Remove support for structured output

CHANGELOG.md

@@ -11,6 +11,7 @@
 
 - **LLMJudge**: Custom evaluator now accepts `ModelConfig` instead of a model name string
 - **Docling upgrade**: docling-core ≥2.70.2 (schema 1.10.0), docling ≥2.81.0. Adds field data model support for structured form/KV content, wide table chunking fixes, and rich table cell hang fix
+- **pydantic-ai ≥1.70.0**: Bumped minimum version. Removed `structured_output_type` helper — all supported providers now handle native structured output, so agents pass result types directly
 
 ## [0.34.1] - 2026-03-16
 

haiku_rag_slim/haiku/rag/agents/qa/agent.py

@@ -13,7 +13,7 @@ from haiku.rag.config import Config
 from haiku.rag.config.models import AppConfig, ModelConfig
 from haiku.rag.store.models import SearchResult
 from haiku.rag.tools.search import create_search_toolset
-from haiku.rag.utils import get_model, structured_output_type
+from haiku.rag.utils import get_model
 
 
 @dataclass
@@ -66,7 +66,7 @@ class QuestionAnswerAgent:
         agent: Agent[_QARunDeps, RawSearchAnswer] = Agent(  # ty: ignore[invalid-assignment]
             model=model,
             deps_type=_QARunDeps,
-            output_type=structured_output_type(RawSearchAnswer, model),
+            output_type=RawSearchAnswer,
             instructions=system_prompt,
             toolsets=[search_toolset],
             retries=3,

haiku_rag_slim/haiku/rag/agents/research/graph.py

@@ -19,7 +19,7 @@ from haiku.rag.agents.research.prompts import (
 from haiku.rag.agents.research.state import ResearchDeps, ResearchState
 from haiku.rag.config import Config
 from haiku.rag.config.models import AppConfig
-from haiku.rag.utils import build_prompt, get_model, structured_output_type
+from haiku.rag.utils import build_prompt, get_model
 
 
 def format_context_for_prompt(context: ResearchContext) -> str:
@@ -68,7 +68,7 @@ async def _iterative_plan_logic(
     model = get_model(model_config, config)
     plan_agent: Agent[ResearchDependencies, IterativePlanResult] = Agent(  # type: ignore[assignment]
         model=model,
-        output_type=structured_output_type(IterativePlanResult, model),
+        output_type=IterativePlanResult,
         instructions=effective_prompt,
         retries=3,
         deps_type=ResearchDependencies,
@@ -118,7 +118,7 @@ async def _search_one_step_logic(
         model = get_model(model_config, config)
         agent: Agent[ResearchDependencies, RawSearchAnswer] = Agent(  # type: ignore[assignment]
             model=model,
-            output_type=structured_output_type(RawSearchAnswer, model),
+            output_type=RawSearchAnswer,
             instructions=search_prompt,
             retries=3,
             deps_type=ResearchDependencies,
@@ -220,7 +220,7 @@ def build_research_graph(
         model = get_model(model_config, config)
         agent: Agent[ResearchDependencies, ResearchReport] = Agent(  # type: ignore[assignment]
             model=model,
-            output_type=structured_output_type(ResearchReport, model),
+            output_type=ResearchReport,
             instructions=synthesis_prompt,
             retries=3,
             deps_type=ResearchDependencies,

haiku_rag_slim/haiku/rag/agents/rlm/agent.py

@@ -4,7 +4,7 @@ from haiku.rag.agents.rlm.dependencies import RLMDeps
 from haiku.rag.agents.rlm.models import CodeExecution, RLMResult
 from haiku.rag.agents.rlm.prompts import RLM_SYSTEM_PROMPT
 from haiku.rag.config.models import AppConfig
-from haiku.rag.utils import get_model, structured_output_type
+from haiku.rag.utils import get_model
 
 
 def create_rlm_agent(config: AppConfig) -> Agent[RLMDeps, RLMResult]:
@@ -25,7 +25,7 @@ def create_rlm_agent(config: AppConfig) -> Agent[RLMDeps, RLMResult]:
     agent: Agent[RLMDeps, RLMResult] = Agent(  # type: ignore[invalid-assignment]
         model,
         deps_type=RLMDeps,
-        output_type=structured_output_type(RLMResult, model),
+        output_type=RLMResult,
         instructions=RLM_SYSTEM_PROMPT,
         retries=3,
     )

haiku_rag_slim/haiku/rag/utils.py

@@ -318,20 +318,6 @@ def get_model(
         return f"{provider}:{model}"
 
 
-def structured_output_type(
-    result_type: type,
-    model: Any,
-    max_retries: int = 3,
-) -> Any:
-    """Return a NativeOutput or ToolOutput wrapper based on model capability."""
-    from pydantic_ai.models import Model
-    from pydantic_ai.output import NativeOutput, ToolOutput
-
-    if isinstance(model, Model) and model.profile.supports_json_schema_output:
-        return NativeOutput(result_type)
-    return ToolOutput(result_type, max_retries=max_retries)
-
-
 def format_bytes(num_bytes: int) -> str:
     """Format bytes as human-readable string."""
     size = float(num_bytes)

tests/agents/rlm/test_agent.py

@@ -15,24 +15,11 @@ def vcr_cassette_dir():
 
 
 class TestCreateRLMAgent:
-    def test_creates_agent_native_output_when_supported(self):
+    def test_creates_agent(self):
-        from pydantic_ai.output import NativeOutput
-
         agent = create_rlm_agent(Config)
         assert isinstance(agent, Agent)
         assert agent.deps_type is RLMDeps
-        assert isinstance(agent.output_type, NativeOutput)
+        assert agent.output_type is RLMResult
-        assert agent.output_type.outputs is RLMResult
-
-    def test_creates_agent_native_output_for_ollama(self):
-        from pydantic_ai.output import NativeOutput
-
-        config = AppConfig()
-        config.rlm.model.name = "qwen3"
-        agent = create_rlm_agent(config)
-        assert isinstance(agent, Agent)
-        assert isinstance(agent.output_type, NativeOutput)
-        assert agent.output_type.outputs is RLMResult
 
     def test_agent_has_execute_code_tool(self):
         agent = create_rlm_agent(Config)

tests/test_utils.py

@@ -139,38 +139,6 @@ Emoji test: 🚀 ✅ 📝"""
     assert "🚀" in result_markdown
 
 
-def test_structured_output_type_native():
-    from pydantic_ai.output import NativeOutput
-
-    from haiku.rag.utils import structured_output_type
-
-    model = get_model(ModelConfig(provider="openai", name="gpt-4o"))
-    result = structured_output_type(str, model)
-    assert isinstance(result, NativeOutput)
-    assert result.outputs is str
-
-
-def test_structured_output_type_ollama_native():
-    from pydantic_ai.output import NativeOutput
-
-    from haiku.rag.utils import structured_output_type
-
-    model = get_model(ModelConfig(provider="ollama", name="qwen3"))
-    result = structured_output_type(str, model)
-    assert isinstance(result, NativeOutput)
-    assert result.outputs is str
-
-
-def test_structured_output_type_string_model():
-    from pydantic_ai.output import ToolOutput
-
-    from haiku.rag.utils import structured_output_type
-
-    result = structured_output_type(str, "unknown:model")
-    assert isinstance(result, ToolOutput)
-    assert result.output is str
-
-
 def test_get_model_ollama():
     """Test get_model returns OpenAIChatModel for Ollama."""
     model_config = ModelConfig(provider="ollama", name="llama3")