fix(genai): ollama loose JSON (#39)

Ollama 0.11.8 segfaults on any Pydantic-shaped structured-output schema
with $ref, anyOf, or pattern — confirmed on the deploy host with the
simplest MVP case (BankStatementHeader alone). The earlier null-stripping
sanitiser wasn't enough.

Switch to format="json", which is "emit valid JSON" mode. We're already
describing the exact JSON shape in the system prompt (via GenAIStep +
the use case's citation instruction appendix) and validating the
response body through Pydantic on parse — which raises IX_002_001 on
schema mismatch, exactly as before.

Stronger guarantees can come back later via a newer Ollama, an API
fix, or a different GenAIClient impl. None of that is needed for the
MVP to work end to end.

Unit tests: the sanitiser left in place (harmless, still tested). The
"happy path" test now asserts format == "json".

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

src/ix/genai/ollama_client.py

@@ -159,7 +159,21 @@ class OllamaClient:
         request_kwargs: dict[str, Any],
         response_schema: type[BaseModel],
     ) -> dict[str, Any]:
-        """Map provider-neutral kwargs to Ollama's /api/chat body."""
+        """Map provider-neutral kwargs to Ollama's /api/chat body.
+
+        Schema strategy for Ollama 0.11.8: we pass ``format="json"`` (loose
+        JSON mode) rather than the full Pydantic schema. The llama.cpp
+        structured-output implementation in 0.11.8 segfaults on schemas
+        involving ``anyOf``, ``$ref``, or ``pattern`` — which Pydantic v2
+        emits for Optional / nested-model / Decimal fields.
+
+        In loose JSON mode Ollama guarantees only syntactically-valid
+        JSON; we enforce the schema on our side by catching the Pydantic
+        ``ValidationError`` at parse time and raising IX_002_001. The
+        system prompt (built upstream in GenAIStep) already tells the
+        model what JSON shape to emit, so loose mode is the right
+        abstraction layer here.
+        """
 
         messages = self._translate_messages(
             list(request_kwargs.get("messages") or [])
@@ -168,9 +182,7 @@ class OllamaClient:
             "model": request_kwargs.get("model"),
             "messages": messages,
             "stream": False,
-            "format": _sanitise_schema_for_ollama(
-                response_schema.model_json_schema()
-            ),
+            "format": "json",
         }
 
         options: dict[str, Any] = {}

tests/unit/test_ollama_client.py

@@ -79,13 +79,10 @@ class TestInvokeHappyPath:
         body_json = json.loads(body)
         assert body_json["model"] == "gpt-oss:20b"
         assert body_json["stream"] is False
-        # Format is the pydantic schema with Optional `anyOf [T, null]`
-        # patterns collapsed to just T — Ollama 0.11.8 segfaults on the
-        # anyOf+null shape, so we sanitise before sending.
-        fmt = body_json["format"]
-        assert fmt["properties"]["bank_name"] == {"title": "Bank Name", "type": "string"}
-        assert fmt["properties"]["account_number"]["type"] == "string"
-        assert "anyOf" not in fmt["properties"]["account_number"]
+        # format is "json" (loose mode): Ollama 0.11.8 segfaults on full
+        # Pydantic schemas. We pass the schema via the system prompt
+        # upstream and validate on parse.
+        assert body_json["format"] == "json"
         assert body_json["options"]["temperature"] == 0.2
         assert "reasoning_effort" not in body_json
         assert body_json["messages"] == [