goose-py 0.9.15__tar.gz → 0.10.0__tar.gz

{goose_py-0.9.15 → goose_py-0.10.0}/.github/workflows/publish.yml

@@ -45,4 +45,3 @@ jobs:
 
       - name: Publish package
         run: uv publish
-        continue-on-error: true

{goose_py-0.9.15 → goose_py-0.10.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: goose-py
-Version: 0.9.15
+Version: 0.10.0
 Summary: A tool for AI workflows based on human-computer collaboration and structured output.
 Author-email: Nash Taylor <nash@chelle.ai>, Joshua Cook <joshua@chelle.ai>, Michael Sankur <michael@chelle.ai>
 Requires-Python: >=3.12

goose_py-0.10.0/goose/_internal/agent.py

@@ -0,0 +1,245 @@
+import logging
+from datetime import datetime
+from typing import Any, Literal, Protocol, overload
+
+from litellm import acompletion
+from pydantic import ValidationError
+
+from goose._internal.types.telemetry import AgentResponse
+from goose.errors import Honk
+
+from .result import FindReplaceResponse, Result, TextResult
+from .types.agent import AIModel, LLMMessage
+
+
+class IAgentLogger(Protocol):
+    async def __call__(self, *, response: AgentResponse[Any]) -> None: ...
+
+
+class Agent:
+    def __init__(
+        self,
+        *,
+        flow_name: str,
+        run_id: str,
+        logger: IAgentLogger | None = None,
+    ) -> None:
+        self.flow_name = flow_name
+        self.run_id = run_id
+        self.logger = logger
+
+    async def generate[R: Result](
+        self,
+        *,
+        messages: list[LLMMessage],
+        model: AIModel,
+        task_name: str,
+        response_model: type[R] = TextResult,
+        system: LLMMessage | None = None,
+    ) -> R:
+        start_time = datetime.now()
+        if system is not None:
+            messages.insert(0, system)
+
+        if response_model is TextResult:
+            response = await acompletion(model=model.value, messages=messages)
+            parsed_response = response_model.model_validate({"text": response.choices[0].message.content})
+        else:
+            response = await acompletion(
+                model=model.value,
+                messages=messages,
+                response_format=response_model,
+            )
+            parsed_response = response_model.model_validate_json(response.choices[0].message.content)
+
+        end_time = datetime.now()
+        agent_response = AgentResponse(
+            response=parsed_response,
+            run_id=self.run_id,
+            flow_name=self.flow_name,
+            task_name=task_name,
+            model=model,
+            system=system,
+            input_messages=messages,
+            input_tokens=response.usage.prompt_tokens,
+            output_tokens=response.usage.completion_tokens,
+            start_time=start_time,
+            end_time=end_time,
+        )
+
+        if self.logger is not None:
+            await self.logger(response=agent_response)
+        else:
+            logging.info(agent_response.model_dump())
+
+        return parsed_response
+
+    async def ask(
+        self, *, messages: list[LLMMessage], model: AIModel, task_name: str, system: LLMMessage | None = None
+    ) -> str:
+        start_time = datetime.now()
+
+        if system is not None:
+            messages.insert(0, system)
+        response = await acompletion(model=model.value, messages=messages)
+
+        end_time = datetime.now()
+        agent_response = AgentResponse(
+            response=response.choices[0].message.content,
+            run_id=self.run_id,
+            flow_name=self.flow_name,
+            task_name=task_name,
+            model=model,
+            system=system,
+            input_messages=messages,
+            input_tokens=response.usage.prompt_tokens,
+            output_tokens=response.usage.completion_tokens,
+            start_time=start_time,
+            end_time=end_time,
+        )
+
+        if self.logger is not None:
+            await self.logger(response=agent_response)
+        else:
+            logging.info(agent_response.model_dump())
+
+        return response.choices[0].message.content
+
+    async def refine[R: Result](
+        self,
+        *,
+        messages: list[LLMMessage],
+        model: AIModel,
+        task_name: str,
+        response_model: type[R],
+        system: LLMMessage | None = None,
+    ) -> R:
+        start_time = datetime.now()
+
+        if system is not None:
+            messages.insert(0, system)
+
+        find_replace_response = await acompletion(
+            model=model.value, messages=messages, response_format=FindReplaceResponse
+        )
+        parsed_find_replace_response = FindReplaceResponse.model_validate_json(
+            find_replace_response.choices[0].message.content
+        )
+
+        end_time = datetime.now()
+        agent_response = AgentResponse(
+            response=parsed_find_replace_response,
+            run_id=self.run_id,
+            flow_name=self.flow_name,
+            task_name=task_name,
+            model=model,
+            system=system,
+            input_messages=messages,
+            input_tokens=find_replace_response.usage.prompt_tokens,
+            output_tokens=find_replace_response.usage.completion_tokens,
+            start_time=start_time,
+            end_time=end_time,
+        )
+
+        if self.logger is not None:
+            await self.logger(response=agent_response)
+        else:
+            logging.info(agent_response.model_dump())
+
+        refined_response = self.__apply_find_replace(
+            result=self.__find_last_result(messages=messages, response_model=response_model),
+            find_replace_response=parsed_find_replace_response,
+            response_model=response_model,
+        )
+
+        return refined_response
+
+    @overload
+    async def __call__[R: Result](
+        self,
+        *,
+        messages: list[LLMMessage],
+        model: AIModel,
+        task_name: str,
+        mode: Literal["generate"],
+        response_model: type[R],
+        system: LLMMessage | None = None,
+    ) -> R: ...
+
+    @overload
+    async def __call__[R: Result](
+        self,
+        *,
+        messages: list[LLMMessage],
+        model: AIModel,
+        task_name: str,
+        mode: Literal["ask"],
+        response_model: type[R] = TextResult,
+        system: LLMMessage | None = None,
+    ) -> str: ...
+
+    @overload
+    async def __call__[R: Result](
+        self,
+        *,
+        messages: list[LLMMessage],
+        model: AIModel,
+        task_name: str,
+        response_model: type[R],
+        mode: Literal["refine"],
+        system: LLMMessage | None = None,
+    ) -> R: ...
+
+    @overload
+    async def __call__[R: Result](
+        self,
+        *,
+        messages: list[LLMMessage],
+        model: AIModel,
+        task_name: str,
+        response_model: type[R],
+        system: LLMMessage | None = None,
+    ) -> R: ...
+
+    async def __call__[R: Result](
+        self,
+        *,
+        messages: list[LLMMessage],
+        model: AIModel,
+        task_name: str,
+        response_model: type[R] = TextResult,
+        mode: Literal["generate", "ask", "refine"] = "generate",
+        system: LLMMessage | None = None,
+    ) -> R | str:
+        match mode:
+            case "generate":
+                return await self.generate(
+                    messages=messages, model=model, task_name=task_name, response_model=response_model, system=system
+                )
+            case "ask":
+                return await self.ask(messages=messages, model=model, task_name=task_name, system=system)
+            case "refine":
+                return await self.refine(
+                    messages=messages, model=model, task_name=task_name, response_model=response_model, system=system
+                )
+
+    def __apply_find_replace[R: Result](
+        self, *, result: R, find_replace_response: FindReplaceResponse, response_model: type[R]
+    ) -> R:
+        dumped_result = result.model_dump_json()
+        for replacement in find_replace_response.replacements:
+            dumped_result = dumped_result.replace(replacement.find, replacement.replace)
+
+        return response_model.model_validate_json(dumped_result)
+
+    def __find_last_result[R: Result](self, *, messages: list[LLMMessage], response_model: type[R]) -> R:
+        for message in reversed(messages):
+            if message["role"] == "assistant":
+                try:
+                    only_part = message["content"][0]
+                    if only_part["type"] == "text":
+                        return response_model.model_validate_json(only_part["text"])
+                except ValidationError:
+                    continue
+
+        raise Honk("No last result found, failed to refine")

goose_py-0.10.0/goose/_internal/conversation.py

@@ -0,0 +1,52 @@
+from typing import Self
+
+from pydantic import BaseModel
+
+from goose.errors import Honk
+
+from .result import Result
+from .types.agent import AssistantMessage, LLMMessage, SystemMessage, UserMessage
+
+
+class Conversation[R: Result](BaseModel):
+    user_messages: list[UserMessage]
+    assistant_messages: list[R | str]
+    context: SystemMessage | None = None
+
+    @property
+    def awaiting_response(self) -> bool:
+        return len(self.user_messages) == len(self.assistant_messages)
+
+    def render(self) -> list[LLMMessage]:
+        messages: list[LLMMessage] = []
+        if self.context is not None:
+            messages.append(self.context.render())
+
+        for message_index in range(len(self.user_messages)):
+            message = self.assistant_messages[message_index]
+            if isinstance(message, str):
+                messages.append(AssistantMessage(text=message).render())
+            else:
+                messages.append(AssistantMessage(text=message.model_dump_json()).render())
+
+            messages.append(self.user_messages[message_index].render())
+
+        if len(self.assistant_messages) > len(self.user_messages):
+            message = self.assistant_messages[-1]
+            if isinstance(message, str):
+                messages.append(AssistantMessage(text=message).render())
+            else:
+                messages.append(AssistantMessage(text=message.model_dump_json()).render())
+
+        return messages
+
+    def undo(self) -> Self:
+        if len(self.user_messages) == 0:
+            raise Honk("Cannot undo, no user messages")
+
+        if len(self.assistant_messages) == 0:
+            raise Honk("Cannot undo, no assistant messages")
+
+        self.user_messages.pop()
+        self.assistant_messages.pop()
+        return self

goose_py-0.10.0/goose/_internal/result.py

@@ -0,0 +1,20 @@
+from pydantic import BaseModel, ConfigDict, Field
+
+
+class Result(BaseModel):
+    model_config = ConfigDict(frozen=True)
+
+
+class TextResult(Result):
+    text: str
+
+
+class Replacement(BaseModel):
+    find: str = Field(description="Text to find, to be replaced with `replace`")
+    replace: str = Field(description="Text to replace `find` with")
+
+
+class FindReplaceResponse(BaseModel):
+    replacements: list[Replacement] = Field(
+        description="List of replacements to make in the previous result to satisfy the user's request"
+    )

{goose_py-0.9.15 → goose_py-0.10.0}/goose/_internal/state.py

@@ -4,15 +4,11 @@ from typing import TYPE_CHECKING, Any, NewType, Self
 
 from pydantic import BaseModel, ConfigDict
 
-from ..errors import Honk
-from .agent import (
-    Agent,
-    IAgentLogger,
-    SystemMessage,
-    UserMessage,
-)
-from .conversation import Conversation
-from .result import Result
+from goose._internal.agent import Agent, IAgentLogger
+from goose._internal.conversation import Conversation
+from goose._internal.result import Result
+from goose._internal.types.agent import SystemMessage, UserMessage
+from goose.errors import Honk
 
 if TYPE_CHECKING:
     from goose._internal.task import Task
@@ -32,10 +28,11 @@ class NodeState[ResultT: Result](BaseModel):
 
     @property
     def result(self) -> ResultT:
-        if len(self.conversation.result_messages) == 0:
-            raise Honk("Node awaiting response, has no result")
+        for message in reversed(self.conversation.assistant_messages):
+            if isinstance(message, Result):
+                return message
 
-        return self.conversation.result_messages[-1]
+        raise Honk("Node awaiting response, has no result")
 
     def set_context(self, *, context: SystemMessage) -> Self:
         self.conversation.context = context
@@ -48,24 +45,33 @@ class NodeState[ResultT: Result](BaseModel):
         new_hash: int | None = None,
         overwrite: bool = False,
     ) -> Self:
-        if overwrite and len(self.conversation.result_messages) > 0:
-            self.conversation.result_messages[-1] = result
+        if overwrite and len(self.conversation.assistant_messages) > 0:
+            self.conversation.assistant_messages[-1] = result
         else:
-            self.conversation.result_messages.append(result)
+            self.conversation.assistant_messages.append(result)
         if new_hash is not None:
             self.last_hash = new_hash
         return self
 
+    def add_answer(self, *, answer: str) -> Self:
+        self.conversation.assistant_messages.append(answer)
+        return self
+
     def add_user_message(self, *, message: UserMessage) -> Self:
         self.conversation.user_messages.append(message)
         return self
 
     def edit_last_result(self, *, result: ResultT) -> Self:
-        if len(self.conversation.result_messages) == 0:
+        if len(self.conversation.assistant_messages) == 0:
             raise Honk("Node awaiting response, has no result")
 
-        self.conversation.result_messages[-1] = result
-        return self
+        for message_index, message in enumerate(reversed(self.conversation.assistant_messages)):
+            if isinstance(message, Result):
+                index = len(self.conversation.assistant_messages) - message_index - 1
+                self.conversation.assistant_messages[index] = result
+                return self
+
+        raise Honk("Node awaiting response, has no result")
 
     def undo(self) -> Self:
         self.conversation.undo()
@@ -117,7 +123,7 @@ class FlowRun[FlowArgumentsT: FlowArguments]:
             return NodeState[task.result_type](
                 task_name=task.name,
                 index=index,
-                conversation=Conversation[task.result_type](user_messages=[], result_messages=[]),
+                conversation=Conversation[task.result_type](user_messages=[], assistant_messages=[]),
                 last_hash=0,
             )
 

{goose_py-0.9.15 → goose_py-0.10.0}/goose/_internal/task.py

@@ -5,11 +5,10 @@ from typing import Any, overload
 from pydantic import BaseModel
 
 from ..errors import Honk
-from .agent import Agent, AIModel, SystemMessage, UserMessage
-from .conversation import Conversation
-from .result import Result, TextResult
+from .agent import Agent, AIModel
+from .result import Result
 from .state import FlowRun, NodeState, get_current_flow_run
-from .types.agent import AssistantMessage
+from .types.agent import SystemMessage, UserMessage
 
 
 class Task[**P, R: Result]:
@@ -19,12 +18,11 @@ class Task[**P, R: Result]:
         /,
         *,
         retries: int = 0,
-        adapter_model: AIModel = AIModel.GEMINI_FLASH,
+        refinement_model: AIModel = AIModel.GEMINI_FLASH,
     ) -> None:
         self._generator = generator
         self._retries = retries
-        self._adapter_model = adapter_model
-        self._adapter_model = adapter_model
+        self._refinement_model = refinement_model
 
     @property
     def result_type(self) -> type[R]:
@@ -46,6 +44,26 @@ class Task[**P, R: Result]:
         else:
             return state.result
 
+    async def ask(self, *, user_message: UserMessage, context: SystemMessage | None = None, index: int = 0) -> str:
+        flow_run = self.__get_current_flow_run()
+        node_state = flow_run.get(task=self, index=index)
+
+        if len(node_state.conversation.assistant_messages) == 0:
+            raise Honk("Cannot ask about a task that has not been initially generated")
+
+        node_state.add_user_message(message=user_message)
+        answer = await flow_run.agent(
+            messages=node_state.conversation.render(),
+            model=self._refinement_model,
+            task_name=f"ask--{self.name}",
+            system=context.render() if context is not None else None,
+            mode="ask",
+        )
+        node_state.add_answer(answer=answer)
+        flow_run.upsert_node_state(node_state)
+
+        return answer
+
     async def refine(
         self,
         *,
@@ -56,14 +74,20 @@ class Task[**P, R: Result]:
         flow_run = self.__get_current_flow_run()
         node_state = flow_run.get(task=self, index=index)
 
-        if len(node_state.conversation.result_messages) == 0:
+        if len(node_state.conversation.assistant_messages) == 0:
             raise Honk("Cannot refine a task that has not been initially generated")
 
         if context is not None:
             node_state.set_context(context=context)
         node_state.add_user_message(message=user_message)
 
-        result = await self.__adapt(conversation=node_state.conversation, agent=flow_run.agent)
+        result = await flow_run.agent(
+            messages=node_state.conversation.render(),
+            model=self._refinement_model,
+            task_name=f"refine--{self.name}",
+            system=context.render() if context is not None else None,
+            response_model=self.result_type,
+        )
         node_state.add_result(result=result)
         flow_run.upsert_node_state(node_state)
 
@@ -88,28 +112,6 @@ class Task[**P, R: Result]:
         flow_run.upsert_node_state(node_state)
         return result
 
-    async def __adapt(self, *, conversation: Conversation[R], agent: Agent) -> R:
-        messages: list[UserMessage | AssistantMessage] = []
-        for message_index in range(len(conversation.user_messages)):
-            user_message = conversation.user_messages[message_index]
-            result = conversation.result_messages[message_index]
-
-            if isinstance(result, TextResult):
-                assistant_text = result.text
-            else:
-                assistant_text = result.model_dump_json()
-            assistant_message = AssistantMessage(text=assistant_text)
-            messages.append(assistant_message)
-            messages.append(user_message)
-
-        return await agent(
-            messages=messages,
-            model=self._adapter_model,
-            task_name=f"adapt--{self.name}",
-            system=conversation.context,
-            response_model=self.result_type,
-        )
-
     def __hash_task_call(self, *args: P.args, **kwargs: P.kwargs) -> int:
         def update_hash(argument: Any, current_hash: Any = hashlib.sha256()) -> None:
             try:
@@ -148,20 +150,20 @@ class Task[**P, R: Result]:
 def task[**P, R: Result](generator: Callable[P, Awaitable[R]], /) -> Task[P, R]: ...
 @overload
 def task[**P, R: Result](
-    *, retries: int = 0, adapter_model: AIModel = AIModel.GEMINI_FLASH
+    *, retries: int = 0, refinement_model: AIModel = AIModel.GEMINI_FLASH
 ) -> Callable[[Callable[P, Awaitable[R]]], Task[P, R]]: ...
 def task[**P, R: Result](
     generator: Callable[P, Awaitable[R]] | None = None,
     /,
     *,
     retries: int = 0,
-    adapter_model: AIModel = AIModel.GEMINI_FLASH,
+    refinement_model: AIModel = AIModel.GEMINI_FLASH,
 ) -> Task[P, R] | Callable[[Callable[P, Awaitable[R]]], Task[P, R]]:
     if generator is None:
 
         def decorator(fn: Callable[P, Awaitable[R]]) -> Task[P, R]:
-            return Task(fn, retries=retries, adapter_model=adapter_model)
+            return Task(fn, retries=retries, refinement_model=refinement_model)
 
         return decorator
 
-    return Task(generator, retries=retries, adapter_model=adapter_model)
+    return Task(generator, retries=retries, refinement_model=refinement_model)

goose_py-0.10.0/goose/_internal/types/telemetry.py

@@ -0,0 +1,113 @@
+import json
+from datetime import datetime
+from typing import ClassVar, TypedDict
+
+from pydantic import BaseModel, computed_field
+
+from ..types.agent import AIModel, LLMMessage
+
+
+class AgentResponseDump(TypedDict):
+    run_id: str
+    flow_name: str
+    task_name: str
+    model: str
+    system_message: str
+    input_messages: list[str]
+    output_message: str
+    input_cost: float
+    output_cost: float
+    total_cost: float
+    input_tokens: int
+    output_tokens: int
+    start_time: datetime
+    end_time: datetime
+    duration_ms: int
+
+
+class AgentResponse[R: BaseModel | str](BaseModel):
+    INPUT_DOLLARS_PER_MILLION_TOKENS: ClassVar[dict[AIModel, float]] = {
+        AIModel.VERTEX_FLASH_8B: 0.30,
+        AIModel.VERTEX_FLASH: 0.15,
+        AIModel.VERTEX_PRO: 5.00,
+        AIModel.GEMINI_FLASH_8B: 0.30,
+        AIModel.GEMINI_FLASH: 0.15,
+        AIModel.GEMINI_PRO: 5.00,
+    }
+    OUTPUT_DOLLARS_PER_MILLION_TOKENS: ClassVar[dict[AIModel, float]] = {
+        AIModel.VERTEX_FLASH_8B: 0.30,
+        AIModel.VERTEX_FLASH: 0.15,
+        AIModel.VERTEX_PRO: 5.00,
+        AIModel.GEMINI_FLASH_8B: 0.30,
+        AIModel.GEMINI_FLASH: 0.15,
+        AIModel.GEMINI_PRO: 5.00,
+    }
+
+    response: R
+    run_id: str
+    flow_name: str
+    task_name: str
+    model: AIModel
+    system: LLMMessage | None = None
+    input_messages: list[LLMMessage]
+    input_tokens: int
+    output_tokens: int
+    start_time: datetime
+    end_time: datetime
+
+    @computed_field
+    @property
+    def duration_ms(self) -> int:
+        return int((self.end_time - self.start_time).total_seconds() * 1000)
+
+    @computed_field
+    @property
+    def input_cost(self) -> float:
+        return self.INPUT_DOLLARS_PER_MILLION_TOKENS[self.model] * self.input_tokens / 1_000_000
+
+    @computed_field
+    @property
+    def output_cost(self) -> float:
+        return self.OUTPUT_DOLLARS_PER_MILLION_TOKENS[self.model] * self.output_tokens / 1_000_000
+
+    @computed_field
+    @property
+    def total_cost(self) -> float:
+        return self.input_cost + self.output_cost
+
+    def minimized_dump(self) -> AgentResponseDump:
+        if self.system is None:
+            minimized_system_message = ""
+        else:
+            minimized_system_message = self.system
+            for part in minimized_system_message["content"]:
+                if part["type"] == "image_url":
+                    part["image_url"] = "__MEDIA__"
+            minimized_system_message = json.dumps(minimized_system_message)
+
+        minimized_input_messages = [message for message in self.input_messages]
+        for message in minimized_input_messages:
+            for part in message["content"]:
+                if part["type"] == "image_url":
+                    part["image_url"] = "__MEDIA__"
+        minimized_input_messages = [json.dumps(message) for message in minimized_input_messages]
+
+        output_message = self.response.model_dump_json() if isinstance(self.response, BaseModel) else self.response
+
+        return {
+            "run_id": self.run_id,
+            "flow_name": self.flow_name,
+            "task_name": self.task_name,
+            "model": self.model.value,
+            "system_message": minimized_system_message,
+            "input_messages": minimized_input_messages,
+            "output_message": output_message,
+            "input_tokens": self.input_tokens,
+            "output_tokens": self.output_tokens,
+            "input_cost": self.input_cost,
+            "output_cost": self.output_cost,
+            "total_cost": self.total_cost,
+            "start_time": self.start_time,
+            "end_time": self.end_time,
+            "duration_ms": self.duration_ms,
+        }

{goose_py-0.9.15 → goose_py-0.10.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "goose-py"
-version = "0.9.15"
+version = "0.10.0"
 description = "A tool for AI workflows based on human-computer collaboration and structured output."
 readme = "README.md"
 authors = [
@@ -24,6 +24,7 @@ packages = ["goose"]
 
 [dependency-groups]
 dev = [
+    "ipykernel>=6.29.5",
     "pyright>=1.1.393",
     "pytest>=8.3.4",
     "pytest-asyncio>=0.25.3",