Remove Ollama reranker, this was a bad idea

docs/benchmarks.md

@@ -15,11 +15,11 @@ The recall obtained is ~0.79 for matching in the top result, raising to ~0.91 fo
 | Embedding Model                       | Document in top 1 | Document in top 3 | Reranker               |
 |---------------------------------------|-------------------|-------------------|------------------------|
 | Ollama / `mxbai-embed-large`          | 0.79              | 0.91              | None                   |
-| Ollama / `mxbai-embed-large`          | 0.81              | 0.91              | `mxbai-rerank-base-v2` |
+| Ollama / `mxbai-embed-large`          | 0.90              | 0.95              | `mxbai-rerank-base-v2` |
-| Ollama / `nomic-embed-text`           | 0.74              | 0.88              | None                   |
+<!-- | Ollama / `nomic-embed-text`           | 0.74              | 0.88              | None                   |
 | OpenAI / `text-embeddings-3-small`    | 0.75              | 0.88              | None                   |
 | OpenAI / `text-embeddings-3-small`    | 0.75              | 0.88              | None                   |
-| OpenAI / `text-embeddings-3-small`    | 0.83              | 0.90              | Cohere / `rerank-v3.5` |
+| OpenAI / `text-embeddings-3-small`    | 0.83              | 0.90              | Cohere / `rerank-v3.5` | -->
 
 ## Question/Answer evaluation
 
@@ -29,5 +29,5 @@ Again using the same dataset, we use a QA agent to answer the question. In addit
 |------------------------------------|-----------------------------------|-----------|------------------------|
 | Ollama / `mxbai-embed-large`       | Ollama / `qwen3`                  | 0.85      | None                   |
 | Ollama / `mxbai-embed-large`       | Ollama / `qwen3`                  | 0.72      | `mxbai-rerank-base-v2` |
-| Ollama / `mxbai-embed-large`       | Anthropic / `Claude Sonnet 3.7`   | 0.79      | None                   |
+<!-- | Ollama / `mxbai-embed-large`       | Anthropic / `Claude Sonnet 3.7`   | 0.79      | None                   |
-| OpenAI / `text-embeddings-3-small` | OpenAI / `gpt-4-turbo`            | 0.62      | None                   |
+| OpenAI / `text-embeddings-3-small` | OpenAI / `gpt-4-turbo`            | 0.62      | None                   | -->

docs/configuration.md

@@ -109,25 +109,7 @@ See the [Pydantic AI documentation](https://ai.pydantic.dev/models/) for the com
 
 Reranking improves search quality by re-ordering the initial search results using specialized models. When enabled, the system retrieves more candidates (3x the requested limit) and then reranks them to return the most relevant results.
 
-Reranking is **disabled by default** for faster searches. You can enable it by configuring a reranking provider.
+Reranking is **disabled by default** (`RERANK_PROVIDER=""`) for faster searches. You can enable it by configuring one of the providers below.
-
-### Disabling Reranking
-
-To disable reranking completely for faster searches:
-
-```bash
-RERANK_PROVIDER=""
-```
-
-### Ollama (Default)
-
-Ollama reranking uses LLMs with structured output to rank documents by relevance:
-
-```bash
-RERANK_PROVIDER="ollama"
-RERANK_MODEL="qwen3:1.7b"  # or any model that supports structured output
-OLLAMA_BASE_URL="http://localhost:11434"
-```
 
 ### MixedBread AI
 

src/haiku/rag/config.py

@@ -20,7 +20,7 @@ class AppConfig(BaseModel):
     EMBEDDINGS_VECTOR_DIM: int = 1024
 
     RERANK_PROVIDER: str = ""
-    RERANK_MODEL: str = "qwen3"
+    RERANK_MODEL: str = ""
 
     QA_PROVIDER: str = "ollama"
     QA_MODEL: str = "qwen3"

src/haiku/rag/reranking/__init__.py

@@ -31,10 +31,4 @@ def get_reranker() -> RerankerBase | None:
         except ImportError:
             return None
 
-    if Config.RERANK_PROVIDER == "ollama":
-        from haiku.rag.reranking.ollama import OllamaReranker
-
-        _reranker = OllamaReranker()
-        return _reranker
-
     return None

src/haiku/rag/reranking/ollama.py

@@ -1,81 +0,0 @@
-from pydantic import BaseModel
-from pydantic_ai import Agent
-from pydantic_ai.models.openai import OpenAIModel
-from pydantic_ai.providers.ollama import OllamaProvider
-
-from haiku.rag.config import Config
-from haiku.rag.reranking.base import RerankerBase
-from haiku.rag.store.models.chunk import Chunk
-
-
-class RerankResult(BaseModel):
-    """Individual rerank result with index and relevance score."""
-
-    index: int
-    relevance_score: float
-
-
-class RerankResponse(BaseModel):
-    """Response from the reranking model containing ranked results."""
-
-    results: list[RerankResult]
-
-
-class OllamaReranker(RerankerBase):
-    def __init__(self, model: str = Config.RERANK_MODEL):
-        self._model = model
-
-        # Create the reranking prompt
-        system_prompt = """You are a document reranking assistant. Given a query and a list of document chunks, you must rank them by relevance to the query.
-
-Return your response as a JSON object with a "results" array. Each result should have:
-- "index": the original index of the document (integer)
-- "relevance_score": a score between 0.0 and 1.0 indicating relevance (float, where 1.0 is most relevant)
-
-Only return the top documents up to the requested limit, ordered by decreasing relevance score.
-/no_think
-"""
-
-        model_obj = OpenAIModel(
-            model_name=model,
-            provider=OllamaProvider(base_url=f"{Config.OLLAMA_BASE_URL}/v1"),
-        )
-
-        self._agent = Agent(
-            model=model_obj,
-            output_type=RerankResponse,
-            system_prompt=system_prompt,
-        )
-
-    async def rerank(
-        self, query: str, chunks: list[Chunk], top_n: int = 10
-    ) -> list[tuple[Chunk, float]]:
-        if not chunks:
-            return []
-
-        documents = []
-        for i, chunk in enumerate(chunks):
-            documents.append({"index": i, "content": chunk.content})
-
-        documents_text = ""
-        for doc in documents:
-            documents_text += f"Index {doc['index']}: {doc['content']}\n\n"
-
-        user_prompt = f"""Query: {query}
-
-Documents to rerank:
-{documents_text.strip()}
-
-Rank these documents by relevance to the query and return the top {top_n} results as JSON."""
-
-        try:
-            result = await self._agent.run(user_prompt)
-
-            return [
-                (chunks[result_item.index], result_item.relevance_score)
-                for result_item in result.output.results[:top_n]
-            ]
-
-        except Exception:
-            # Fallback: return chunks in original order with same score
-            return [(chunks[i], 1.0) for i in range(min(top_n, len(chunks)))]

tests/test_reranker.py

@@ -24,7 +24,7 @@ chunks = [
 @pytest.mark.asyncio
 async def test_reranker_base():
     reranker = RerankerBase()
-    assert reranker._model == "qwen3"
+    assert reranker._model == ""
 
     with pytest.raises(NotImplementedError):
         await reranker.rerank("query", [])
@@ -35,12 +35,16 @@ async def test_mxbai_reranker():
     try:
         from haiku.rag.reranking.mxbai import MxBAIReranker
 
+        Config.RERANK_MODEL = "mixedbread-ai/mxbai-rerank-base-v2"
         reranker = MxBAIReranker()
+        # reranker._model = "mixedbread-ai/mxbai-rerank-base-v2"
         reranked = await reranker.rerank(
             "Who wrote 'To Kill a Mockingbird'?", chunks, top_n=2
         )
         assert [chunk.document_id for chunk, score in reranked] == ["0", "2"]
         assert all(isinstance(score, float) for chunk, score in reranked)
+        Config.RERANK_MODEL = ""
+
     except ImportError:
         pytest.skip("MxBAI package not installed")
 
@@ -62,16 +66,3 @@ async def test_cohere_reranker():
 
     except ImportError:
         pytest.skip("Cohere package not installed")
-
-
-@pytest.mark.asyncio
-async def test_ollama_reranker():
-    from haiku.rag.reranking.ollama import OllamaReranker
-
-    reranker = OllamaReranker()
-    reranked = await reranker.rerank(
-        "Who wrote 'To Kill a Mockingbird'?", chunks, top_n=2
-    )
-
-    assert [chunk.document_id for chunk, score in reranked] == ["0", "2"]
-    assert all(isinstance(score, float) for chunk, score in reranked)