ccproxy-api 0.1.6__py3-none-any.whl → 0.2.0__py3-none-any.whl

ccproxy/llms/formatters/openai_to_anthropic/streams.py

@@ -0,0 +1,530 @@
+"""OpenAI→Anthropic streaming conversion entry points."""
+
+from __future__ import annotations
+
+import json
+from collections.abc import AsyncGenerator, AsyncIterator
+from typing import Any, cast
+
+import ccproxy.core.logging
+from ccproxy.llms.formatters.common import (
+    IndexedToolCallTracker,
+    ToolCallTracker,
+    emit_anthropic_tool_use_events,
+)
+from ccproxy.llms.formatters.utils import (
+    map_openai_finish_to_anthropic_stop,
+    openai_usage_to_anthropic_usage,
+)
+from ccproxy.llms.models import anthropic as anthropic_models
+from ccproxy.llms.models import openai as openai_models
+from ccproxy.llms.streaming.accumulators import OpenAIAccumulator
+
+
+logger = ccproxy.core.logging.get_logger(__name__)
+
+
+class OpenAIResponsesToAnthropicStreamAdapter:
+    """Stateful adapter for OpenAI Responses → Anthropic streaming."""
+
+    async def run(
+        self,
+        stream: AsyncIterator[Any],
+    ) -> AsyncGenerator[anthropic_models.MessageStreamEvent, None]:
+        async for event in self._convert_responses_stream(stream):
+            yield event
+
+    async def _convert_responses_stream(
+        self,
+        stream: AsyncIterator[Any],
+    ) -> AsyncGenerator[anthropic_models.MessageStreamEvent, None]:
+        """Translate OpenAI Responses streaming events into Anthropic message events."""
+
+        def _event_to_dict(raw: Any) -> dict[str, Any]:
+            if isinstance(raw, dict):
+                return raw
+            if hasattr(raw, "root"):
+                return _event_to_dict(raw.root)
+            if hasattr(raw, "model_dump"):
+                return cast(dict[str, Any], raw.model_dump(mode="json"))
+            return cast(dict[str, Any], {})
+
+        def _parse_tool_input(text: str) -> dict[str, Any]:
+            if not text:
+                return cast(dict[str, Any], {})
+            try:
+                parsed = json.loads(text)
+                return parsed if isinstance(parsed, dict) else {"arguments": text}
+            except Exception:
+                return {"arguments": text}
+
+        message_started = False
+        text_block_active = False
+        current_index = 0
+        final_stop_reason: str | None = None
+        final_stop_sequence: str | None = None
+        usage = anthropic_models.Usage(input_tokens=0, output_tokens=0)
+        reasoning_buffer: list[str] = []
+        tool_states = IndexedToolCallTracker()
+
+        async for raw_event in stream:
+            event = _event_to_dict(raw_event)
+            event_type = event.get("type") or event.get("event")
+            if not event_type:
+                continue
+
+            if event_type == "error":
+                payload = event.get("error") or {}
+                detail = (
+                    anthropic_models.ErrorDetail(**payload)
+                    if isinstance(payload, dict)
+                    else anthropic_models.ErrorDetail(message=str(payload))
+                )
+                yield anthropic_models.ErrorEvent(type="error", error=detail)
+                return
+
+            if not message_started:
+                response_meta = event.get("response") or {}
+                yield anthropic_models.MessageStartEvent(
+                    type="message_start",
+                    message=anthropic_models.MessageResponse(
+                        id=response_meta.get("id", "resp_stream"),
+                        type="message",
+                        role="assistant",
+                        content=[],
+                        model=response_meta.get("model", ""),
+                        stop_reason=None,
+                        stop_sequence=None,
+                        usage=usage,
+                    ),
+                )
+                message_started = True
+
+            if event_type == "response.output_text.delta":
+                delta = event.get("delta")
+                text = ""
+                if isinstance(delta, dict):
+                    text = delta.get("text") or ""
+                elif isinstance(delta, str):
+                    text = delta
+                if text:
+                    if not text_block_active:
+                        yield anthropic_models.ContentBlockStartEvent(
+                            type="content_block_start",
+                            index=current_index,
+                            content_block=anthropic_models.TextBlock(
+                                type="text", text=""
+                            ),
+                        )
+                        text_block_active = True
+                    yield anthropic_models.ContentBlockDeltaEvent(
+                        type="content_block_delta",
+                        index=current_index,
+                        delta=anthropic_models.TextBlock(type="text", text=text),
+                    )
+            elif event_type == "response.output_text.done":
+                if text_block_active:
+                    yield anthropic_models.ContentBlockStopEvent(
+                        type="content_block_stop", index=current_index
+                    )
+                    text_block_active = False
+                    current_index += 1
+
+            elif event_type == "response.reasoning_summary_text.delta":
+                delta = event.get("delta")
+                summary_piece = delta.get("text") if isinstance(delta, dict) else delta
+                if isinstance(summary_piece, str):
+                    reasoning_buffer.append(summary_piece)
+
+            elif event_type == "response.reasoning_summary_text.done":
+                if text_block_active:
+                    yield anthropic_models.ContentBlockStopEvent(
+                        type="content_block_stop", index=current_index
+                    )
+                    text_block_active = False
+                    current_index += 1
+                summary = "".join(reasoning_buffer)
+                reasoning_buffer.clear()
+                if summary:
+                    yield anthropic_models.ContentBlockStartEvent(
+                        type="content_block_start",
+                        index=current_index,
+                        content_block=anthropic_models.ThinkingBlock(
+                            type="thinking",
+                            thinking=summary,
+                            signature="",
+                        ),
+                    )
+                    yield anthropic_models.ContentBlockStopEvent(
+                        type="content_block_stop", index=current_index
+                    )
+                    current_index += 1
+
+            elif event_type == "response.function_call_arguments.delta":
+                output_index = int(event.get("output_index", 0) or 0)
+                state = tool_states.ensure(output_index)
+                if state.output_index < 0:
+                    state.output_index = output_index
+
+                delta = event.get("delta") or {}
+                delta_text = (
+                    delta.get("arguments") if isinstance(delta, dict) else delta
+                )
+                if isinstance(delta_text, str):
+                    state.add_arguments_part(delta_text)
+                    state.append_arguments(delta_text)
+
+                item_id_val = event.get("item_id")
+                if isinstance(item_id_val, str) and item_id_val:
+                    if not state.item_id:
+                        state.item_id = item_id_val
+
+                name_val = event.get("name")
+                if isinstance(name_val, str) and name_val:
+                    if not state.name:
+                        state.name = name_val
+
+                call_id_val = event.get("call_id")
+                if isinstance(call_id_val, str) and call_id_val:
+                    if not state.call_id:
+                        state.call_id = call_id_val
+
+            elif event_type == "response.function_call_arguments.done":
+                output_index = int(event.get("output_index", 0) or 0)
+                state = tool_states.ensure(output_index)
+                if state.output_index < 0:
+                    state.output_index = output_index
+
+                arguments = event.get("arguments") if isinstance(event, dict) else None
+                if isinstance(arguments, str) and arguments:
+                    state.append_arguments(arguments)
+                    state.final_arguments = arguments
+                elif not state.final_arguments:
+                    combined = state.arguments or "".join(state.arguments_parts)
+                    if combined:
+                        state.final_arguments = combined
+
+                item_id_val = event.get("item_id")
+                if isinstance(item_id_val, str) and item_id_val:
+                    state.item_id = state.item_id or item_id_val
+
+                name_val = event.get("name")
+                if isinstance(name_val, str) and name_val:
+                    state.name = state.name or name_val
+
+                call_id_val = event.get("call_id")
+                if isinstance(call_id_val, str) and call_id_val:
+                    state.call_id = state.call_id or call_id_val
+
+            elif event_type == "response.output_item.added":
+                item = event.get("item") or {}
+                item_type = item.get("type")
+                if item_type == "output_tool_call":
+                    output_index = int(
+                        item.get("output_index", event.get("output_index", 0)) or 0
+                    )
+                    state = tool_states.ensure(output_index)
+                    if state.output_index < 0:
+                        state.output_index = output_index
+                    item_id_val = item.get("id")
+                    if isinstance(item_id_val, str) and item_id_val:
+                        state.item_id = state.item_id or item_id_val
+                    name_val = item.get("name")
+                    if isinstance(name_val, str) and name_val:
+                        state.name = state.name or name_val
+                    call_id_val = item.get("call_id")
+                    if isinstance(call_id_val, str) and call_id_val:
+                        state.call_id = state.call_id or call_id_val
+
+            elif event_type == "response.output_item.done":
+                item = event.get("item") or {}
+                item_type = item.get("type")
+                if item_type == "output_text" and text_block_active:
+                    yield anthropic_models.ContentBlockStopEvent(
+                        type="content_block_stop", index=current_index
+                    )
+                    text_block_active = False
+                    current_index += 1
+                elif item_type == "output_tool_call":
+                    output_index = int(
+                        item.get("output_index", event.get("output_index", 0)) or 0
+                    )
+                    state = tool_states.ensure(output_index)
+                    if state.output_index < 0:
+                        state.output_index = output_index
+
+                    item_name = item.get("name")
+                    if isinstance(item_name, str) and item_name:
+                        state.name = state.name or item_name
+                    item_call_id = item.get("call_id")
+                    if isinstance(item_call_id, str) and item_call_id:
+                        state.call_id = state.call_id or item_call_id
+                    if item.get("id") and not state.item_id:
+                        state.item_id = item.get("id")
+
+                    args_str = (
+                        state.final_arguments
+                        or state.arguments
+                        or "".join(state.arguments_parts)
+                    )
+
+                    yield anthropic_models.ContentBlockStartEvent(
+                        type="content_block_start",
+                        index=current_index,
+                        content_block=anthropic_models.ToolUseBlock(
+                            type="tool_use",
+                            id=state.item_id or f"call_{state.index}",
+                            name=state.name or "tool",
+                            input=_parse_tool_input(args_str),
+                        ),
+                    )
+                    yield anthropic_models.ContentBlockStopEvent(
+                        type="content_block_stop", index=current_index
+                    )
+                    current_index += 1
+
+            elif event_type == "response.completed":
+                response = event.get("response") or {}
+                usage_data = response.get("usage") or {}
+                try:
+                    usage = anthropic_models.Usage.model_validate(usage_data)
+                except Exception:
+                    usage = anthropic_models.Usage(
+                        input_tokens=usage_data.get("input_tokens", 0),
+                        output_tokens=usage_data.get("output_tokens", 0),
+                    )
+                final_stop_reason = response.get("stop_reason")
+                final_stop_sequence = response.get("stop_sequence")
+                break
+
+        if text_block_active:
+            yield anthropic_models.ContentBlockStopEvent(
+                type="content_block_stop", index=current_index
+            )
+
+        if message_started:
+            yield anthropic_models.MessageDeltaEvent(
+                type="message_delta",
+                delta=anthropic_models.MessageDelta(
+                    stop_reason=map_openai_finish_to_anthropic_stop(final_stop_reason),
+                    stop_sequence=final_stop_sequence,
+                ),
+                usage=usage,
+            )
+            yield anthropic_models.MessageStopEvent(type="message_stop")
+
+
+class OpenAIChatToAnthropicStreamAdapter:
+    """Stateful adapter for OpenAI Chat → Anthropic streaming."""
+
+    def run(
+        self,
+        stream: AsyncIterator[Any],
+    ) -> AsyncGenerator[anthropic_models.MessageStreamEvent, None]:
+        return self._convert_chat_stream(stream)
+
+    def _convert_chat_stream(
+        self,
+        stream: AsyncIterator[openai_models.ChatCompletionChunk],
+    ) -> AsyncGenerator[anthropic_models.MessageStreamEvent, None]:
+        """Convert OpenAI ChatCompletion stream to Anthropic MessageStreamEvent stream."""
+
+        async def generator() -> AsyncGenerator[
+            anthropic_models.MessageStreamEvent, None
+        ]:
+            message_started = False
+            text_block_started = False
+            accumulated_content = ""
+            model_id = ""
+            current_index = 0
+            tool_tracker = ToolCallTracker()
+            openai_accumulator = OpenAIAccumulator()
+
+            def _parse_tool_input(text: str) -> dict[str, Any]:
+                if not text:
+                    return {}
+                try:
+                    parsed = json.loads(text)
+                    return parsed if isinstance(parsed, dict) else {"arguments": text}
+                except Exception:
+                    return {"arguments": text}
+
+            async for chunk in stream:
+                if isinstance(chunk, dict):
+                    chunk_payload = chunk
+                else:
+                    try:
+                        chunk_payload = chunk.model_dump(mode="json", exclude_none=True)
+                    except Exception:
+                        chunk_payload = chunk.model_dump(exclude_none=True)
+
+                choices = chunk_payload.get("choices")
+                if not choices:
+                    continue
+
+                choice = choices[0]
+                openai_accumulator.accumulate("", chunk_payload)
+
+                if chunk_payload.get("model"):
+                    model_id = chunk_payload["model"]
+                elif not isinstance(chunk, dict):
+                    model_id = getattr(chunk, "model", model_id) or model_id
+
+                delta = choice.get("delta") if isinstance(choice, dict) else {}
+                if not isinstance(delta, dict):
+                    delta = {}
+                finish_reason = choice.get("finish_reason")
+                content_piece = delta.get("content")
+
+                if not message_started:
+                    chunk_id = chunk_payload.get("id")
+                    if chunk_id is None and not isinstance(chunk, dict):
+                        chunk_id = getattr(chunk, "id", None)
+                    chunk_id = chunk_id or "msg_stream"
+
+                    yield anthropic_models.MessageStartEvent(
+                        type="message_start",
+                        message=anthropic_models.MessageResponse(
+                            id=chunk_id,
+                            type="message",
+                            role="assistant",
+                            content=[],
+                            model=model_id,
+                            stop_reason=None,
+                            stop_sequence=None,
+                            usage=anthropic_models.Usage(
+                                input_tokens=0, output_tokens=0
+                            ),
+                        ),
+                    )
+                    message_started = True
+
+                if content_piece:
+                    content_text = (
+                        content_piece
+                        if isinstance(content_piece, str)
+                        else str(content_piece)
+                    )
+                    if not text_block_started:
+                        yield anthropic_models.ContentBlockStartEvent(
+                            type="content_block_start",
+                            index=current_index,
+                            content_block=anthropic_models.TextBlock(
+                                type="text", text=""
+                            ),
+                        )
+                        text_block_started = True
+
+                    yield anthropic_models.ContentBlockDeltaEvent(
+                        type="content_block_delta",
+                        index=current_index,
+                        delta=anthropic_models.TextBlock(
+                            type="text", text=content_text
+                        ),
+                    )
+                    accumulated_content += content_text
+
+                if finish_reason:
+                    if text_block_started:
+                        yield anthropic_models.ContentBlockStopEvent(
+                            type="content_block_stop",
+                            index=current_index,
+                        )
+                        text_block_started = False
+                        current_index += 1
+
+                    complete_calls = openai_accumulator.get_complete_tool_calls()
+                    for tool in complete_calls:
+                        tool_id_value = tool.get("id")
+                        call_identifier = str(
+                            tool_id_value
+                            or f"call_{tool.get('index', len(tool_tracker.values()))}"
+                        )
+                        state = tool_tracker.ensure(call_identifier)
+                        if state.name is None:
+                            state.name = tool.get("function", {}).get(
+                                "name"
+                            ) or tool.get("name")
+                        if state.call_id is None and isinstance(tool_id_value, str):
+                            state.call_id = tool_id_value
+                        function_payload = tool.get("function", {})
+                        arguments_payload = function_payload.get("arguments")
+                        if isinstance(arguments_payload, str) and arguments_payload:
+                            state.final_arguments = arguments_payload
+                            state.arguments = arguments_payload
+                        if state.item_id is None:
+                            state.item_id = call_identifier
+
+                    for state in tool_tracker.values():
+                        if state.item_id is None:
+                            state.item_id = (
+                                state.call_id or state.id or f"call_{state.index}"
+                            )
+                        for event in emit_anthropic_tool_use_events(
+                            current_index,
+                            state,
+                            parser=_parse_tool_input,
+                        ):
+                            yield event
+                        current_index += 1
+
+                    usage_payload = chunk_payload.get("usage")
+                    if usage_payload is None and not isinstance(chunk, dict):
+                        usage_obj = getattr(chunk, "usage", None)
+                        if usage_obj is not None:
+                            if hasattr(usage_obj, "model_dump"):
+                                usage_payload = usage_obj.model_dump()
+                            else:
+                                usage_payload = {
+                                    "prompt_tokens": getattr(
+                                        usage_obj, "prompt_tokens", 0
+                                    ),
+                                    "completion_tokens": getattr(
+                                        usage_obj, "completion_tokens", 0
+                                    ),
+                                }
+
+                    anthropic_usage = (
+                        openai_usage_to_anthropic_usage(usage_payload)
+                        if usage_payload is not None
+                        else anthropic_models.Usage(input_tokens=0, output_tokens=0)
+                    )
+
+                    mapped_stop = map_openai_finish_to_anthropic_stop(finish_reason)
+
+                    yield anthropic_models.MessageDeltaEvent(
+                        type="message_delta",
+                        delta=anthropic_models.MessageDelta(stop_reason=mapped_stop),
+                        usage=anthropic_usage,
+                    )
+                    yield anthropic_models.MessageStopEvent(type="message_stop")
+                    break
+
+        return generator()
+
+
+async def convert__openai_responses_to_anthropic_messages__stream(
+    stream: AsyncIterator[Any],
+) -> AsyncGenerator[anthropic_models.MessageStreamEvent, None]:
+    """Translate OpenAI Responses streaming events into Anthropic message events."""
+
+    adapter = OpenAIResponsesToAnthropicStreamAdapter()
+    async for event in adapter.run(stream):
+        yield event
+
+
+def convert__openai_chat_to_anthropic_messages__stream(
+    stream: AsyncIterator[Any],
+) -> AsyncGenerator[anthropic_models.MessageStreamEvent, None]:
+    """Translate OpenAI ChatCompletion streams into Anthropic message events."""
+
+    adapter = OpenAIChatToAnthropicStreamAdapter()
+    return adapter.run(stream)
+
+
+__all__ = [
+    "OpenAIChatToAnthropicStreamAdapter",
+    "OpenAIResponsesToAnthropicStreamAdapter",
+    "convert__openai_chat_to_anthropic_messages__stream",
+    "convert__openai_responses_to_anthropic_messages__stream",
+]

ccproxy/llms/formatters/openai_to_openai/__init__.py

@@ -0,0 +1,53 @@
+"""Facade module exposing OpenAI↔OpenAI formatter entry points."""
+
+import sys
+from types import ModuleType
+
+from . import streams as _streams
+from .requests import (
+    convert__openai_chat_to_openai_responses__request,
+    convert__openai_responses_to_openaichat__request,
+)
+from .responses import (
+    convert__openai_chat_to_openai_responses__response,
+    convert__openai_completion_usage_to_openai_responses__usage,
+    convert__openai_responses_to_openai_chat__response,
+    convert__openai_responses_usage_to_openai_completion__usage,
+)
+from .streams import (
+    OpenAIChatToResponsesStreamAdapter,
+    OpenAIResponsesToChatStreamAdapter,
+    convert__openai_chat_to_openai_responses__stream,
+    convert__openai_responses_to_openai_chat__stream,
+)
+
+
+__all__ = [
+    "convert__openai_chat_to_openai_responses__request",
+    "convert__openai_responses_to_openaichat__request",
+    "convert__openai_chat_to_openai_responses__response",
+    "convert__openai_completion_usage_to_openai_responses__usage",
+    "convert__openai_responses_to_openai_chat__response",
+    "convert__openai_responses_usage_to_openai_completion__usage",
+    "OpenAIChatToResponsesStreamAdapter",
+    "OpenAIResponsesToChatStreamAdapter",
+    "convert__openai_chat_to_openai_responses__stream",
+    "convert__openai_responses_to_openai_chat__stream",
+]
+
+
+class _OpenAIToOpenAIModule(ModuleType):
+    _propagated_names = {
+        "OpenAIChatToResponsesStreamAdapter",
+        "OpenAIResponsesToChatStreamAdapter",
+    }
+
+    def __setattr__(self, name: str, value: object) -> None:
+        super().__setattr__(name, value)
+        if name in self._propagated_names and hasattr(_streams, name):
+            setattr(_streams, name, value)
+
+
+_module = sys.modules[__name__]
+if not isinstance(_module, _OpenAIToOpenAIModule):
+    _module.__class__ = _OpenAIToOpenAIModule