fix: replace exact-phrase wrap with AND-required-terms in /search

Wrapping every multi-word phrase in exact-phrase quotes (added for the
"uyuşturucu madde ticareti" false-positive problem) was too strict for
independent keyword queries like "bıçak yaralaması beraat" or the AI
query-optimizer's extracted keywords — those words rarely appear
verbatim adjacent to each other, so exact-phrase returned 0 results.

Now each word is prefixed with + instead (AND semantics: all words
must appear somewhere in the decision, not necessarily adjacent).
This still fixes the original "madde" noise problem while no longer
breaking loose multi-keyword searches.

.env.example

@@ -75,11 +75,12 @@ JWT_SECRET_KEY=your_jwt_secret_key_here
 # =============================================================================
 
 # Embedding provider for the semantic_search tool.
-# Pick exactly one of: OpenRouter (hosted) or Local (your own server).
+# Pick exactly one of: OpenRouter (hosted), OrcaRouter (hosted), or Local.
 
 # --- Option A: OpenRouter (hosted, default) -----------------------------------
 # Get your API key from: https://openrouter.ai/keys
-# If neither this nor EMBEDDING_PROVIDER=local is set, semantic search is off.
+# If neither this, nor ORCAROUTER_API_KEY, nor EMBEDDING_PROVIDER=local is set,
+# semantic search is off.
 OPENROUTER_API_KEY=sk-or-v1-your_openrouter_api_key_here
 
 # Optional: override the OpenRouter embedding model and dimension.
@@ -89,6 +90,17 @@ OPENROUTER_API_KEY=sk-or-v1-your_openrouter_api_key_here
 # OPENROUTER_EMBEDDING_MODEL=google/gemini-embedding-001
 # OPENROUTER_EMBEDDING_DIMENSION=3072
 
+# --- Option A2: OrcaRouter (hosted) ------------------------------------------
+# OrcaRouter is a production AI gateway with one OpenAI-compatible endpoint
+# (https://api.orcarouter.ai/v1). Get your API key from: https://www.orcarouter.ai
+# Set ORCAROUTER_API_KEY instead of OPENROUTER_API_KEY to use it.
+# ORCAROUTER_API_KEY=sk-orca-your_orcarouter_api_key_here
+
+# Optional: override the OrcaRouter embedding model and dimension.
+# Defaults: google/gemini-embedding-001 at 3072 dims (multilingual).
+# ORCAROUTER_EMBEDDING_MODEL=google/gemini-embedding-001
+# ORCAROUTER_EMBEDDING_DIMENSION=3072
+
 # --- Option B: Local OpenAI-compatible server (no API key required) ----------
 # Recommended for Turkish: intfloat/multilingual-e5-large served by HuggingFace
 # Text Embeddings Inference (TEI). One-line setup:

Dockerfile

@@ -18,6 +18,7 @@ COPY README.md ./
 COPY app.py ./
 COPY asgi_app.py ./
 COPY mcp_server_main.py ./
+COPY rest_api.py ./
 
 # Copy MCP modules and shared packages
 COPY anayasa_mcp_module ./anayasa_mcp_module
@@ -51,4 +52,4 @@ HEALTHCHECK --interval=30s --timeout=10s --start-period=40s --retries=3 \
     CMD python -c "import httpx; httpx.get('http://localhost:8000/health', timeout=5)" || exit 1
 
 # Run the ASGI application
-CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "8000"]
+CMD ["uvicorn", "rest_api:app", "--host", "0.0.0.0", "--port", "8000"]

README.md

@@ -1,5 +1,7 @@
 # Yargı MCP: Türk Hukuk Kaynakları için MCP Sunucusu
 
+[![MCP Toplist](https://mcptoplist.com/badge/glama%2Fsaidsurucu%2Fyargi-mcp.svg)](https://mcptoplist.com/server/glama%2Fsaidsurucu%2Fyargi-mcp)
+
 > ## ✨ Profesyonel Sürüm Hazır: Yargı MCP Pro
 >
 > **Mevzuat ve içtihatı tek bir MCP sunucusunda birleştiren** profesyonel sürüm yayında:
@@ -275,7 +277,7 @@ Yargı MCP'yi Gemini CLI ile kullanmak için:
 Yargı MCP, **semantik arama** özelliği ile kararları anlamsal olarak sıralayabilir. Opsiyoneldir; iki yoldan biri yapılandırıldığında otomatik etkinleşir:
 
 - **Yerel** (önerilen, ücretsiz): kendi makinenizdeki OpenAI-uyumlu embedding sunucusu (HuggingFace TEI, llama.cpp, Ollama, vLLM, LM Studio…)
-- **Hosted**: OpenRouter API anahtarı
+- **Hosted**: OpenRouter ya da [OrcaRouter](https://www.orcarouter.ai) API anahtarı
 
 ### Semantik Arama Nasıl Çalışır?
 1. `initial_keyword` ile Bedesten API'den 100 karar çekilir
@@ -351,11 +353,25 @@ OPENROUTER_API_KEY=sk-or-v1-xxx...
 
 API anahtarınızı [openrouter.ai/keys](https://openrouter.ai/keys) adresinden alın. Varsayılan model `google/gemini-embedding-001` artık ücretli — ücretsiz bir model seçerseniz `OPENROUTER_EMBEDDING_MODEL`, `OPENROUTER_EMBEDDING_DIMENSION` ve uygun `EMBEDDING_PROMPT_STYLE` değerlerini birlikte ayarlayın.
 
+### Alternatif 3: OrcaRouter (hosted)
+
+[OrcaRouter](https://www.orcarouter.ai), 200+ modeli tek OpenAI-uyumlu uçta toplayan bir üretim AI ağ geçididir (ağ geçidi seviyesinde, sıfır-güven AI ajan güvenliği de içerir). Mevcut SDK kodu `base_url` değiştirilerek aynen çalışır.
+
+```bash
+ORCAROUTER_API_KEY=sk-orca-xxx...
+# İsteğe bağlı — varsayılan google/gemini-embedding-001 (3072 dim, çok dilli)
+# ORCAROUTER_EMBEDDING_MODEL=...
+# ORCAROUTER_EMBEDDING_DIMENSION=...
+# EMBEDDING_PROMPT_STYLE=gemini   # varsayılan
+```
+
+API anahtarınızı [www.orcarouter.ai](https://www.orcarouter.ai) adresinden alın. `OPENROUTER_API_KEY` yerine `ORCAROUTER_API_KEY` ayarlamanız yeterli — semantik arama aynı OpenAI-uyumlu akışı OrcaRouter ucu üzerinden kullanır.
+
 ### Yapılandırma Referansı
 
 | Env Var | Açıklama | Örnek |
 |---|---|---|
-| `EMBEDDING_PROVIDER` | `local` ise yerel sunucu, boş ise OpenRouter | `local` |
+| `EMBEDDING_PROVIDER` | `local` ise yerel sunucu, boş ise hosted (OpenRouter/OrcaRouter) | `local` |
 | `EMBEDDING_PROMPT_STYLE` | `gemini` / `e5` / `raw` — modelin beklediği önek | `e5` |
 | `LOCAL_EMBEDDING_BASE_URL` | Yerel sunucunun OpenAI-uyumlu URL'i | `http://localhost:8080/v1` |
 | `LOCAL_EMBEDDING_MODEL` | Model adı | `intfloat/multilingual-e5-large` |
@@ -363,6 +379,9 @@ API anahtarınızı [openrouter.ai/keys](https://openrouter.ai/keys) adresinden
 | `OPENROUTER_API_KEY` | OpenRouter anahtarı (sadece hosted için) | `sk-or-v1-…` |
 | `OPENROUTER_EMBEDDING_MODEL` | OpenRouter model id'si | `google/gemini-embedding-001` |
 | `OPENROUTER_EMBEDDING_DIMENSION` | OpenRouter modelinin çıktı boyutu | `3072` |
+| `ORCAROUTER_API_KEY` | OrcaRouter anahtarı (sadece hosted için) | `sk-orca-…` |
+| `ORCAROUTER_EMBEDDING_MODEL` | OrcaRouter model id'si | `google/gemini-embedding-001` |
+| `ORCAROUTER_EMBEDDING_DIMENSION` | OrcaRouter modelinin çıktı boyutu | `3072` |
 
 > 💡 **Not:** Hiçbir embedding sağlayıcı yapılandırılmazsa semantik arama aracı görünmez, diğer 28 araç normal şekilde çalışır.
 

mcp_server_main.py

@@ -266,14 +266,24 @@ from bedesten_mcp_module.models import (
 from bedesten_mcp_module.enums import BirimAdiEnum
 
 # Semantic Search Module Imports (enabled if any embedding provider is configured)
-from semantic_search.embedder import is_semantic_search_available, is_local_embedding_configured
+from semantic_search.embedder import (
+    is_semantic_search_available,
+    is_local_embedding_configured,
+    is_openrouter_available,
+    is_orcarouter_available,
+)
 SEMANTIC_SEARCH_AVAILABLE = is_semantic_search_available()
 
 if SEMANTIC_SEARCH_AVAILABLE:
     from semantic_search.embedder import get_embedder
     from semantic_search.vector_store import VectorStore
     from semantic_search.processor import DocumentProcessor
-    provider = "local" if is_local_embedding_configured() else "openrouter"
+    if is_local_embedding_configured():
+        provider = "local"
+    elif is_orcarouter_available():
+        provider = "orcarouter"
+    elif is_openrouter_available():
+        provider = "openrouter"
     logger.info(f"Semantic search enabled (provider={provider})")
 else:
     logger.info("Semantic search disabled (no embedding provider configured)")

rest_api.py

@@ -0,0 +1,243 @@
+# rest_api.py
+#
+# Bu dosya, yargi-mcp fork'unun kok dizinine eklenir (mcp_server_main.py ile ayni seviyede).
+# Amac: FastMCP/MCP protokolunu (JSON-RPC) tamamen atlayip, mcp_server_main.py'deki
+# gercek client nesnelerini (bedesten_client_instance vb.) dogrudan cagiran duz bir
+# REST API sunmak. Boylece Node backend, MCP istemcisi konusmadan basit HTTP ile
+# bu servise istek atabilir.
+#
+# Calistirma:
+#   uvicorn rest_api:app --host 0.0.0.0 --port 8001
+#
+# NOT: mcp_server_main.py'deki bedesten_client_instance nesnesini yeniden olusturmak
+# yerine dogrudan ayni siniflari import edip burada kendi instance'imizi kuruyoruz.
+# Boylece mcp_server_main.py'yi (FastMCP/app.tool decorator'lariyla) hic calistirmaya
+# gerek kalmiyor, sadece alttaki client katmanini kullaniyoruz.
+
+from fastapi import FastAPI, HTTPException, Query
+from pydantic import BaseModel, Field
+from typing import List, Optional
+import logging
+import re
+
+from bedesten_mcp_module.client import BedestenApiClient, BedestenRateLimited
+from bedesten_mcp_module.models import (
+    BedestenSearchRequest,
+    BedestenSearchData,
+    BedestenCourtTypeEnum,
+)
+from bedesten_mcp_module.enums import BirimAdiEnum
+import httpx
+
+logging.basicConfig(level=logging.INFO)
+logger = logging.getLogger("rest_api")
+
+app = FastAPI(title="LegalOS - Yargi Karar Arama REST API")
+
+bedesten_client = BedestenApiClient()
+
+
+# --- Request/Response semalari ---
+
+class SearchRequest(BaseModel):
+    phrase: str = Field(..., description="Arama ifadesi (Turkce). Ornek: 'mulkiyet hakki'")
+    court_types: List[str] = Field(
+        default=["YARGITAYKARARI", "DANISTAYKARAR"],
+        description="YARGITAYKARARI, DANISTAYKARAR, YERELHUKUK, ISTINAFHUKUK, KYB",
+    )
+    page_number: int = Field(default=1, ge=1)
+    birim_adi: str = Field(default="ALL", description="Daire filtresi, orn. H1, C3, HGK")
+    karar_tarihi_start: Optional[str] = Field(default=None, description="YYYY-MM-DD")
+    karar_tarihi_end: Optional[str] = Field(default=None, description="YYYY-MM-DD")
+    exact_phrase: bool = Field(
+        default=True,
+        description=(
+            "True ise coklu kelimeli phrase'deki her kelime otomatik olarak '+' ile zorunlu "
+            "kilinir (orn. 'a b' -> '+a +b') — tum kelimeler karar icinde gecmeli ama yan yana "
+            "olmalari gerekmez. Bedesten API'de tirnaksiz coklu kelime aramasi kelimeleri gevsek "
+            "eslestirir (orn. 'madde' gibi her kararda gecen ortak kelimeler alakasiz sonuclari "
+            "one cikarir); bu ayar bu gurultuyu onler. Kullanici zaten tirnak/AND/OR/NOT/+/- "
+            "kullaniyorsa dokunulmaz."
+        ),
+    )
+
+
+class SearchResultItem(BaseModel):
+    document_id: str
+    raw: dict
+
+
+class SearchResponse(BaseModel):
+    decisions: List[dict]
+    total_records: int
+    requested_page: int
+    page_size: int
+    searched_courts: List[str]
+    error: Optional[str] = None
+    retry_after: Optional[float] = None
+
+
+class DocumentResponse(BaseModel):
+    document_id: str
+    markdown_content: Optional[str]
+    source_url: Optional[str]
+    mime_type: Optional[str]
+
+
+# --- Yardimci fonksiyon: tarih formatlama (mcp_server_main.py'deki mantikla ayni) ---
+
+def _format_date(value: Optional[str], end_of_day: bool = False) -> str:
+    if not value:
+        return ""
+    if value.endswith("Z"):
+        return value
+    if "T" not in value:
+        suffix = "T23:59:59.999Z" if end_of_day else "T00:00:00.000Z"
+        return f"{value}{suffix}"
+    return value
+
+
+# --- Yardimci fonksiyon: coklu kelimeli aramalari daha isabetli hale getirme ---
+#
+# Bedesten API'de tirnaksiz coklu kelime aramasi kelimeleri ayri ayri (gevsek) eslestiriyor.
+# Ornek: "uyusturucu madde ticareti" -> "madde" gibi her kararda gecen (kanun maddesi
+# anlaminda) ortak bir kelime yuzunden alakasiz sonuclar (ic icra/iflas kararlari) one
+# cikabiliyor.
+#
+# Once tum ifadeyi tirnaklayip "tam bitisik ifade" aramasi denendi, ama bu cok kati
+# cikti: "bicak yaralamasi beraat" gibi bagimsiz anahtar kelimelerden olusan (AI'nin
+# uzun sorulardan cikardigi turden) aramalar, bu 3 kelime kararlarda hic yan yana/aynen
+# gecmedigi icin 0 sonuc donduruyordu. Bunun yerine her kelimeyi "+" ile ayri ayri
+# zorunlu kiliyoruz (AND semantigi): tum kelimeler karar icinde herhangi bir yerde
+# gecmeli ama bitisik/ayni sirada olmalari gerekmiyor. Bu hem orijinal "madde" sorununu
+# cozuyor (uyusturucu VE madde VE ticaret hepsi gecmeli) hem de bagimsiz anahtar kelime
+# aramalarini kirmiyor.
+_OPERATOR_PATTERN = re.compile(r'"|\bAND\b|\bOR\b|\bNOT\b|(?:^|\s)[+-]\S', re.IGNORECASE)
+
+
+def _apply_required_terms(phrase: str, exact_phrase: bool) -> str:
+    stripped = phrase.strip()
+    if not exact_phrase or not stripped:
+        return phrase
+    words = stripped.split()
+    if len(words) < 2:
+        return phrase
+    if _OPERATOR_PATTERN.search(stripped):
+        return phrase
+    return " ".join(f"+{w}" for w in words)
+
+
+# --- Endpoint 1: Arama ---
+
+@app.post("/search", response_model=SearchResponse)
+async def search(req: SearchRequest):
+    """
+    Yargitay / Danistay / yerel mahkeme / istinaf / KYB kararlarinda arama yapar.
+    mcp_server_main.py'deki search_bedesten_unified aracinin dogrudan REST karsiligi.
+    """
+    karar_tarihi_start = _format_date(req.karar_tarihi_start)
+    karar_tarihi_end = _format_date(req.karar_tarihi_end, end_of_day=True)
+    phrase = _apply_required_terms(req.phrase, req.exact_phrase)
+
+    search_data = BedestenSearchData(
+        pageSize=10,
+        pageNumber=req.page_number,
+        itemTypeList=req.court_types,
+        phrase=phrase,
+        birimAdi=req.birim_adi,
+        kararTarihiStart=karar_tarihi_start,
+        kararTarihiEnd=karar_tarihi_end,
+    )
+    search_request = BedestenSearchRequest(data=search_data)
+
+    logger.info(f"search: phrase={req.phrase!r} -> sent={phrase!r} courts={req.court_types} page={req.page_number}")
+
+    try:
+        response = await bedesten_client.search_documents(search_request)
+
+        if response.data is None:
+            return SearchResponse(
+                decisions=[],
+                total_records=0,
+                requested_page=req.page_number,
+                page_size=10,
+                searched_courts=req.court_types,
+                error="no_data",
+            )
+
+        decisions = response.data.emsalKararList or []
+        total = response.data.total or 0
+
+        return SearchResponse(
+            decisions=[d.model_dump() for d in decisions],
+            total_records=total,
+            requested_page=req.page_number,
+            page_size=10,
+            searched_courts=req.court_types,
+        )
+
+    except BedestenRateLimited as e:
+        logger.warning(f"local rate limit hit, retry_after={e.retry_after}")
+        raise HTTPException(
+            status_code=429,
+            detail={"error": "rate_limit_exceeded", "retry_after": e.retry_after},
+        )
+    except httpx.HTTPStatusError as e:
+        if e.response.status_code == 429:
+            retry_after = e.response.headers.get("Retry-After", "")
+            raise HTTPException(
+                status_code=429,
+                detail={"error": "upstream_rate_limit", "retry_after": retry_after},
+            )
+        logger.exception("Bedesten search error")
+        raise HTTPException(status_code=502, detail="Bedesten API hatasi")
+    except Exception:
+        logger.exception("Beklenmeyen arama hatasi")
+        raise HTTPException(status_code=500, detail="Sunucu hatasi")
+
+
+# --- Endpoint 2: Belge getirme (tam metin, Markdown) ---
+
+@app.get("/document/{document_id}", response_model=DocumentResponse)
+async def get_document(document_id: str):
+    """
+    Bir kararin tam metnini Markdown formatinda getirir.
+    mcp_server_main.py'deki get_bedesten_document_markdown aracinin REST karsiligi.
+    """
+    if not document_id.strip():
+        raise HTTPException(status_code=400, detail="document_id bos olamaz")
+
+    logger.info(f"get_document: id={document_id}")
+
+    try:
+        doc = await bedesten_client.get_document_as_markdown(document_id)
+        return DocumentResponse(
+            document_id=document_id,
+            markdown_content=doc.markdown_content,
+            source_url=doc.source_url,
+            mime_type=doc.mime_type,
+        )
+    except BedestenRateLimited as e:
+        raise HTTPException(
+            status_code=429,
+            detail={"error": "rate_limit_exceeded", "retry_after": e.retry_after},
+        )
+    except httpx.HTTPStatusError as e:
+        if e.response.status_code == 429:
+            retry_after = e.response.headers.get("Retry-After", "")
+            raise HTTPException(
+                status_code=429,
+                detail={"error": "upstream_rate_limit", "retry_after": retry_after},
+            )
+        logger.exception("Bedesten document fetch error")
+        raise HTTPException(status_code=502, detail="Bedesten API hatasi")
+    except Exception:
+        logger.exception("Beklenmeyen belge getirme hatasi")
+        raise HTTPException(status_code=500, detail="Sunucu hatasi")
+
+
+# --- Endpoint 3: Health check (Coolify icin) ---
+
+@app.get("/health")
+async def health():
+    return {"status": "ok"}

semantic_search/__init__.py

@@ -2,9 +2,11 @@
 
 from .embedder import (
     OpenRouterEmbedder,
+    OrcaRouterEmbedder,
     LocalEmbedder,
     get_embedder,
     is_openrouter_available,
+    is_orcarouter_available,
     is_local_embedding_configured,
     is_semantic_search_available,
 )
@@ -13,9 +15,11 @@ from .processor import DocumentProcessor
 
 __all__ = [
     'OpenRouterEmbedder',
+    'OrcaRouterEmbedder',
     'LocalEmbedder',
     'get_embedder',
     'is_openrouter_available',
+    'is_orcarouter_available',
     'is_local_embedding_configured',
     'is_semantic_search_available',
     'VectorStore',

semantic_search/embedder.py

@@ -64,6 +64,11 @@ def is_openrouter_available() -> bool:
     return bool(os.getenv("OPENROUTER_API_KEY"))
 
 
+def is_orcarouter_available() -> bool:
+    """Check if OrcaRouter API key is available."""
+    return bool(os.getenv("ORCAROUTER_API_KEY"))
+
+
 def is_local_embedding_configured() -> bool:
     """Check if the user opted into a local embedding endpoint."""
     return os.getenv("EMBEDDING_PROVIDER", "").strip().lower() == "local"
@@ -71,7 +76,11 @@ def is_local_embedding_configured() -> bool:
 
 def is_semantic_search_available() -> bool:
     """Returns True if any embedding provider is configured."""
-    return is_local_embedding_configured() or is_openrouter_available()
+    return (
+        is_local_embedding_configured()
+        or is_openrouter_available()
+        or is_orcarouter_available()
+    )
 
 
 def _coerce_dimension(value, env_name: str, default: int) -> int:
@@ -261,6 +270,59 @@ class OpenRouterEmbedder(_BaseOpenAICompatibleEmbedder):
         )
 
 
+class OrcaRouterEmbedder(_BaseOpenAICompatibleEmbedder):
+    """
+    Embedder using OrcaRouter's OpenAI-compatible embedding API.
+
+    OrcaRouter is a production AI gateway that proxies 200+ models on a single
+    OpenAI-compatible endpoint. The model and dimension are configurable so
+    users can pick any embedding model the gateway routes. Configuration
+    precedence: explicit constructor args > environment variables > defaults.
+
+    Environment variables:
+        ORCAROUTER_API_KEY (required): OrcaRouter credential (sk-orca-...)
+        ORCAROUTER_EMBEDDING_MODEL (optional): override the embedding model id
+        ORCAROUTER_EMBEDDING_DIMENSION (optional): override the vector size
+
+    Defaults: ``google/gemini-embedding-001`` at 3072 dimensions (multilingual,
+    matches the OpenRouter default — good for Turkish legal text).
+    """
+
+    def __init__(
+        self,
+        model: Optional[str] = None,
+        dimension: Optional[int] = None,
+        prompt_style: Optional[str] = None,
+    ):
+        api_key = os.getenv("ORCAROUTER_API_KEY")
+        if not api_key:
+            raise ValueError("ORCAROUTER_API_KEY environment variable is not set")
+
+        try:
+            from openai import OpenAI
+        except ImportError:
+            raise ImportError("openai package is required. Install with: pip install openai")
+
+        self.client = OpenAI(
+            base_url="https://api.orcarouter.ai/v1",
+            api_key=api_key,
+        )
+        self.model = model or os.getenv("ORCAROUTER_EMBEDDING_MODEL") or DEFAULT_MODEL
+        self.dimension = _coerce_dimension(
+            dimension if dimension is not None else os.getenv("ORCAROUTER_EMBEDDING_DIMENSION"),
+            "ORCAROUTER_EMBEDDING_DIMENSION",
+            DEFAULT_DIMENSION,
+        )
+        # Same gemini-style default as the OpenRouter embedder — matches the
+        # multilingual google/gemini-embedding-001 default model.
+        self.prompt_style = _resolve_prompt_style(prompt_style, "gemini")
+
+        logger.info(
+            f"OrcaRouter Embedder initialized with model: {self.model} "
+            f"(dimension={self.dimension}, prompt_style={self.prompt_style})"
+        )
+
+
 class LocalEmbedder(_BaseOpenAICompatibleEmbedder):
     """
     Embedder for a local OpenAI-compatible embedding server — Ollama,
@@ -332,17 +394,22 @@ def get_embedder():
     Factory that picks the embedder based on EMBEDDING_PROVIDER.
 
     - ``EMBEDDING_PROVIDER=local`` -> ``LocalEmbedder``
+    - ``ORCAROUTER_API_KEY`` set -> ``OrcaRouterEmbedder``
     - otherwise -> ``OpenRouterEmbedder`` (requires OPENROUTER_API_KEY)
 
     Raises:
-        ValueError: If no provider is configured (neither local nor OpenRouter).
+        ValueError: If no provider is configured (neither local, OpenRouter,
+            nor OrcaRouter).
     """
     if is_local_embedding_configured():
         return LocalEmbedder()
+    if is_orcarouter_available():
+        return OrcaRouterEmbedder()
     if is_openrouter_available():
         return OpenRouterEmbedder()
     raise ValueError(
-        "No embedding provider configured. Set OPENROUTER_API_KEY for hosted "
-        "embeddings, or EMBEDDING_PROVIDER=local (with LOCAL_EMBEDDING_* "
-        "env vars) for a local OpenAI-compatible server like Ollama."
+        "No embedding provider configured. Set OPENROUTER_API_KEY or "
+        "ORCAROUTER_API_KEY for hosted embeddings, or EMBEDDING_PROVIDER=local "
+        "(with LOCAL_EMBEDDING_* env vars) for a local OpenAI-compatible "
+        "server like Ollama."
     )