replaces LobeChat search intent)