prompture 0.0.47.dev2__py3-none-any.whl → 0.0.47.dev3__py3-none-any.whl

prompture/_version.py

@@ -28,7 +28,7 @@ version_tuple: VERSION_TUPLE
 commit_id: COMMIT_ID
 __commit_id__: COMMIT_ID
 
-__version__ = version = '0.0.47.dev2'
-__version_tuple__ = version_tuple = (0, 0, 47, 'dev2')
+__version__ = version = '0.0.47.dev3'
+__version_tuple__ = version_tuple = (0, 0, 47, 'dev3')
 
 __commit_id__ = commit_id = None

prompture/async_conversation.py

@@ -109,6 +109,9 @@ class AsyncConversation:
         self._max_tool_rounds = max_tool_rounds
         self._simulated_tools = simulated_tools
 
+        # Reasoning content from last response
+        self._last_reasoning: str | None = None
+
         # Persistence
         self._conversation_id = conversation_id or str(uuid.uuid4())
         self._auto_save = Path(auto_save) if auto_save else None
@@ -121,6 +124,11 @@ class AsyncConversation:
     # Public helpers
     # ------------------------------------------------------------------
 
+    @property
+    def last_reasoning(self) -> str | None:
+        """The reasoning/thinking content from the last LLM response, if any."""
+        return self._last_reasoning
+
     @property
     def messages(self) -> list[dict[str, Any]]:
         """Read-only view of the conversation history."""
@@ -326,6 +334,8 @@ class AsyncConversation:
         If tools are registered and the driver supports tool use,
         dispatches to the async tool execution loop.
         """
+        self._last_reasoning = None
+
         # Route to appropriate tool handling
         if self._tools:
             use_native = getattr(self._driver, "supports_tool_use", False)
@@ -340,6 +350,7 @@ class AsyncConversation:
 
         text = resp.get("text", "")
         meta = resp.get("meta", {})
+        self._last_reasoning = resp.get("reasoning_content")
 
         user_content = self._build_content_with_images(content, images)
         self._messages.append({"role": "user", "content": user_content})
@@ -372,6 +383,7 @@ class AsyncConversation:
             text = resp.get("text", "")
 
             if not tool_calls:
+                self._last_reasoning = resp.get("reasoning_content")
                 self._messages.append({"role": "assistant", "content": text})
                 return text
 
@@ -526,6 +538,8 @@ class AsyncConversation:
         images: list[ImageInput] | None = None,
     ) -> dict[str, Any]:
         """Send a message with schema enforcement and get structured JSON back (async)."""
+        self._last_reasoning = None
+
         merged = {**self._options, **(options or {})}
 
         schema_string = json.dumps(json_schema, indent=2)
@@ -563,6 +577,7 @@ class AsyncConversation:
 
         text = resp.get("text", "")
         meta = resp.get("meta", {})
+        self._last_reasoning = resp.get("reasoning_content")
 
         user_content = self._build_content_with_images(content, images)
         self._messages.append({"role": "user", "content": user_content})
@@ -597,6 +612,7 @@ class AsyncConversation:
             "json_object": json_obj,
             "usage": usage,
             "output_format": output_format,
+            "reasoning": self._last_reasoning,
         }
 
         if output_format == "toon":

prompture/conversation.py

@@ -112,6 +112,9 @@ class Conversation:
         self._max_tool_rounds = max_tool_rounds
         self._simulated_tools = simulated_tools
 
+        # Reasoning content from last response
+        self._last_reasoning: str | None = None
+
         # Persistence
         self._conversation_id = conversation_id or str(uuid.uuid4())
         self._auto_save = Path(auto_save) if auto_save else None
@@ -124,6 +127,11 @@ class Conversation:
     # Public helpers
     # ------------------------------------------------------------------
 
+    @property
+    def last_reasoning(self) -> str | None:
+        """The reasoning/thinking content from the last LLM response, if any."""
+        return self._last_reasoning
+
     @property
     def messages(self) -> list[dict[str, Any]]:
         """Read-only view of the conversation history."""
@@ -340,6 +348,8 @@ class Conversation:
             images: Optional list of images to include (bytes, path, URL,
                 base64 string, or :class:`ImageContent`).
         """
+        self._last_reasoning = None
+
         # Route to appropriate tool handling
         if self._tools:
             use_native = getattr(self._driver, "supports_tool_use", False)
@@ -354,6 +364,7 @@ class Conversation:
 
         text = resp.get("text", "")
         meta = resp.get("meta", {})
+        self._last_reasoning = resp.get("reasoning_content")
 
         # Record in history — store content with images for context
         user_content = self._build_content_with_images(content, images)
@@ -389,6 +400,7 @@ class Conversation:
 
             if not tool_calls:
                 # No tool calls -> final response
+                self._last_reasoning = resp.get("reasoning_content")
                 self._messages.append({"role": "assistant", "content": text})
                 return text
 
@@ -553,6 +565,8 @@ class Conversation:
         context clean for subsequent turns.
         """
 
+        self._last_reasoning = None
+
         merged = {**self._options, **(options or {})}
 
         # Build the full prompt with schema instructions inline (handled by ask_for_json)
@@ -594,6 +608,7 @@ class Conversation:
 
         text = resp.get("text", "")
         meta = resp.get("meta", {})
+        self._last_reasoning = resp.get("reasoning_content")
 
         # Store original content (without schema boilerplate) for cleaner context
         # Include images in history so subsequent turns can reference them
@@ -632,6 +647,7 @@ class Conversation:
             "json_object": json_obj,
             "usage": usage,
             "output_format": output_format,
+            "reasoning": self._last_reasoning,
         }
 
         if output_format == "toon":

prompture/drivers/async_grok_driver.py

@@ -95,8 +95,17 @@ class AsyncGrokDriver(CostMixin, AsyncDriver):
             "model_name": model,
         }
 
-        text = resp["choices"][0]["message"]["content"]
-        return {"text": text, "meta": meta}
+        message = resp["choices"][0]["message"]
+        text = message.get("content") or ""
+        reasoning_content = message.get("reasoning_content")
+
+        if not text and reasoning_content:
+            text = reasoning_content
+
+        result: dict[str, Any] = {"text": text, "meta": meta}
+        if reasoning_content is not None:
+            result["reasoning_content"] = reasoning_content
+        return result
 
     # ------------------------------------------------------------------
     # Tool use
@@ -173,15 +182,20 @@ class AsyncGrokDriver(CostMixin, AsyncDriver):
                 args = json.loads(tc["function"]["arguments"])
             except (json.JSONDecodeError, TypeError):
                 args = {}
-            tool_calls_out.append({
-                "id": tc["id"],
-                "name": tc["function"]["name"],
-                "arguments": args,
-            })
-
-        return {
+            tool_calls_out.append(
+                {
+                    "id": tc["id"],
+                    "name": tc["function"]["name"],
+                    "arguments": args,
+                }
+            )
+
+        result: dict[str, Any] = {
             "text": text,
             "meta": meta,
             "tool_calls": tool_calls_out,
             "stop_reason": stop_reason,
         }
+        if choice["message"].get("reasoning_content") is not None:
+            result["reasoning_content"] = choice["message"]["reasoning_content"]
+        return result

prompture/drivers/async_groq_driver.py

@@ -88,8 +88,16 @@ class AsyncGroqDriver(CostMixin, AsyncDriver):
             "model_name": model,
         }
 
-        text = resp.choices[0].message.content
-        return {"text": text, "meta": meta}
+        text = resp.choices[0].message.content or ""
+        reasoning_content = getattr(resp.choices[0].message, "reasoning_content", None)
+
+        if not text and reasoning_content:
+            text = reasoning_content
+
+        result: dict[str, Any] = {"text": text, "meta": meta}
+        if reasoning_content is not None:
+            result["reasoning_content"] = reasoning_content
+        return result
 
     # ------------------------------------------------------------------
     # Tool use
@@ -152,15 +160,21 @@ class AsyncGroqDriver(CostMixin, AsyncDriver):
                     args = json.loads(tc.function.arguments)
                 except (json.JSONDecodeError, TypeError):
                     args = {}
-                tool_calls_out.append({
-                    "id": tc.id,
-                    "name": tc.function.name,
-                    "arguments": args,
-                })
-
-        return {
+                tool_calls_out.append(
+                    {
+                        "id": tc.id,
+                        "name": tc.function.name,
+                        "arguments": args,
+                    }
+                )
+
+        result: dict[str, Any] = {
             "text": text,
             "meta": meta,
             "tool_calls": tool_calls_out,
             "stop_reason": stop_reason,
         }
+        reasoning_content = getattr(choice.message, "reasoning_content", None)
+        if reasoning_content is not None:
+            result["reasoning_content"] = reasoning_content
+        return result

prompture/drivers/async_lmstudio_driver.py

@@ -98,7 +98,12 @@ class AsyncLMStudioDriver(AsyncDriver):
         if "choices" not in response_data or not response_data["choices"]:
             raise ValueError(f"Unexpected response format: {response_data}")
 
-        text = response_data["choices"][0]["message"]["content"]
+        message = response_data["choices"][0]["message"]
+        text = message.get("content") or ""
+        reasoning_content = message.get("reasoning_content")
+
+        if not text and reasoning_content:
+            text = reasoning_content
 
         usage = response_data.get("usage", {})
         prompt_tokens = usage.get("prompt_tokens", 0)
@@ -114,7 +119,10 @@ class AsyncLMStudioDriver(AsyncDriver):
             "model_name": merged_options.get("model", self.model),
         }
 
-        return {"text": text, "meta": meta}
+        result: dict[str, Any] = {"text": text, "meta": meta}
+        if reasoning_content is not None:
+            result["reasoning_content"] = reasoning_content
+        return result
 
     # -- Model management (LM Studio 0.4.0+) ----------------------------------
 

prompture/drivers/async_moonshot_driver.py

@@ -138,10 +138,11 @@ class AsyncMoonshotDriver(CostMixin, AsyncDriver):
 
         message = resp["choices"][0]["message"]
         text = message.get("content") or ""
+        reasoning_content = message.get("reasoning_content")
 
         # Reasoning models may return content in reasoning_content when content is empty
-        if not text and message.get("reasoning_content"):
-            text = message["reasoning_content"]
+        if not text and reasoning_content:
+            text = reasoning_content
 
         # Structured output fallback: if we used json_schema mode and got an
         # empty response, retry with json_object mode and schema in the prompt.
@@ -184,8 +185,9 @@ class AsyncMoonshotDriver(CostMixin, AsyncDriver):
                     resp = fb_resp
                     fb_message = fb_resp["choices"][0]["message"]
                     text = fb_message.get("content") or ""
-                    if not text and fb_message.get("reasoning_content"):
-                        text = fb_message["reasoning_content"]
+                    reasoning_content = fb_message.get("reasoning_content")
+                    if not text and reasoning_content:
+                        text = reasoning_content
 
         total_cost = self._calculate_cost("moonshot", model, prompt_tokens, completion_tokens)
 
@@ -198,7 +200,10 @@ class AsyncMoonshotDriver(CostMixin, AsyncDriver):
             "model_name": model,
         }
 
-        return {"text": text, "meta": meta}
+        result: dict[str, Any] = {"text": text, "meta": meta}
+        if reasoning_content is not None:
+            result["reasoning_content"] = reasoning_content
+        return result
 
     # ------------------------------------------------------------------
     # Tool use
@@ -334,6 +339,7 @@ class AsyncMoonshotDriver(CostMixin, AsyncDriver):
             data["temperature"] = opts["temperature"]
 
         full_text = ""
+        full_reasoning = ""
         prompt_tokens = 0
         completion_tokens = 0
 
@@ -368,9 +374,11 @@ class AsyncMoonshotDriver(CostMixin, AsyncDriver):
                 if choices:
                     delta = choices[0].get("delta", {})
                     content = delta.get("content") or ""
-                    # Reasoning models stream thinking via reasoning_content
-                    if not content:
-                        content = delta.get("reasoning_content") or ""
+                    reasoning_chunk = delta.get("reasoning_content") or ""
+                    if reasoning_chunk:
+                        full_reasoning += reasoning_chunk
+                    if not content and reasoning_chunk:
+                        content = reasoning_chunk
                     if content:
                         full_text += content
                         yield {"type": "delta", "text": content}
@@ -378,7 +386,7 @@ class AsyncMoonshotDriver(CostMixin, AsyncDriver):
         total_tokens = prompt_tokens + completion_tokens
         total_cost = self._calculate_cost("moonshot", model, prompt_tokens, completion_tokens)
 
-        yield {
+        done_chunk: dict[str, Any] = {
             "type": "done",
             "text": full_text,
             "meta": {
@@ -390,3 +398,6 @@ class AsyncMoonshotDriver(CostMixin, AsyncDriver):
                 "model_name": model,
             },
         }
+        if full_reasoning:
+            done_chunk["reasoning_content"] = full_reasoning
+        yield done_chunk

prompture/drivers/async_openrouter_driver.py

@@ -122,8 +122,17 @@ class AsyncOpenRouterDriver(CostMixin, AsyncDriver):
             "model_name": model,
         }
 
-        text = resp["choices"][0]["message"]["content"]
-        return {"text": text, "meta": meta}
+        message = resp["choices"][0]["message"]
+        text = message.get("content") or ""
+        reasoning_content = message.get("reasoning_content")
+
+        if not text and reasoning_content:
+            text = reasoning_content
+
+        result: dict[str, Any] = {"text": text, "meta": meta}
+        if reasoning_content is not None:
+            result["reasoning_content"] = reasoning_content
+        return result
 
     # ------------------------------------------------------------------
     # Tool use
@@ -196,18 +205,23 @@ class AsyncOpenRouterDriver(CostMixin, AsyncDriver):
                 args = json.loads(tc["function"]["arguments"])
             except (json.JSONDecodeError, TypeError):
                 args = {}
-            tool_calls_out.append({
-                "id": tc["id"],
-                "name": tc["function"]["name"],
-                "arguments": args,
-            })
+            tool_calls_out.append(
+                {
+                    "id": tc["id"],
+                    "name": tc["function"]["name"],
+                    "arguments": args,
+                }
+            )
 
-        return {
+        result: dict[str, Any] = {
             "text": text,
             "meta": meta,
             "tool_calls": tool_calls_out,
             "stop_reason": stop_reason,
         }
+        if choice["message"].get("reasoning_content") is not None:
+            result["reasoning_content"] = choice["message"]["reasoning_content"]
+        return result
 
     # ------------------------------------------------------------------
     # Streaming
@@ -238,21 +252,25 @@ class AsyncOpenRouterDriver(CostMixin, AsyncDriver):
             data["temperature"] = opts["temperature"]
 
         full_text = ""
+        full_reasoning = ""
         prompt_tokens = 0
         completion_tokens = 0
 
-        async with httpx.AsyncClient() as client, client.stream(
-            "POST",
-            f"{self.base_url}/chat/completions",
-            headers=self.headers,
-            json=data,
-            timeout=120,
-        ) as response:
+        async with (
+            httpx.AsyncClient() as client,
+            client.stream(
+                "POST",
+                f"{self.base_url}/chat/completions",
+                headers=self.headers,
+                json=data,
+                timeout=120,
+            ) as response,
+        ):
             response.raise_for_status()
             async for line in response.aiter_lines():
                 if not line or not line.startswith("data: "):
                     continue
-                payload = line[len("data: "):]
+                payload = line[len("data: ") :]
                 if payload.strip() == "[DONE]":
                     break
                 try:
@@ -270,6 +288,11 @@ class AsyncOpenRouterDriver(CostMixin, AsyncDriver):
                 if choices:
                     delta = choices[0].get("delta", {})
                     content = delta.get("content", "")
+                    reasoning_chunk = delta.get("reasoning_content") or ""
+                    if reasoning_chunk:
+                        full_reasoning += reasoning_chunk
+                    if not content and reasoning_chunk:
+                        content = reasoning_chunk
                     if content:
                         full_text += content
                         yield {"type": "delta", "text": content}
@@ -277,7 +300,7 @@ class AsyncOpenRouterDriver(CostMixin, AsyncDriver):
         total_tokens = prompt_tokens + completion_tokens
         total_cost = self._calculate_cost("openrouter", model, prompt_tokens, completion_tokens)
 
-        yield {
+        done_chunk: dict[str, Any] = {
             "type": "done",
             "text": full_text,
             "meta": {
@@ -289,3 +312,6 @@ class AsyncOpenRouterDriver(CostMixin, AsyncDriver):
                 "model_name": model,
             },
         }
+        if full_reasoning:
+            done_chunk["reasoning_content"] = full_reasoning
+        yield done_chunk

prompture/drivers/grok_driver.py

@@ -154,8 +154,17 @@ class GrokDriver(CostMixin, Driver):
             "model_name": model,
         }
 
-        text = resp["choices"][0]["message"]["content"]
-        return {"text": text, "meta": meta}
+        message = resp["choices"][0]["message"]
+        text = message.get("content") or ""
+        reasoning_content = message.get("reasoning_content")
+
+        if not text and reasoning_content:
+            text = reasoning_content
+
+        result: dict[str, Any] = {"text": text, "meta": meta}
+        if reasoning_content is not None:
+            result["reasoning_content"] = reasoning_content
+        return result
 
     # ------------------------------------------------------------------
     # Tool use
@@ -227,15 +236,20 @@ class GrokDriver(CostMixin, Driver):
                 args = json.loads(tc["function"]["arguments"])
             except (json.JSONDecodeError, TypeError):
                 args = {}
-            tool_calls_out.append({
-                "id": tc["id"],
-                "name": tc["function"]["name"],
-                "arguments": args,
-            })
-
-        return {
+            tool_calls_out.append(
+                {
+                    "id": tc["id"],
+                    "name": tc["function"]["name"],
+                    "arguments": args,
+                }
+            )
+
+        result: dict[str, Any] = {
             "text": text,
             "meta": meta,
             "tool_calls": tool_calls_out,
             "stop_reason": stop_reason,
         }
+        if choice["message"].get("reasoning_content") is not None:
+            result["reasoning_content"] = choice["message"]["reasoning_content"]
+        return result

prompture/drivers/groq_driver.py

@@ -122,8 +122,16 @@ class GroqDriver(CostMixin, Driver):
         }
 
         # Extract generated text
-        text = resp.choices[0].message.content
-        return {"text": text, "meta": meta}
+        text = resp.choices[0].message.content or ""
+        reasoning_content = getattr(resp.choices[0].message, "reasoning_content", None)
+
+        if not text and reasoning_content:
+            text = reasoning_content
+
+        result: dict[str, Any] = {"text": text, "meta": meta}
+        if reasoning_content is not None:
+            result["reasoning_content"] = reasoning_content
+        return result
 
     # ------------------------------------------------------------------
     # Tool use
@@ -186,15 +194,21 @@ class GroqDriver(CostMixin, Driver):
                     args = json.loads(tc.function.arguments)
                 except (json.JSONDecodeError, TypeError):
                     args = {}
-                tool_calls_out.append({
-                    "id": tc.id,
-                    "name": tc.function.name,
-                    "arguments": args,
-                })
-
-        return {
+                tool_calls_out.append(
+                    {
+                        "id": tc.id,
+                        "name": tc.function.name,
+                        "arguments": args,
+                    }
+                )
+
+        result: dict[str, Any] = {
             "text": text,
             "meta": meta,
             "tool_calls": tool_calls_out,
             "stop_reason": stop_reason,
         }
+        reasoning_content = getattr(choice.message, "reasoning_content", None)
+        if reasoning_content is not None:
+            result["reasoning_content"] = reasoning_content
+        return result

prompture/drivers/lmstudio_driver.py

@@ -123,7 +123,13 @@ class LMStudioDriver(Driver):
             raise RuntimeError(f"LM Studio request failed: {e}") from e
 
         # Extract text
-        text = response_data["choices"][0]["message"]["content"]
+        message = response_data["choices"][0]["message"]
+        text = message.get("content") or ""
+        reasoning_content = message.get("reasoning_content")
+
+        # Reasoning models (e.g. DeepSeek R1) may return content in reasoning_content
+        if not text and reasoning_content:
+            text = reasoning_content
 
         # Meta info
         usage = response_data.get("usage", {})
@@ -140,7 +146,10 @@ class LMStudioDriver(Driver):
             "model_name": merged_options.get("model", self.model),
         }
 
-        return {"text": text, "meta": meta}
+        result: dict[str, Any] = {"text": text, "meta": meta}
+        if reasoning_content is not None:
+            result["reasoning_content"] = reasoning_content
+        return result
 
     # -- Model management (LM Studio 0.4.0+) ----------------------------------
 

prompture/drivers/moonshot_driver.py

@@ -228,10 +228,11 @@ class MoonshotDriver(CostMixin, Driver):
 
         message = resp["choices"][0]["message"]
         text = message.get("content") or ""
+        reasoning_content = message.get("reasoning_content")
 
         # Reasoning models may return content in reasoning_content when content is empty
-        if not text and message.get("reasoning_content"):
-            text = message["reasoning_content"]
+        if not text and reasoning_content:
+            text = reasoning_content
 
         # Structured output fallback: if we used json_schema mode and got an
         # empty response, retry with json_object mode and schema in the prompt.
@@ -275,8 +276,9 @@ class MoonshotDriver(CostMixin, Driver):
                 resp = fb_resp
                 fb_message = fb_resp["choices"][0]["message"]
                 text = fb_message.get("content") or ""
-                if not text and fb_message.get("reasoning_content"):
-                    text = fb_message["reasoning_content"]
+                reasoning_content = fb_message.get("reasoning_content")
+                if not text and reasoning_content:
+                    text = reasoning_content
 
         total_cost = self._calculate_cost("moonshot", model, prompt_tokens, completion_tokens)
 
@@ -289,7 +291,10 @@ class MoonshotDriver(CostMixin, Driver):
             "model_name": model,
         }
 
-        return {"text": text, "meta": meta}
+        result: dict[str, Any] = {"text": text, "meta": meta}
+        if reasoning_content is not None:
+            result["reasoning_content"] = reasoning_content
+        return result
 
     # ------------------------------------------------------------------
     # Tool use
@@ -439,6 +444,7 @@ class MoonshotDriver(CostMixin, Driver):
         response.raise_for_status()
 
         full_text = ""
+        full_reasoning = ""
         prompt_tokens = 0
         completion_tokens = 0
 
@@ -462,9 +468,11 @@ class MoonshotDriver(CostMixin, Driver):
             if choices:
                 delta = choices[0].get("delta", {})
                 content = delta.get("content") or ""
-                # Reasoning models stream thinking via reasoning_content
-                if not content:
-                    content = delta.get("reasoning_content") or ""
+                reasoning_chunk = delta.get("reasoning_content") or ""
+                if reasoning_chunk:
+                    full_reasoning += reasoning_chunk
+                if not content and reasoning_chunk:
+                    content = reasoning_chunk
                 if content:
                     full_text += content
                     yield {"type": "delta", "text": content}
@@ -472,7 +480,7 @@ class MoonshotDriver(CostMixin, Driver):
         total_tokens = prompt_tokens + completion_tokens
         total_cost = self._calculate_cost("moonshot", model, prompt_tokens, completion_tokens)
 
-        yield {
+        done_chunk: dict[str, Any] = {
             "type": "done",
             "text": full_text,
             "meta": {
@@ -484,3 +492,6 @@ class MoonshotDriver(CostMixin, Driver):
                 "model_name": model,
             },
         }
+        if full_reasoning:
+            done_chunk["reasoning_content"] = full_reasoning
+        yield done_chunk

prompture/drivers/openrouter_driver.py

@@ -181,8 +181,18 @@ class OpenRouterDriver(CostMixin, Driver):
             "model_name": model,
         }
 
-        text = resp["choices"][0]["message"]["content"]
-        return {"text": text, "meta": meta}
+        message = resp["choices"][0]["message"]
+        text = message.get("content") or ""
+        reasoning_content = message.get("reasoning_content")
+
+        # Reasoning models may return content in reasoning_content when content is empty
+        if not text and reasoning_content:
+            text = reasoning_content
+
+        result: dict[str, Any] = {"text": text, "meta": meta}
+        if reasoning_content is not None:
+            result["reasoning_content"] = reasoning_content
+        return result
 
     # ------------------------------------------------------------------
     # Tool use
@@ -257,18 +267,23 @@ class OpenRouterDriver(CostMixin, Driver):
                 args = json.loads(tc["function"]["arguments"])
             except (json.JSONDecodeError, TypeError):
                 args = {}
-            tool_calls_out.append({
-                "id": tc["id"],
-                "name": tc["function"]["name"],
-                "arguments": args,
-            })
+            tool_calls_out.append(
+                {
+                    "id": tc["id"],
+                    "name": tc["function"]["name"],
+                    "arguments": args,
+                }
+            )
 
-        return {
+        result: dict[str, Any] = {
             "text": text,
             "meta": meta,
             "tool_calls": tool_calls_out,
             "stop_reason": stop_reason,
         }
+        if choice["message"].get("reasoning_content") is not None:
+            result["reasoning_content"] = choice["message"]["reasoning_content"]
+        return result
 
     # ------------------------------------------------------------------
     # Streaming
@@ -311,13 +326,14 @@ class OpenRouterDriver(CostMixin, Driver):
         response.raise_for_status()
 
         full_text = ""
+        full_reasoning = ""
         prompt_tokens = 0
         completion_tokens = 0
 
         for line in response.iter_lines(decode_unicode=True):
             if not line or not line.startswith("data: "):
                 continue
-            payload = line[len("data: "):]
+            payload = line[len("data: ") :]
             if payload.strip() == "[DONE]":
                 break
             try:
@@ -335,6 +351,11 @@ class OpenRouterDriver(CostMixin, Driver):
             if choices:
                 delta = choices[0].get("delta", {})
                 content = delta.get("content", "")
+                reasoning_chunk = delta.get("reasoning_content") or ""
+                if reasoning_chunk:
+                    full_reasoning += reasoning_chunk
+                if not content and reasoning_chunk:
+                    content = reasoning_chunk
                 if content:
                     full_text += content
                     yield {"type": "delta", "text": content}
@@ -342,7 +363,7 @@ class OpenRouterDriver(CostMixin, Driver):
         total_tokens = prompt_tokens + completion_tokens
         total_cost = self._calculate_cost("openrouter", model, prompt_tokens, completion_tokens)
 
-        yield {
+        done_chunk: dict[str, Any] = {
             "type": "done",
             "text": full_text,
             "meta": {
@@ -354,3 +375,6 @@ class OpenRouterDriver(CostMixin, Driver):
                 "model_name": model,
             },
         }
+        if full_reasoning:
+            done_chunk["reasoning_content"] = full_reasoning
+        yield done_chunk

{prompture-0.0.47.dev2.dist-info → prompture-0.0.47.dev3.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: prompture
-Version: 0.0.47.dev2
+Version: 0.0.47.dev3
 Summary: Ask LLMs to return structured JSON and run cross-model tests. API-first.
 Author-email: Juan Denis <juan@vene.co>
 License-Expression: MIT

{prompture-0.0.47.dev2.dist-info → prompture-0.0.47.dev3.dist-info}/RECORD

@@ -1,16 +1,16 @@
 prompture/__init__.py,sha256=cJnkefDpiyFbU77juw4tXPdKJQWoJ-c6XBFt2v-e5Q4,7455
-prompture/_version.py,sha256=O-euPmgX7xDW2XYNpQwEZBbxUjaoxG5WwalrpUa34CM,719
+prompture/_version.py,sha256=u1zVumMYiBTlJQmr6vfEgpcsD6-vEHBNzFbXS_Hdi_Q,719
 prompture/agent.py,sha256=-8qdo_Lz20GGssCe5B_QPxb5Kct71YtKHh5vZgrSYik,34748
 prompture/agent_types.py,sha256=Icl16PQI-ThGLMFCU43adtQA6cqETbsPn4KssKBI4xc,4664
 prompture/async_agent.py,sha256=_6_IRb-LGzZxGxfPVy43SIWByUoQfN-5XnUWahVP6r8,33110
-prompture/async_conversation.py,sha256=dqUqozCHSmquu-x9hDtvdzpKNyOrYx0osQFymHLKUKc,34227
+prompture/async_conversation.py,sha256=tzOy2jiCOVkRHIJ9H6VeRwwcUtbjF33--igdxMC6F7s,34811
 prompture/async_core.py,sha256=hbRXLvsBJv3JAnUwGZbazsL6x022FrsJU6swmZolgxY,29745
 prompture/async_driver.py,sha256=4VQ9Q_tI6Ufw6W1CYJ5j8hVtgVdqFGuk6e2tLaSceWE,8581
 prompture/async_groups.py,sha256=pceKrt0UayQjMLFs1dFGoxOHpgD948aEjIY61r608C4,22459
 prompture/cache.py,sha256=4dfQDMsEZ9JMQDXLOkiugPmmMJQIfKVE8rTAKDH4oL8,14401
 prompture/callbacks.py,sha256=JPDqWGzPIzv44l54ocmezlYVBnbKPDEEXRrLdluWGAo,1731
 prompture/cli.py,sha256=tNiIddRmgC1BomjY5O1VVVAwvqHVzF8IHmQrM-cG2wQ,2902
-prompture/conversation.py,sha256=7GU_VrC5Ht2EcziNFxTjFcVE6_D4McM0sUvGYEAXx-Y,35844
+prompture/conversation.py,sha256=uxstayJjgY6a39DtU0YxQl0Dt3JBo2UVCyMPJW95MNI,36428
 prompture/core.py,sha256=5FHwX7fNPwFHMbFCMvV-RH7LpPpTToLAmcyDnKbrN0E,57202
 prompture/cost_mixin.py,sha256=Qx7gPgPsWgTHiaFeI7q_p9cfe95ccjgN8Mi56d_AVX0,4563
 prompture/discovery.py,sha256=K-svbO-qJraHinCbFVS64vEo5McWX5pURv26ZMmuL6U,10295
@@ -40,31 +40,31 @@ prompture/drivers/async_airllm_driver.py,sha256=1hIWLXfyyIg9tXaOE22tLJvFyNwHnOi1
 prompture/drivers/async_azure_driver.py,sha256=s__y_EGQkK7UZjxiyF08uql8F09cnbJ0q7aFuxzreIw,7328
 prompture/drivers/async_claude_driver.py,sha256=oawbFVVMtRlikQOmu3jRjbdpoeu95JqTF1YHLKO3ybE,10576
 prompture/drivers/async_google_driver.py,sha256=LTUgCXJjzuTDGzsCsmY2-xH2KdTLJD7htwO49ZNFOdE,13711
-prompture/drivers/async_grok_driver.py,sha256=4oOGT4SzsheulU_QK0ZSqj4-THrFAOCeZwIqIslnW14,6858
-prompture/drivers/async_groq_driver.py,sha256=iORpf0wcqPfS4zKCg4BTWpQCoHV2klkQVTQ1W-jhjUE,5755
+prompture/drivers/async_grok_driver.py,sha256=lj160GHARe0fqTms4ovWhkpgt0idsGt55xnuc6JlH1w,7413
+prompture/drivers/async_groq_driver.py,sha256=5G0rXAEAmsLNftI9YfGAh4E8X3B4Hb6_0cXBhf9LZMk,6348
 prompture/drivers/async_hugging_driver.py,sha256=IblxqU6TpNUiigZ0BCgNkAgzpUr2FtPHJOZnOZMnHF0,2152
-prompture/drivers/async_lmstudio_driver.py,sha256=rPn2qVPm6UE2APzAn7ZHYTELUwr0dQMi8XHv6gAhyH8,5782
+prompture/drivers/async_lmstudio_driver.py,sha256=4bz8NFFiZiFFkzlYDcS7abnwmEbbvbKb-CQhHeTGlU8,6102
 prompture/drivers/async_local_http_driver.py,sha256=qoigIf-w3_c2dbVdM6m1e2RMAWP4Gk4VzVs5hM3lPvQ,1609
 prompture/drivers/async_modelscope_driver.py,sha256=wzHYGLf9qE9KXRFZYtN1hZS10Bw1m1Wy6HcmyUD67HM,10170
-prompture/drivers/async_moonshot_driver.py,sha256=ghSD0UDpbEuE2sCH3DKuRJkQ97p7PGM67r-bgLmVZvU,15189
+prompture/drivers/async_moonshot_driver.py,sha256=a9gr3T_4NiDFd7foM1mSHJRvXYb43iqqJnQ0FVRyI2E,15669
 prompture/drivers/async_ollama_driver.py,sha256=pFtCvh5bHe_qwGy-jIJbyG_zmnPbNbagJCGxCTJMdPU,8244
 prompture/drivers/async_openai_driver.py,sha256=COa_JE-AgKowKJpmRnfDJp4RSQKZel_7WswxOzvLksM,9044
-prompture/drivers/async_openrouter_driver.py,sha256=GnOMY67CCV3HV83lCC-CxcngwrUnuc7G-AX7fb1DYpg,10698
+prompture/drivers/async_openrouter_driver.py,sha256=N7s72HuXHLs_RWmJO9P3pCayWE98ommfqVeAfru8Bl0,11758
 prompture/drivers/async_registry.py,sha256=JFEnXNPm-8AAUCiNLoKuYBSCYEK-4BmAen5t55QrMvg,5223
 prompture/drivers/async_zai_driver.py,sha256=zXHxske1CtK8dDTGY-D_kiyZZ_NfceNTJlyTpKn0R4c,10727
 prompture/drivers/azure_driver.py,sha256=gQFffA29gOr-GZ25fNXTokV8-mEmffeV9CT_UBZ3yXc,8565
 prompture/drivers/claude_driver.py,sha256=C8Av3DXP2x3f35jEv8BRwEM_4vh0cfmLsy3t5dsR6aM,11837
 prompture/drivers/google_driver.py,sha256=Zck5VUsW37kDgohXz3cUWRmZ88OfhmTpVD-qzAVMp-8,16318
-prompture/drivers/grok_driver.py,sha256=mNfPgOsJR53_5Ep6aYnfKGy7lnZMqN8bxrqKep4CiF0,8408
-prompture/drivers/groq_driver.py,sha256=olr1t7V71ET8Z-7VyRwb75_iYEiZg8-n5qs1edZ2erw,6897
+prompture/drivers/grok_driver.py,sha256=fxl5Gx9acFq7BlOh_N9U66oJvG3y8YX4QuSAgZWHJmU,8963
+prompture/drivers/groq_driver.py,sha256=7YEok1BQlsDZGkA-l9yrjTDapqIWX3yq_Ctgbhu8jSI,7490
 prompture/drivers/hugging_driver.py,sha256=gZir3XnM77VfYIdnu3S1pRftlZJM6G3L8bgGn5esg-Q,2346
-prompture/drivers/lmstudio_driver.py,sha256=9ZnJ1l5LuWAjkH2WKfFjZprNMVIXoSC7qXDNDTxm-tA,6748
+prompture/drivers/lmstudio_driver.py,sha256=nZ5SvBC0kTDNDzsupIW_H7YK92dcYta_xSPUNs52gyM,7154
 prompture/drivers/local_http_driver.py,sha256=QJgEf9kAmy8YZ5fb8FHnWuhoDoZYNd8at4jegzNVJH0,1658
 prompture/drivers/modelscope_driver.py,sha256=yTxTG7j5f7zz4CjbrV8J0VKeoBmxv69F40bfp8nq6AE,10651
-prompture/drivers/moonshot_driver.py,sha256=0FJtOH0kjVW6awEW7ulLpjj6XzhPx_EDffqIi2bq-aw,18655
+prompture/drivers/moonshot_driver.py,sha256=cm1XpU6EPFjcZaneXjfetRNSUxN9daP6hkJ1y99kqLI,19123
 prompture/drivers/ollama_driver.py,sha256=SJtMRtAr8geUB4y5GIZxPr-RJ0C3q7yqigYei2b4luM,13710
 prompture/drivers/openai_driver.py,sha256=DqdMhxF8M2HdOY5vfsFrz0h23lqBoQlbxV3xUdHvZho,10548
-prompture/drivers/openrouter_driver.py,sha256=DaG1H99s8GaOgJXZK4TP28HM7U4wiLu9wHXzWZleW_U,12589
+prompture/drivers/openrouter_driver.py,sha256=m2I5E9L5YYE_bV8PruKnAwjL63SIFEXevN_ThUzxQaA,13657
 prompture/drivers/registry.py,sha256=Dg_5w9alnIPKhOnsR9Xspuf5T7roBGu0r_L2Cf-UhXs,9926
 prompture/drivers/vision_helpers.py,sha256=l5iYXHJLR_vLFvqDPPPK1QqK7YPKh5GwocpbSyt0R04,5403
 prompture/drivers/zai_driver.py,sha256=Wkur0HfwKJt8ugYErpvz1Gy6e9an8vt4R7U3i6HWV_s,11038
@@ -77,9 +77,9 @@ prompture/scaffold/templates/env.example.j2,sha256=eESKr1KWgyrczO6d-nwAhQwSpf_G-
 prompture/scaffold/templates/main.py.j2,sha256=TEgc5OvsZOEX0JthkSW1NI_yLwgoeVN_x97Ibg-vyWY,2632
 prompture/scaffold/templates/models.py.j2,sha256=JrZ99GCVK6TKWapskVRSwCssGrTu5cGZ_r46fOhY2GE,858
 prompture/scaffold/templates/requirements.txt.j2,sha256=m3S5fi1hq9KG9l_9j317rjwWww0a43WMKd8VnUWv2A4,102
-prompture-0.0.47.dev2.dist-info/licenses/LICENSE,sha256=0HgDepH7aaHNFhHF-iXuW6_GqDfYPnVkjtiCAZ4yS8I,1060
-prompture-0.0.47.dev2.dist-info/METADATA,sha256=sWcR-K1ML6e_u8s41LcfNHIln4BLrywR1R_Kv9u6g5w,12153
-prompture-0.0.47.dev2.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
-prompture-0.0.47.dev2.dist-info/entry_points.txt,sha256=AFPG3lJR86g4IJMoWQUW5Ph7G6MLNWG3A2u2Tp9zkp8,48
-prompture-0.0.47.dev2.dist-info/top_level.txt,sha256=to86zq_kjfdoLeAxQNr420UWqT0WzkKoZ509J7Qr2t4,10
-prompture-0.0.47.dev2.dist-info/RECORD,,
+prompture-0.0.47.dev3.dist-info/licenses/LICENSE,sha256=0HgDepH7aaHNFhHF-iXuW6_GqDfYPnVkjtiCAZ4yS8I,1060
+prompture-0.0.47.dev3.dist-info/METADATA,sha256=GCknitr30odleYVAoFwVD9zZvMaHqed8z_3omEhs3Vc,12153
+prompture-0.0.47.dev3.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
+prompture-0.0.47.dev3.dist-info/entry_points.txt,sha256=AFPG3lJR86g4IJMoWQUW5Ph7G6MLNWG3A2u2Tp9zkp8,48
+prompture-0.0.47.dev3.dist-info/top_level.txt,sha256=to86zq_kjfdoLeAxQNr420UWqT0WzkKoZ509J7Qr2t4,10
+prompture-0.0.47.dev3.dist-info/RECORD,,