docs(search): add ingest-content Claude command

.claude/commands/search/ingest-content.md

@@ -0,0 +1,66 @@
+# Ingest Content to Hi-RAG
+
+Ingest text content into Hi-RAG v2 for knowledge retrieval.
+
+## Pipeline Paths
+
+### Path 1: YouTube → PMOVES.YT → Extract Worker → Qdrant
+```bash
+# Step 1: Ingest video (downloads + transcribes)
+curl -X POST http://localhost:8077/yt/ingest \
+  -H "Content-Type: application/json" \
+  -d '{"url": "https://www.youtube.com/watch?v=VIDEO_ID"}'
+
+# Step 2: Push transcript to extract-worker (chunks + embeds + indexes)
+curl -X POST http://localhost:8083/ingest \
+  -H "Content-Type: application/json" \
+  -d '{"chunks": [{"text": "...", "chunk_id": "unique-id"}]}'
+
+# Step 3: Query via Hi-RAG
+curl -X POST http://localhost:8086/hirag/query \
+  -H "Content-Type: application/json" \
+  -d '{"query": "search terms", "top_k": 5}'
+```
+
+### Path 2: Direct text → Extract Worker
+```bash
+curl -X POST http://localhost:8083/ingest \
+  -H "Content-Type: application/json" \
+  -d '{"chunks": [{"text": "content here", "chunk_id": "doc-1"}]}'
+```
+
+## Known Issues
+
+- Extract-worker requires TensorZero + Ollama for Qwen3 embeddings
+- If TensorZero embedding fails (500), fall back to `EXTRACT_WORKER_EMBEDDING_BACKEND=sentence-transformers`
+- `QDRANT_COLLECTION` must match between extract-worker and Hi-RAG (both should be `pmoves_chunks_qwen3`)
+- PMOVES.YT downloads + transcribes but does NOT auto-trigger extract-worker — manual POST required
+
+## Dependencies
+
+| Service | Port | Role |
+|---------|------|------|
+| PMOVES.YT | 8077 | YouTube download + transcribe |
+| Extract Worker | 8083 | Chunk + embed + index |
+| Hi-RAG v2 | 8086 | Query (hybrid search) |
+| Qdrant | 6333 | Vector storage (Docker internal) |
+| Meilisearch | 7700 | Full-text search (Docker internal) |
+| TensorZero | 3030 | Embedding model routing |
+| Ollama | 11434 | Local embedding model (qwen3-embedding:4b) |
+
+## HuggingFace Alternatives (No TensorZero dependency)
+
+For environments without TensorZero/Ollama, use direct HuggingFace models:
+- `BAAI/bge-large-en-v1.5` (1024d, high quality)
+- `sentence-transformers/all-MiniLM-L6-v2` (384d, fast)
+- `Alibaba-NLP/gte-Qwen2-1.5B-instruct` (1536d, multilingual)
+
+Set `EXTRACT_WORKER_EMBEDDING_BACKEND=sentence-transformers` in env to bypass TensorZero.
+
+## Make Targets
+
+```bash
+make -C pmoves up-hirag       # Start Hi-RAG v2
+make -C pmoves up-workers     # Start extract-worker + media workers
+make -C pmoves brand-defaults # Set QDRANT_COLLECTION default
+```