livekit-plugins-anthropic 0.2.13__tar.gz → 1.0.0.dev4__tar.gz

livekit_plugins_anthropic-1.0.0.dev4/.gitignore

@@ -0,0 +1,168 @@
+**/.vscode
+**/.DS_Store
+
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+cover/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+.pybuilder/
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pyenv
+#   For a library or package, you might want to ignore these files since the code is
+#   intended to run in multiple environments; otherwise, check them in:
+# .python-version
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# poetry
+#   Similar to Pipfile.lock, it is generally recommended to include poetry.lock in version control.
+#   This is especially recommended for binary packages to ensure reproducibility, and is more
+#   commonly ignored for libraries.
+#   https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
+#poetry.lock
+
+# pdm
+#   Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
+#pdm.lock
+#   pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
+#   in version control.
+#   https://pdm.fming.dev/#use-with-ide
+.pdm.toml
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# pytype static type analyzer
+.pytype/
+
+# Cython debug symbols
+cython_debug/
+
+# PyCharm
+#  JetBrains specific template is maintained in a separate JetBrains.gitignore that can
+#  be found at https://github.com/github/gitignore/blob/main/Global/JetBrains.gitignore
+#  and can be added to the global gitignore or merged into this file.  For a more nuclear
+#  option (not recommended) you can uncomment the following to ignore the entire idea folder.
+.idea/
+
+node_modules
+
+credentials.json
+pyrightconfig.json

{livekit_plugins_anthropic-0.2.13 → livekit_plugins_anthropic-1.0.0.dev4}/PKG-INFO

@@ -1,36 +1,25 @@
-Metadata-Version: 2.2
+Metadata-Version: 2.4
 Name: livekit-plugins-anthropic
-Version: 0.2.13
+Version: 1.0.0.dev4
 Summary: Agent Framework plugin for services from Anthropic
-Home-page: https://github.com/livekit/agents
-License: Apache-2.0
 Project-URL: Documentation, https://docs.livekit.io
 Project-URL: Website, https://livekit.io/
 Project-URL: Source, https://github.com/livekit/agents
-Keywords: webrtc,realtime,audio,video,livekit
+Author-email: LiveKit <support@livekit.io>
+License-Expression: Apache-2.0
+Keywords: audio,livekit,realtime,video,webrtc
 Classifier: Intended Audience :: Developers
 Classifier: License :: OSI Approved :: Apache Software License
-Classifier: Topic :: Multimedia :: Sound/Audio
-Classifier: Topic :: Multimedia :: Video
-Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
 Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3 :: Only
 Classifier: Programming Language :: Python :: 3.9
 Classifier: Programming Language :: Python :: 3.10
-Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Topic :: Multimedia :: Sound/Audio
+Classifier: Topic :: Multimedia :: Video
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
 Requires-Python: >=3.9.0
+Requires-Dist: livekit-agents>=1.0.0.dev4
 Description-Content-Type: text/markdown
-Requires-Dist: livekit-agents<1.0.0,>=0.12.16
-Requires-Dist: anthropic>=0.34
-Dynamic: classifier
-Dynamic: description
-Dynamic: description-content-type
-Dynamic: home-page
-Dynamic: keywords
-Dynamic: license
-Dynamic: project-url
-Dynamic: requires-dist
-Dynamic: requires-python
-Dynamic: summary
 
 # LiveKit Plugins Anthropic
 

livekit_plugins_anthropic-1.0.0.dev4/livekit/plugins/anthropic/llm.py

@@ -0,0 +1,320 @@
+# Copyright 2023 LiveKit, Inc.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from __future__ import annotations
+
+import os
+from collections.abc import Awaitable
+from dataclasses import dataclass
+from typing import Any, Literal
+
+import httpx
+
+import anthropic
+from livekit.agents import APIConnectionError, APIStatusError, APITimeoutError, llm
+from livekit.agents.llm import ToolChoice
+from livekit.agents.llm.chat_context import ChatContext
+from livekit.agents.llm.tool_context import FunctionTool
+from livekit.agents.types import (
+    DEFAULT_API_CONNECT_OPTIONS,
+    NOT_GIVEN,
+    APIConnectOptions,
+    NotGivenOr,
+)
+from livekit.agents.utils import is_given
+
+from .models import ChatModels
+from .utils import to_chat_ctx, to_fnc_ctx
+
+
+@dataclass
+class _LLMOptions:
+    model: str | ChatModels
+    user: NotGivenOr[str]
+    temperature: NotGivenOr[float]
+    parallel_tool_calls: NotGivenOr[bool]
+    tool_choice: NotGivenOr[ToolChoice | Literal["auto", "required", "none"]]
+    caching: NotGivenOr[Literal["ephemeral"]]
+    top_k: NotGivenOr[int]
+    max_tokens: NotGivenOr[int]
+    """If set to "ephemeral", the system prompt, tools, and chat history will be cached."""
+
+
+class LLM(llm.LLM):
+    def __init__(
+        self,
+        *,
+        model: str | ChatModels = "claude-3-5-sonnet-20241022",
+        api_key: NotGivenOr[str] = NOT_GIVEN,
+        base_url: NotGivenOr[str] = NOT_GIVEN,
+        user: NotGivenOr[str] = NOT_GIVEN,
+        client: anthropic.AsyncClient | None = None,
+        top_k: NotGivenOr[int] = NOT_GIVEN,
+        max_tokens: NotGivenOr[int] = NOT_GIVEN,
+        temperature: NotGivenOr[float] = NOT_GIVEN,
+        parallel_tool_calls: NotGivenOr[bool] = NOT_GIVEN,
+        tool_choice: NotGivenOr[ToolChoice | Literal["auto", "required", "none"]] = NOT_GIVEN,
+        caching: NotGivenOr[Literal["ephemeral"]] = NOT_GIVEN,
+    ) -> None:
+        """
+        Create a new instance of Anthropic LLM.
+
+        ``api_key`` must be set to your Anthropic API key, either using the argument or by setting
+        the ``ANTHROPIC_API_KEY`` environmental variable.
+
+        model (str | ChatModels): The model to use. Defaults to "claude-3-5-sonnet-20241022".
+        api_key (str | None): The Anthropic API key. Defaults to the ANTHROPIC_API_KEY environment variable.
+        base_url (str | None): The base URL for the Anthropic API. Defaults to None.
+        user (str | None): The user for the Anthropic API. Defaults to None.
+        client (anthropic.AsyncClient | None): The Anthropic client to use. Defaults to None.
+        temperature (float | None): The temperature for the Anthropic API. Defaults to None.
+        parallel_tool_calls (bool | None): Whether to parallelize tool calls. Defaults to None.
+        tool_choice (Union[ToolChoice, Literal["auto", "required", "none"]] | None): The tool choice for the Anthropic API. Defaults to "auto".
+        caching (Literal["ephemeral"] | None): If set to "ephemeral", caching will be enabled for the system prompt, tools, and chat history.
+        """
+
+        super().__init__()
+
+        self._opts = _LLMOptions(
+            model=model,
+            user=user,
+            temperature=temperature,
+            parallel_tool_calls=parallel_tool_calls,
+            tool_choice=tool_choice,
+            caching=caching,
+            top_k=top_k,
+            max_tokens=max_tokens,
+        )
+        api_key = api_key or os.environ.get("ANTHROPIC_API_KEY")
+        if not is_given(api_key):
+            raise ValueError("Anthropic API key is required")
+
+        self._client = anthropic.AsyncClient(
+            api_key=api_key or None,
+            base_url=base_url or None,
+            http_client=httpx.AsyncClient(
+                timeout=5.0,
+                follow_redirects=True,
+                limits=httpx.Limits(
+                    max_connections=1000,
+                    max_keepalive_connections=100,
+                    keepalive_expiry=120,
+                ),
+            ),
+        )
+
+    def chat(
+        self,
+        *,
+        chat_ctx: ChatContext,
+        tools: list[FunctionTool] | None = None,
+        conn_options: APIConnectOptions = DEFAULT_API_CONNECT_OPTIONS,
+        parallel_tool_calls: NotGivenOr[bool] = NOT_GIVEN,
+        tool_choice: NotGivenOr[ToolChoice | Literal["auto", "required", "none"]] = NOT_GIVEN,
+        extra_kwargs: NotGivenOr[dict[str, Any]] = NOT_GIVEN,
+    ) -> LLMStream:
+        extra = {}
+
+        if is_given(extra_kwargs):
+            extra.update(extra_kwargs)
+
+        if is_given(self._opts.user):
+            extra["user"] = self._opts.user
+
+        if is_given(self._opts.temperature):
+            extra["temperature"] = self._opts.temperature
+
+        if is_given(self._opts.top_k):
+            extra["top_k"] = self._opts.top_k
+
+        extra["max_tokens"] = self._opts.max_tokens if is_given(self._opts.max_tokens) else 1024
+
+        if tools:
+            extra["tools"] = to_fnc_ctx(tools, self._opts.caching)
+            tool_choice = tool_choice if is_given(tool_choice) else self._opts.tool_choice
+            if is_given(tool_choice):
+                anthropic_tool_choice: dict[str, Any] | None = {"type": "auto"}
+                if isinstance(tool_choice, ToolChoice):
+                    if tool_choice.type == "function":
+                        anthropic_tool_choice = {
+                            "type": "tool",
+                            "name": tool_choice.name,
+                        }
+                elif isinstance(tool_choice, str):
+                    if tool_choice == "required":
+                        anthropic_tool_choice = {"type": "any"}
+                    elif tool_choice == "none":
+                        extra["tools"] = []
+                        anthropic_tool_choice = None
+                if anthropic_tool_choice is not None:
+                    parallel_tool_calls = (
+                        parallel_tool_calls
+                        if is_given(parallel_tool_calls)
+                        else self._opts.parallel_tool_calls
+                    )
+                    if is_given(parallel_tool_calls):
+                        anthropic_tool_choice["disable_parallel_tool_use"] = not parallel_tool_calls
+                    extra["tool_choice"] = anthropic_tool_choice
+
+        anthropic_ctx, system_message = to_chat_ctx(chat_ctx, id(self), caching=self._opts.caching)
+
+        if system_message:
+            extra["system"] = [system_message]
+
+        stream = self._client.messages.create(
+            messages=anthropic_ctx,
+            model=self._opts.model,
+            stream=True,
+            **extra,
+        )
+
+        return LLMStream(
+            self,
+            anthropic_stream=stream,
+            chat_ctx=chat_ctx,
+            tools=tools,
+            conn_options=conn_options,
+        )
+
+
+class LLMStream(llm.LLMStream):
+    def __init__(
+        self,
+        llm: LLM,
+        *,
+        anthropic_stream: Awaitable[anthropic.AsyncStream[anthropic.types.RawMessageStreamEvent]],
+        chat_ctx: llm.ChatContext,
+        tools: list[FunctionTool] | None,
+        conn_options: APIConnectOptions,
+    ) -> None:
+        super().__init__(llm, chat_ctx=chat_ctx, tools=tools, conn_options=conn_options)
+        self._awaitable_anthropic_stream = anthropic_stream
+        self._anthropic_stream: (
+            anthropic.AsyncStream[anthropic.types.RawMessageStreamEvent] | None
+        ) = None
+
+        # current function call that we're waiting for full completion (args are streamed)
+        self._tool_call_id: str | None = None
+        self._fnc_name: str | None = None
+        self._fnc_raw_arguments: str | None = None
+
+        self._request_id: str = ""
+        self._ignoring_cot = False  # ignore chain of thought
+        self._input_tokens = 0
+        self._cache_creation_tokens = 0
+        self._cache_read_tokens = 0
+        self._output_tokens = 0
+
+    async def _run(self) -> None:
+        retryable = True
+        try:
+            if not self._anthropic_stream:
+                self._anthropic_stream = await self._awaitable_anthropic_stream
+
+            async with self._anthropic_stream as stream:
+                async for event in stream:
+                    chat_chunk = self._parse_event(event)
+                    if chat_chunk is not None:
+                        self._event_ch.send_nowait(chat_chunk)
+                        retryable = False
+
+                self._event_ch.send_nowait(
+                    llm.ChatChunk(
+                        id=self._request_id,
+                        usage=llm.CompletionUsage(
+                            completion_tokens=self._output_tokens,
+                            prompt_tokens=self._input_tokens,
+                            total_tokens=self._input_tokens
+                            + self._output_tokens
+                            + self._cache_creation_tokens
+                            + self._cache_read_tokens,
+                            cache_creation_input_tokens=self._cache_creation_tokens,
+                            cache_read_input_tokens=self._cache_read_tokens,
+                        ),
+                    )
+                )
+        except anthropic.APITimeoutError as e:
+            raise APITimeoutError(retryable=retryable) from e
+        except anthropic.APIStatusError as e:
+            raise APIStatusError(
+                e.message,
+                status_code=e.status_code,
+                request_id=e.request_id,
+                body=e.body,
+            ) from e
+        except Exception as e:
+            raise APIConnectionError(retryable=retryable) from e
+
+    def _parse_event(self, event: anthropic.types.RawMessageStreamEvent) -> llm.ChatChunk | None:
+        if event.type == "message_start":
+            self._request_id = event.message.id
+            self._input_tokens = event.message.usage.input_tokens
+            self._output_tokens = event.message.usage.output_tokens
+            if event.message.usage.cache_creation_input_tokens:
+                self._cache_creation_tokens = event.message.usage.cache_creation_input_tokens
+            if event.message.usage.cache_read_input_tokens:
+                self._cache_read_tokens = event.message.usage.cache_read_input_tokens
+        elif event.type == "message_delta":
+            self._output_tokens += event.usage.output_tokens
+        elif event.type == "content_block_start":
+            if event.content_block.type == "tool_use":
+                self._tool_call_id = event.content_block.id
+                self._fnc_name = event.content_block.name
+                self._fnc_raw_arguments = ""
+        elif event.type == "content_block_delta":
+            delta = event.delta
+            if delta.type == "text_delta":
+                text = delta.text
+
+                if self._tools is not None:
+                    # anthropic may inject COC when using functions
+                    if text.startswith("<thinking>"):
+                        self._ignoring_cot = True
+                    elif self._ignoring_cot and "</thinking>" in text:
+                        text = text.split("</thinking>")[-1]
+                        self._ignoring_cot = False
+
+                if self._ignoring_cot:
+                    return None
+
+                return llm.ChatChunk(
+                    id=self._request_id,
+                    delta=llm.ChoiceDelta(content=text, role="assistant"),
+                )
+            elif delta.type == "input_json_delta":
+                assert self._fnc_raw_arguments is not None
+                self._fnc_raw_arguments += delta.partial_json
+
+        elif event.type == "content_block_stop":
+            if self._tool_call_id is not None:
+                assert self._fnc_name is not None
+                assert self._fnc_raw_arguments is not None
+
+                chat_chunk = llm.ChatChunk(
+                    id=self._request_id,
+                    delta=llm.ChoiceDelta(
+                        role="assistant",
+                        tool_calls=[
+                            llm.FunctionToolCall(
+                                arguments=self._fnc_raw_arguments or "",
+                                name=self._fnc_name or "",
+                                call_id=self._tool_call_id or "",
+                            )
+                        ],
+                    ),
+                )
+                self._tool_call_id = self._fnc_raw_arguments = self._fnc_name = None
+                return chat_chunk
+
+        return None

livekit_plugins_anthropic-1.0.0.dev4/livekit/plugins/anthropic/utils.py

@@ -0,0 +1,140 @@
+import base64
+import json
+from typing import Any, Literal
+
+import anthropic
+from livekit.agents import llm
+from livekit.agents.llm import FunctionTool
+
+CACHE_CONTROL_EPHEMERAL = anthropic.types.CacheControlEphemeralParam(type="ephemeral")
+
+__all__ = ["to_fnc_ctx", "to_chat_ctx"]
+
+
+def to_fnc_ctx(
+    fncs: list[FunctionTool], caching: Literal["ephemeral"] | None
+) -> list[anthropic.types.ToolParam]:
+    tools: list[anthropic.types.ToolParam] = []
+    for i, fnc in enumerate(fncs):
+        cache_ctrl = (
+            CACHE_CONTROL_EPHEMERAL if (i == len(fncs) - 1) and caching == "ephemeral" else None
+        )
+        tools.append(_build_anthropic_schema(fnc, cache_ctrl=cache_ctrl))
+
+    return tools
+
+
+def to_chat_ctx(
+    chat_ctx: llm.ChatContext,
+    cache_key: Any,
+    caching: Literal["ephemeral"] | None,
+) -> list[anthropic.types.MessageParam]:
+    messages: list[anthropic.types.MessageParam] = []
+    system_message: anthropic.types.TextBlockParam | None = None
+    current_role: str | None = None
+    content: list[anthropic.types.TextBlockParam] = []
+    for i, msg in enumerate(chat_ctx.items):
+        if msg.type == "message" and msg.role == "system":
+            for content in msg.content:
+                if isinstance(content, str):
+                    system_message = anthropic.types.TextBlockParam(
+                        text=content,
+                        type="text",
+                        cache_control=CACHE_CONTROL_EPHEMERAL if caching == "ephemeral" else None,
+                    )
+            continue
+
+        cache_ctrl = (
+            CACHE_CONTROL_EPHEMERAL
+            if (i == len(chat_ctx.items) - 1) and caching == "ephemeral"
+            else None
+        )
+        if msg.type == "message":
+            role = "assistant" if msg.role == "assistant" else "user"
+        elif msg.type == "function_call":
+            role = "assistant"
+        elif msg.type == "function_call_output":
+            role = "user"
+
+        if role != current_role:
+            if current_role is not None and content:
+                messages.append(anthropic.types.MessageParam(role=current_role, content=content))
+            content = []
+            current_role = role
+
+        if msg.type == "message":
+            for c in msg.content:
+                if isinstance(c, str):
+                    content.append(
+                        anthropic.types.TextBlockParam(
+                            text=c, type="text", cache_control=cache_ctrl
+                        )
+                    )
+                elif isinstance(c, llm.ImageContent):
+                    content.append(_to_image_content(c, cache_key, cache_ctrl=cache_ctrl))
+        elif msg.type == "function_call":
+            content.append(
+                anthropic.types.ToolUseBlockParam(
+                    id=msg.call_id,
+                    type="tool_use",
+                    name=msg.name,
+                    input=json.loads(msg.arguments or "{}"),
+                    cache_control=cache_ctrl,
+                )
+            )
+        elif msg.type == "function_call_output":
+            content.append(
+                anthropic.types.ToolResultBlockParam(
+                    tool_use_id=msg.call_id,
+                    type="tool_result",
+                    content=msg.output,
+                    cache_control=cache_ctrl,
+                )
+            )
+
+    if current_role is not None and content:
+        messages.append(anthropic.types.MessageParam(role=current_role, content=content))
+
+    # ensure the messages starts with a "user" message
+    if not messages or messages[0]["role"] != "user":
+        messages.insert(
+            0,
+            anthropic.types.MessageParam(
+                role="user", content=[anthropic.types.TextBlockParam(text="(empty)", type="text")]
+            ),
+        )
+
+    return messages, system_message
+
+
+def _to_image_content(
+    image: llm.ImageContent,
+    cache_key: Any,
+    cache_ctrl: anthropic.types.CacheControlEphemeralParam | None,
+) -> anthropic.types.ImageBlockParam:
+    img = llm.utils.serialize_image(image)
+    if cache_key not in image._cache:
+        image._cache[cache_key] = img.data_bytes
+    b64_data = base64.b64encode(image._cache[cache_key]).decode("utf-8")
+    return {
+        "type": "image",
+        "source": {
+            "type": "base64",
+            "data": f"data:{img.media_type};base64,{b64_data}",
+            "media_type": img.media_type,
+        },
+        "cache_control": cache_ctrl,
+    }
+
+
+def _build_anthropic_schema(
+    function_tool: FunctionTool,
+    cache_ctrl: anthropic.types.CacheControlEphemeralParam | None = None,
+) -> anthropic.types.ToolParam:
+    fnc = llm.utils.build_legacy_openai_schema(function_tool, internally_tagged=True)
+    return anthropic.types.ToolParam(
+        name=fnc["name"],
+        description=fnc["description"] or "",
+        input_schema=fnc["parameters"],
+        cache_control=cache_ctrl,
+    )

{livekit_plugins_anthropic-0.2.13 → livekit_plugins_anthropic-1.0.0.dev4}/livekit/plugins/anthropic/version.py

@@ -12,4 +12,4 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-__version__ = "0.2.13"
+__version__ = "1.0.0.dev4"