goose-py 0.5.1__tar.gz → 0.6.0__tar.gz

{goose_py-0.5.1 → goose_py-0.6.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: goose-py
-Version: 0.5.1
+Version: 0.6.0
 Summary: A tool for AI workflows based on human-computer collaboration and structured output.
 Home-page: https://github.com/chelle-ai/goose
 Keywords: ai,yaml,configuration,llm

goose_py-0.6.0/goose/__init__.py

@@ -0,0 +1,5 @@
+from goose._agent import Agent, AgentResponse, IAgentLogger
+from goose._flow import flow
+from goose._result import Result, TextResult
+from goose._store import IFlowRunStore
+from goose._task import task

goose_py-0.5.1/goose/agent.py → goose_py-0.6.0/goose/_agent.py

@@ -1,100 +1,13 @@
 import json
 import logging
 from datetime import datetime
-from enum import StrEnum
-from typing import Any, ClassVar, Literal, NotRequired, Protocol, TypedDict
+from typing import Any, ClassVar, Protocol, TypedDict
 
 from litellm import acompletion
 from pydantic import BaseModel, computed_field
-from goose.result import Result, TextResult
 
-
-class GeminiModel(StrEnum):
-    PRO = "vertex_ai/gemini-1.5-pro"
-    FLASH = "vertex_ai/gemini-1.5-flash"
-    FLASH_8B = "vertex_ai/gemini-1.5-flash-8b"
-
-
-class UserMediaContentType(StrEnum):
-    # images
-    JPEG = "image/jpeg"
-    PNG = "image/png"
-    WEBP = "image/webp"
-
-    # audio
-    MP3 = "audio/mp3"
-    WAV = "audio/wav"
-
-    # files
-    PDF = "application/pdf"
-
-
-class LLMTextMessagePart(TypedDict):
-    type: Literal["text"]
-    text: str
-
-
-class LLMMediaMessagePart(TypedDict):
-    type: Literal["image_url"]
-    image_url: str
-
-
-class CacheControl(TypedDict):
-    type: Literal["ephemeral"]
-
-
-class LLMMessage(TypedDict):
-    role: Literal["user", "assistant", "system"]
-    content: list[LLMTextMessagePart | LLMMediaMessagePart]
-    cache_control: NotRequired[CacheControl]
-
-
-class TextMessagePart(BaseModel):
-    text: str
-
-    def render(self) -> LLMTextMessagePart:
-        return {"type": "text", "text": self.text}
-
-
-class MediaMessagePart(BaseModel):
-    content_type: UserMediaContentType
-    content: str
-
-    def render(self) -> LLMMediaMessagePart:
-        return {
-            "type": "image_url",
-            "image_url": f"data:{self.content_type};base64,{self.content}",
-        }
-
-
-class UserMessage(BaseModel):
-    parts: list[TextMessagePart | MediaMessagePart]
-
-    def render(self) -> LLMMessage:
-        content: LLMMessage = {
-            "role": "user",
-            "content": [part.render() for part in self.parts],
-        }
-        if any(isinstance(part, MediaMessagePart) for part in self.parts):
-            content["cache_control"] = {"type": "ephemeral"}
-        return content
-
-
-class AssistantMessage(BaseModel):
-    text: str
-
-    def render(self) -> LLMMessage:
-        return {"role": "assistant", "content": [{"type": "text", "text": self.text}]}
-
-
-class SystemMessage(BaseModel):
-    parts: list[TextMessagePart | MediaMessagePart]
-
-    def render(self) -> LLMMessage:
-        return {
-            "role": "system",
-            "content": [part.render() for part in self.parts],
-        }
+from goose._result import Result, TextResult
+from goose.types.agent import AssistantMessage, GeminiModel, SystemMessage, UserMessage
 
 
 class AgentResponseDump(TypedDict):

goose_py-0.6.0/goose/_conversation.py

@@ -0,0 +1,36 @@
+from pydantic import BaseModel
+
+from goose._result import Result
+from goose.types.agent import AssistantMessage, LLMMessage, SystemMessage, UserMessage
+
+
+class Conversation[R: Result](BaseModel):
+    user_messages: list[UserMessage]
+    result_messages: list[R]
+    context: SystemMessage | None = None
+
+    @property
+    def awaiting_response(self) -> bool:
+        return len(self.user_messages) == len(self.result_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)):
+            messages.append(
+                AssistantMessage(
+                    text=self.result_messages[message_index].model_dump_json()
+                ).render()
+            )
+            messages.append(self.user_messages[message_index].render())
+
+        if len(self.result_messages) > len(self.user_messages):
+            messages.append(
+                AssistantMessage(
+                    text=self.result_messages[-1].model_dump_json()
+                ).render()
+            )
+
+        return messages

goose_py-0.6.0/goose/_flow.py

@@ -0,0 +1,106 @@
+from contextlib import asynccontextmanager
+from types import CodeType
+from typing import AsyncIterator, Awaitable, Callable, Protocol, overload
+
+from goose._agent import Agent, IAgentLogger
+from goose._conversation import Conversation
+from goose._result import Result
+from goose._state import FlowRun, get_current_flow_run, set_current_flow_run
+from goose._store import IFlowRunStore, InMemoryFlowRunStore
+from goose.errors import Honk
+
+
+class IAdapter[ResultT: Result](Protocol):
+    __code__: CodeType
+
+    async def __call__(
+        self, *, conversation: Conversation[ResultT], agent: Agent
+    ) -> ResultT: ...
+
+
+class Flow[**P]:
+    def __init__(
+        self,
+        fn: Callable[P, Awaitable[None]],
+        /,
+        *,
+        name: str | None = None,
+        store: IFlowRunStore | None = None,
+        agent_logger: IAgentLogger | None = None,
+    ) -> None:
+        self._fn = fn
+        self._name = name
+        self._agent_logger = agent_logger
+        self._store = store or InMemoryFlowRunStore(flow_name=self.name)
+
+    @property
+    def name(self) -> str:
+        return self._name or self._fn.__name__
+
+    @property
+    def current_run(self) -> FlowRun:
+        run = get_current_flow_run()
+        if run is None:
+            raise Honk("No current flow run")
+        return run
+
+    @asynccontextmanager
+    async def start_run(self, *, run_id: str) -> AsyncIterator[FlowRun]:
+        existing_run = await self._store.get(run_id=run_id)
+        if existing_run is None:
+            run = FlowRun()
+        else:
+            run = existing_run
+
+        old_run = get_current_flow_run()
+        set_current_flow_run(run)
+
+        run.start(flow_name=self.name, run_id=run_id, agent_logger=self._agent_logger)
+        yield run
+        await self._store.save(run=run)
+        run.end()
+
+        set_current_flow_run(old_run)
+
+    async def generate(self, *args: P.args, **kwargs: P.kwargs) -> None:
+        flow_run = get_current_flow_run()
+        if flow_run is None:
+            raise Honk("No current flow run")
+
+        flow_run.set_flow_inputs(*args, **kwargs)
+        await self._fn(*args, **kwargs)
+
+    async def regenerate(self) -> None:
+        flow_run = get_current_flow_run()
+        if flow_run is None:
+            raise Honk("No current flow run")
+
+        flow_args, flow_kwargs = flow_run.flow_inputs
+        await self._fn(*flow_args, **flow_kwargs)
+
+
+@overload
+def flow[**P](fn: Callable[P, Awaitable[None]], /) -> Flow[P]: ...
+@overload
+def flow[**P](
+    *,
+    name: str | None = None,
+    store: IFlowRunStore | None = None,
+    agent_logger: IAgentLogger | None = None,
+) -> Callable[[Callable[P, Awaitable[None]]], Flow[P]]: ...
+def flow[**P](
+    fn: Callable[P, Awaitable[None]] | None = None,
+    /,
+    *,
+    name: str | None = None,
+    store: IFlowRunStore | None = None,
+    agent_logger: IAgentLogger | None = None,
+) -> Flow[P] | Callable[[Callable[P, Awaitable[None]]], Flow[P]]:
+    if fn is None:
+
+        def decorator(fn: Callable[P, Awaitable[None]]) -> Flow[P]:
+            return Flow(fn, name=name, store=store, agent_logger=agent_logger)
+
+        return decorator
+
+    return Flow(fn, name=name, store=store, agent_logger=agent_logger)

goose_py-0.6.0/goose/_state.py

@@ -0,0 +1,190 @@
+from contextvars import ContextVar
+from dataclasses import dataclass
+from typing import TYPE_CHECKING, Any, Self
+
+from pydantic import BaseModel
+
+from goose._agent import (
+    Agent,
+    IAgentLogger,
+    SystemMessage,
+    UserMessage,
+)
+from goose._conversation import Conversation
+from goose._result import Result
+from goose.errors import Honk
+
+if TYPE_CHECKING:
+    from goose._task import Task
+
+
+@dataclass
+class FlowRunState:
+    node_states: dict[tuple[str, int], str]
+    flow_args: tuple[Any, ...]
+    flow_kwargs: dict[str, Any]
+
+
+class NodeState[ResultT: Result](BaseModel):
+    task_name: str
+    index: int
+    conversation: Conversation[ResultT]
+    last_hash: int
+
+    @property
+    def result(self) -> ResultT:
+        if len(self.conversation.result_messages) == 0:
+            raise Honk("Node awaiting response, has no result")
+
+        return self.conversation.result_messages[-1]
+
+    def set_context(self, *, context: SystemMessage) -> Self:
+        self.conversation.context = context
+        return self
+
+    def add_result(
+        self,
+        *,
+        result: ResultT,
+        new_hash: int | None = None,
+        overwrite: bool = False,
+    ) -> Self:
+        if overwrite and len(self.conversation.result_messages) > 0:
+            self.conversation.result_messages[-1] = result
+        else:
+            self.conversation.result_messages.append(result)
+        if new_hash is not None:
+            self.last_hash = new_hash
+        return self
+
+    def add_user_message(self, *, message: UserMessage) -> Self:
+        self.conversation.user_messages.append(message)
+        return self
+
+
+class FlowRun:
+    def __init__(self) -> None:
+        self._node_states: dict[tuple[str, int], str] = {}
+        self._last_requested_indices: dict[str, int] = {}
+        self._flow_name = ""
+        self._id = ""
+        self._agent: Agent | None = None
+        self._flow_args: tuple[Any, ...] | None = None
+        self._flow_kwargs: dict[str, Any] | None = None
+
+    @property
+    def flow_name(self) -> str:
+        return self._flow_name
+
+    @property
+    def id(self) -> str:
+        return self._id
+
+    @property
+    def agent(self) -> Agent:
+        if self._agent is None:
+            raise Honk("Agent is only accessible once a run is started")
+        return self._agent
+
+    @property
+    def flow_inputs(self) -> tuple[tuple[Any, ...], dict[str, Any]]:
+        if self._flow_args is None or self._flow_kwargs is None:
+            raise Honk("This Flow run has not been executed before")
+
+        return self._flow_args, self._flow_kwargs
+
+    def get_all[R: Result](self, *, task: "Task[Any, R]") -> list[NodeState[R]]:
+        matching_nodes: list[NodeState[R]] = []
+        for key, node_state in self._node_states.items():
+            if key[0] == task.name:
+                matching_nodes.append(
+                    NodeState[task.result_type].model_validate_json(node_state)
+                )
+        return sorted(matching_nodes, key=lambda node: node.index)
+
+    def get[R: Result](self, *, task: "Task[Any, R]", index: int = 0) -> NodeState[R]:
+        if (
+            existing_node_state := self._node_states.get((task.name, index))
+        ) is not None:
+            return NodeState[task.result_type].model_validate_json(existing_node_state)
+        else:
+            return NodeState[task.result_type](
+                task_name=task.name,
+                index=index,
+                conversation=Conversation[task.result_type](
+                    user_messages=[], result_messages=[]
+                ),
+                last_hash=0,
+            )
+
+    def set_flow_inputs(self, *args: Any, **kwargs: Any) -> None:
+        self._flow_args = args
+        self._flow_kwargs = kwargs
+
+    def add_node_state(self, node_state: NodeState[Any], /) -> None:
+        key = (node_state.task_name, node_state.index)
+        self._node_states[key] = node_state.model_dump_json()
+
+    def get_next[R: Result](self, *, task: "Task[Any, R]") -> NodeState[R]:
+        if task.name not in self._last_requested_indices:
+            self._last_requested_indices[task.name] = 0
+        else:
+            self._last_requested_indices[task.name] += 1
+
+        return self.get(task=task, index=self._last_requested_indices[task.name])
+
+    def start(
+        self,
+        *,
+        flow_name: str,
+        run_id: str,
+        agent_logger: IAgentLogger | None = None,
+    ) -> None:
+        self._last_requested_indices = {}
+        self._flow_name = flow_name
+        self._id = run_id
+        self._agent = Agent(
+            flow_name=self.flow_name, run_id=self.id, logger=agent_logger
+        )
+
+    def end(self) -> None:
+        self._last_requested_indices = {}
+        self._flow_name = ""
+        self._id = ""
+        self._agent = None
+
+    def clear_node(self, *, task: "Task[Any, Result]", index: int) -> None:
+        key = (task.name, index)
+        if key in self._node_states:
+            del self._node_states[key]
+
+    def dump(self) -> FlowRunState:
+        flow_args, flow_kwargs = self.flow_inputs
+
+        return FlowRunState(
+            node_states=self._node_states,
+            flow_args=flow_args,
+            flow_kwargs=flow_kwargs,
+        )
+
+    @classmethod
+    def load(cls, flow_run_state: FlowRunState, /) -> Self:
+        flow_run = cls()
+        flow_run._node_states = flow_run_state.node_states
+        flow_run._flow_args = flow_run_state.flow_args
+        flow_run._flow_kwargs = flow_run_state.flow_kwargs
+
+        return flow_run
+
+
+_current_flow_run: ContextVar[FlowRun | None] = ContextVar(
+    "current_flow_run", default=None
+)
+
+
+def get_current_flow_run() -> FlowRun | None:
+    return _current_flow_run.get()
+
+
+def set_current_flow_run(flow_run: FlowRun | None) -> None:
+    _current_flow_run.set(flow_run)

goose_py-0.5.1/goose/store.py → goose_py-0.6.0/goose/_store.py

@@ -1,9 +1,9 @@
 from __future__ import annotations
 
-from typing import TYPE_CHECKING, Protocol
+from typing import Protocol
 
-if TYPE_CHECKING:
-    from goose.flow import FlowRun
+from goose._flow import FlowRun
+from goose._state import FlowRunState
 
 
 class IFlowRunStore(Protocol):
@@ -16,13 +16,15 @@ class IFlowRunStore(Protocol):
 class InMemoryFlowRunStore(IFlowRunStore):
     def __init__(self, *, flow_name: str) -> None:
         self._flow_name = flow_name
-        self._runs: dict[str, FlowRun] = {}
+        self._runs: dict[str, FlowRunState] = {}
 
     async def get(self, *, run_id: str) -> FlowRun | None:
-        return self._runs.get(run_id)
+        state = self._runs.get(run_id)
+        if state is not None:
+            return FlowRun.load(state)
 
     async def save(self, *, run: FlowRun) -> None:
-        self._runs[run.id] = run
+        self._runs[run.id] = run.dump()
 
     async def delete(self, *, run_id: str) -> None:
         self._runs.pop(run_id, None)

goose_py-0.6.0/goose/_task.py

@@ -0,0 +1,136 @@
+from typing import Awaitable, Callable, overload
+
+from goose._agent import Agent, GeminiModel, SystemMessage, UserMessage
+from goose._conversation import Conversation
+from goose._result import Result, TextResult
+from goose._state import FlowRun, NodeState, get_current_flow_run
+from goose.errors import Honk
+from goose.types.agent import AssistantMessage
+
+
+class Task[**P, R: Result]:
+    def __init__(
+        self,
+        generator: Callable[P, Awaitable[R]],
+        /,
+        *,
+        retries: int = 0,
+        adapter_model: GeminiModel = GeminiModel.FLASH,
+    ) -> None:
+        self._generator = generator
+        self._retries = retries
+        self._adapter_model = adapter_model
+        self._adapter_model = adapter_model
+
+    @property
+    def result_type(self) -> type[R]:
+        result_type = self._generator.__annotations__.get("return")
+        if result_type is None:
+            raise Honk(f"Task {self.name} has no return type annotation")
+        return result_type
+
+    @property
+    def name(self) -> str:
+        return self._generator.__name__
+
+    async def generate(
+        self, state: NodeState[R], *args: P.args, **kwargs: P.kwargs
+    ) -> R:
+        state_hash = self.__hash_task_call(*args, **kwargs)
+        if state_hash != state.last_hash:
+            result = await self._generator(*args, **kwargs)
+            state.add_result(result=result, new_hash=state_hash, overwrite=True)
+            return result
+        else:
+            return state.result
+
+    async def jam(
+        self,
+        *,
+        user_message: UserMessage,
+        context: SystemMessage | None = None,
+        index: int = 0,
+    ) -> R:
+        flow_run = self.__get_current_flow_run()
+        node_state = flow_run.get(task=self, index=index)
+
+        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
+        )
+        node_state.add_result(result=result)
+        flow_run.add_node_state(node_state)
+
+        return result
+
+    async def __call__(self, *args: P.args, **kwargs: P.kwargs) -> R:
+        flow_run = self.__get_current_flow_run()
+        node_state = flow_run.get_next(task=self)
+        result = await self.generate(node_state, *args, **kwargs)
+        flow_run.add_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:
+        try:
+            to_hash = str(
+                tuple(args)
+                + tuple(kwargs.values())
+                + (self._generator.__code__, self._adapter_model)
+            )
+            return hash(to_hash)
+        except TypeError:
+            raise Honk(f"Unhashable argument to task {self.name}: {args} {kwargs}")
+
+    def __get_current_flow_run(self) -> FlowRun:
+        run = get_current_flow_run()
+        if run is None:
+            raise Honk("No current flow run")
+        return run
+
+
+@overload
+def task[**P, R: Result](generator: Callable[P, Awaitable[R]], /) -> Task[P, R]: ...
+@overload
+def task[**P, R: Result](
+    *, retries: int = 0, adapter_model: GeminiModel = GeminiModel.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: GeminiModel = GeminiModel.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 decorator
+
+    return Task(generator, retries=retries, adapter_model=adapter_model)

goose_py-0.6.0/goose/types/agent.py

@@ -0,0 +1,92 @@
+from enum import StrEnum
+from typing import Literal, NotRequired, TypedDict
+
+from pydantic import BaseModel
+
+
+class GeminiModel(StrEnum):
+    PRO = "vertex_ai/gemini-1.5-pro"
+    FLASH = "vertex_ai/gemini-1.5-flash"
+    FLASH_8B = "vertex_ai/gemini-1.5-flash-8b"
+
+
+class UserMediaContentType(StrEnum):
+    # images
+    JPEG = "image/jpeg"
+    PNG = "image/png"
+    WEBP = "image/webp"
+
+    # audio
+    MP3 = "audio/mp3"
+    WAV = "audio/wav"
+
+    # files
+    PDF = "application/pdf"
+
+
+class LLMTextMessagePart(TypedDict):
+    type: Literal["text"]
+    text: str
+
+
+class LLMMediaMessagePart(TypedDict):
+    type: Literal["image_url"]
+    image_url: str
+
+
+class CacheControl(TypedDict):
+    type: Literal["ephemeral"]
+
+
+class LLMMessage(TypedDict):
+    role: Literal["user", "assistant", "system"]
+    content: list[LLMTextMessagePart | LLMMediaMessagePart]
+    cache_control: NotRequired[CacheControl]
+
+
+class TextMessagePart(BaseModel):
+    text: str
+
+    def render(self) -> LLMTextMessagePart:
+        return {"type": "text", "text": self.text}
+
+
+class MediaMessagePart(BaseModel):
+    content_type: UserMediaContentType
+    content: str
+
+    def render(self) -> LLMMediaMessagePart:
+        return {
+            "type": "image_url",
+            "image_url": f"data:{self.content_type};base64,{self.content}",
+        }
+
+
+class UserMessage(BaseModel):
+    parts: list[TextMessagePart | MediaMessagePart]
+
+    def render(self) -> LLMMessage:
+        content: LLMMessage = {
+            "role": "user",
+            "content": [part.render() for part in self.parts],
+        }
+        if any(isinstance(part, MediaMessagePart) for part in self.parts):
+            content["cache_control"] = {"type": "ephemeral"}
+        return content
+
+
+class AssistantMessage(BaseModel):
+    text: str
+
+    def render(self) -> LLMMessage:
+        return {"role": "assistant", "content": [{"type": "text", "text": self.text}]}
+
+
+class SystemMessage(BaseModel):
+    parts: list[TextMessagePart | MediaMessagePart]
+
+    def render(self) -> LLMMessage:
+        return {
+            "role": "system",
+            "content": [part.render() for part in self.parts],
+        }

{goose_py-0.5.1 → goose_py-0.6.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "goose-py"
-version = "0.5.1"
+version = "0.6.0"
 description = "A tool for AI workflows based on human-computer collaboration and structured output."
 authors = [
     "Nash Taylor <nash@chelle.ai>",
@@ -53,6 +53,10 @@ reportImportCycles = false
 reportUnknownMemberType = false
 reportUnknownVariableType = false
 stubPath = ".stubs"
+exclude = [
+    "goose/__init__.py",
+    ".venv",
+]
 
 [tool.pytest.ini_options]
 filterwarnings = [

goose_py-0.5.1/goose/flow.py

@@ -1,458 +0,0 @@
-import json
-from contextlib import asynccontextmanager
-from contextvars import ContextVar
-from types import CodeType
-from typing import (
-    Any,
-    AsyncIterator,
-    Awaitable,
-    Callable,
-    NewType,
-    Protocol,
-    Self,
-    overload,
-)
-
-from pydantic import BaseModel
-
-from goose.agent import (
-    Agent,
-    AssistantMessage,
-    GeminiModel,
-    IAgentLogger,
-    LLMMessage,
-    SystemMessage,
-    UserMessage,
-)
-from goose.errors import Honk
-from goose.result import Result, TextResult
-from goose.store import IFlowRunStore, InMemoryFlowRunStore
-
-SerializedFlowRun = NewType("SerializedFlowRun", str)
-
-
-class Conversation[R: Result](BaseModel):
-    user_messages: list[UserMessage]
-    result_messages: list[R]
-    context: SystemMessage | None = None
-
-    @property
-    def awaiting_response(self) -> bool:
-        return len(self.user_messages) == len(self.result_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)):
-            messages.append(
-                AssistantMessage(
-                    text=self.result_messages[message_index].model_dump_json()
-                ).render()
-            )
-            messages.append(self.user_messages[message_index].render())
-
-        if len(self.result_messages) > len(self.user_messages):
-            messages.append(
-                AssistantMessage(
-                    text=self.result_messages[-1].model_dump_json()
-                ).render()
-            )
-
-        return messages
-
-
-class IAdapter[ResultT: Result](Protocol):
-    __code__: CodeType
-
-    async def __call__(
-        self, *, conversation: Conversation[ResultT], agent: Agent
-    ) -> ResultT: ...
-
-
-class NodeState[ResultT: Result](BaseModel):
-    task_name: str
-    index: int
-    conversation: Conversation[ResultT]
-    last_hash: int
-
-    @property
-    def result(self) -> ResultT:
-        if len(self.conversation.result_messages) == 0:
-            raise Honk("Node awaiting response, has no result")
-
-        return self.conversation.result_messages[-1]
-
-    def set_context(self, *, context: SystemMessage) -> Self:
-        self.conversation.context = context
-        return self
-
-    def add_result(
-        self,
-        *,
-        result: ResultT,
-        new_hash: int | None = None,
-        overwrite: bool = False,
-    ) -> Self:
-        if overwrite and len(self.conversation.result_messages) > 0:
-            self.conversation.result_messages[-1] = result
-        else:
-            self.conversation.result_messages.append(result)
-        if new_hash is not None:
-            self.last_hash = new_hash
-        return self
-
-    def add_user_message(self, *, message: UserMessage) -> Self:
-        self.conversation.user_messages.append(message)
-        return self
-
-
-class FlowRun:
-    def __init__(self) -> None:
-        self._node_states: dict[tuple[str, int], str] = {}
-        self._last_requested_indices: dict[str, int] = {}
-        self._flow_name = ""
-        self._id = ""
-        self._agent: Agent | None = None
-        self._flow_args: tuple[Any, ...] | None = None
-        self._flow_kwargs: dict[str, Any] | None = None
-
-    @property
-    def flow_name(self) -> str:
-        return self._flow_name
-
-    @property
-    def id(self) -> str:
-        return self._id
-
-    @property
-    def agent(self) -> Agent:
-        if self._agent is None:
-            raise Honk("Agent is only accessible once a run is started")
-        return self._agent
-
-    @property
-    def flow_inputs(self) -> tuple[tuple[Any, ...], dict[str, Any]]:
-        if self._flow_args is None or self._flow_kwargs is None:
-            raise Honk("This Flow run has not been executed before")
-
-        return self._flow_args, self._flow_kwargs
-
-    def get_all[R: Result](self, *, task: "Task[Any, R]") -> list[NodeState[R]]:
-        matching_nodes: list[NodeState[R]] = []
-        for key, node_state in self._node_states.items():
-            if key[0] == task.name:
-                matching_nodes.append(
-                    NodeState[task.result_type].model_validate_json(node_state)
-                )
-        return sorted(matching_nodes, key=lambda node: node.index)
-
-    def get[R: Result](self, *, task: "Task[Any, R]", index: int = 0) -> NodeState[R]:
-        if (
-            existing_node_state := self._node_states.get((task.name, index))
-        ) is not None:
-            return NodeState[task.result_type].model_validate_json(existing_node_state)
-        else:
-            return NodeState[task.result_type](
-                task_name=task.name,
-                index=index,
-                conversation=Conversation[task.result_type](
-                    user_messages=[], result_messages=[]
-                ),
-                last_hash=0,
-            )
-
-    def set_flow_inputs(self, *args: Any, **kwargs: Any) -> None:
-        self._flow_args = args
-        self._flow_kwargs = kwargs
-
-    def add_node_state(self, node_state: NodeState[Any], /) -> None:
-        key = (node_state.task_name, node_state.index)
-        self._node_states[key] = node_state.model_dump_json()
-
-    def get_next[R: Result](self, *, task: "Task[Any, R]") -> NodeState[R]:
-        if task.name not in self._last_requested_indices:
-            self._last_requested_indices[task.name] = 0
-        else:
-            self._last_requested_indices[task.name] += 1
-
-        return self.get(task=task, index=self._last_requested_indices[task.name])
-
-    def start(
-        self,
-        *,
-        flow_name: str,
-        run_id: str,
-        agent_logger: IAgentLogger | None = None,
-    ) -> None:
-        self._last_requested_indices = {}
-        self._flow_name = flow_name
-        self._id = run_id
-        self._agent = Agent(
-            flow_name=self.flow_name, run_id=self.id, logger=agent_logger
-        )
-
-    def end(self) -> None:
-        self._last_requested_indices = {}
-        self._flow_name = ""
-        self._id = ""
-        self._agent = None
-
-    def clear_node(self, *, task: "Task[Any, Result]", index: int) -> None:
-        key = (task.name, index)
-        if key in self._node_states:
-            del self._node_states[key]
-
-    def dump(self) -> SerializedFlowRun:
-        flow_args, flow_kwargs = self.flow_inputs
-
-        return SerializedFlowRun(
-            json.dumps(
-                {
-                    "node_states": {
-                        ":".join([task_name, str(index)]): value
-                        for (task_name, index), value in self._node_states.items()
-                    },
-                    "flow_args": list(flow_args),
-                    "flow_kwargs": flow_kwargs,
-                }
-            )
-        )
-
-    @classmethod
-    def load(cls, serialized_flow_run: SerializedFlowRun, /) -> Self:
-        flow_run = cls()
-        run = json.loads(serialized_flow_run)
-
-        new_node_states: dict[tuple[str, int], str] = {}
-        for key, node_state in run["node_states"].items():
-            task_name, index = tuple(key.split(":"))
-            new_node_states[(task_name, int(index))] = node_state
-        flow_run._node_states = new_node_states
-
-        flow_run._flow_args = tuple(run["flow_args"])
-        flow_run._flow_kwargs = run["flow_kwargs"]
-
-        return flow_run
-
-
-_current_flow_run: ContextVar[FlowRun | None] = ContextVar(
-    "current_flow_run", default=None
-)
-
-
-class Flow[**P]:
-    def __init__(
-        self,
-        fn: Callable[P, Awaitable[None]],
-        /,
-        *,
-        name: str | None = None,
-        store: IFlowRunStore | None = None,
-        agent_logger: IAgentLogger | None = None,
-    ) -> None:
-        self._fn = fn
-        self._name = name
-        self._agent_logger = agent_logger
-        self._store = store or InMemoryFlowRunStore(flow_name=self.name)
-
-    @property
-    def name(self) -> str:
-        return self._name or self._fn.__name__
-
-    @property
-    def current_run(self) -> FlowRun:
-        run = _current_flow_run.get()
-        if run is None:
-            raise Honk("No current flow run")
-        return run
-
-    @asynccontextmanager
-    async def start_run(self, *, run_id: str) -> AsyncIterator[FlowRun]:
-        existing_run = await self._store.get(run_id=run_id)
-        if existing_run is None:
-            run = FlowRun()
-        else:
-            run = existing_run
-
-        old_run = _current_flow_run.get()
-        _current_flow_run.set(run)
-
-        run.start(flow_name=self.name, run_id=run_id, agent_logger=self._agent_logger)
-        yield run
-        await self._store.save(run=run)
-        run.end()
-
-        _current_flow_run.set(old_run)
-
-    async def generate(self, *args: P.args, **kwargs: P.kwargs) -> None:
-        flow_run = _current_flow_run.get()
-        if flow_run is None:
-            raise Honk("No current flow run")
-
-        flow_run.set_flow_inputs(*args, **kwargs)
-        await self._fn(*args, **kwargs)
-
-    async def regenerate(self) -> None:
-        flow_run = _current_flow_run.get()
-        if flow_run is None:
-            raise Honk("No current flow run")
-
-        flow_args, flow_kwargs = flow_run.flow_inputs
-        await self._fn(*flow_args, **flow_kwargs)
-
-
-class Task[**P, R: Result]:
-    def __init__(
-        self,
-        generator: Callable[P, Awaitable[R]],
-        /,
-        *,
-        retries: int = 0,
-        adapter_model: GeminiModel = GeminiModel.FLASH,
-    ) -> None:
-        self._generator = generator
-        self._retries = retries
-        self._adapter_model = adapter_model
-        self._adapter_model = adapter_model
-
-    @property
-    def result_type(self) -> type[R]:
-        result_type = self._generator.__annotations__.get("return")
-        if result_type is None:
-            raise Honk(f"Task {self.name} has no return type annotation")
-        return result_type
-
-    @property
-    def name(self) -> str:
-        return self._generator.__name__
-
-    async def generate(
-        self, state: NodeState[R], *args: P.args, **kwargs: P.kwargs
-    ) -> R:
-        state_hash = self.__hash_task_call(*args, **kwargs)
-        if state_hash != state.last_hash:
-            result = await self._generator(*args, **kwargs)
-            state.add_result(result=result, new_hash=state_hash, overwrite=True)
-            return result
-        else:
-            return state.result
-
-    async def jam(
-        self,
-        *,
-        user_message: UserMessage,
-        context: SystemMessage | None = None,
-        index: int = 0,
-    ) -> R:
-        flow_run = self.__get_current_flow_run()
-        node_state = flow_run.get(task=self, index=index)
-
-        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
-        )
-        node_state.add_result(result=result)
-        flow_run.add_node_state(node_state)
-
-        return result
-
-    async def __call__(self, *args: P.args, **kwargs: P.kwargs) -> R:
-        flow_run = self.__get_current_flow_run()
-        node_state = flow_run.get_next(task=self)
-        result = await self.generate(node_state, *args, **kwargs)
-        flow_run.add_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:
-        try:
-            to_hash = str(
-                tuple(args)
-                + tuple(kwargs.values())
-                + (self._generator.__code__, self._adapter_model)
-            )
-            return hash(to_hash)
-        except TypeError:
-            raise Honk(f"Unhashable argument to task {self.name}: {args} {kwargs}")
-
-    def __get_current_flow_run(self) -> FlowRun:
-        run = _current_flow_run.get()
-        if run is None:
-            raise Honk("No current flow run")
-        return run
-
-
-@overload
-def task[**P, R: Result](generator: Callable[P, Awaitable[R]], /) -> Task[P, R]: ...
-@overload
-def task[**P, R: Result](
-    *, retries: int = 0, adapter_model: GeminiModel = GeminiModel.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: GeminiModel = GeminiModel.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 decorator
-
-    return Task(generator, retries=retries, adapter_model=adapter_model)
-
-
-@overload
-def flow[**P](fn: Callable[P, Awaitable[None]], /) -> Flow[P]: ...
-@overload
-def flow[**P](
-    *,
-    name: str | None = None,
-    store: IFlowRunStore | None = None,
-    agent_logger: IAgentLogger | None = None,
-) -> Callable[[Callable[P, Awaitable[None]]], Flow[P]]: ...
-def flow[**P](
-    fn: Callable[P, Awaitable[None]] | None = None,
-    /,
-    *,
-    name: str | None = None,
-    store: IFlowRunStore | None = None,
-    agent_logger: IAgentLogger | None = None,
-) -> Flow[P] | Callable[[Callable[P, Awaitable[None]]], Flow[P]]:
-    if fn is None:
-
-        def decorator(fn: Callable[P, Awaitable[None]]) -> Flow[P]:
-            return Flow(fn, name=name, store=store, agent_logger=agent_logger)
-
-        return decorator
-
-    return Flow(fn, name=name, store=store, agent_logger=agent_logger)