intentkit 1.2.11__tar.gz → 1.2.13__tar.gz

{intentkit-1.2.11 → intentkit-1.2.13}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: intentkit
-Version: 1.2.11
+Version: 1.2.13
 Summary: Intent-based AI Agent Platform - Core Package
 Project-URL: Homepage, https://github.com/crestalnetwork/intentkit
 Project-URL: Repository, https://github.com/crestalnetwork/intentkit
@@ -55,20 +55,20 @@ Requires-Dist: fastapi>=0.115.8
 Requires-Dist: filetype<2.0.0,>=1.2.0
 Requires-Dist: httpx>=0.28.1
 Requires-Dist: jsonref>=1.1.0
-Requires-Dist: langchain-anthropic>=0.3.12
-Requires-Dist: langchain-community>=0.3.19
-Requires-Dist: langchain-core>=0.3.43
-Requires-Dist: langchain-deepseek>=0.1.4
-Requires-Dist: langchain-google-genai>=3.2.0
-Requires-Dist: langchain-openai>=0.3.8
-Requires-Dist: langchain-openrouter>=0.2.1
-Requires-Dist: langchain-text-splitters>=0.3.8
-Requires-Dist: langchain-xai>=0.2.1
-Requires-Dist: langchain>=0.3.25
-Requires-Dist: langgraph-checkpoint-postgres>=2.0.16
-Requires-Dist: langgraph-checkpoint>=2.0.18
-Requires-Dist: langgraph-prebuilt>=0.6.1
-Requires-Dist: langgraph>=0.6.1
+Requires-Dist: langchain-anthropic<2,>=1.4
+Requires-Dist: langchain-community<1,>=0.4
+Requires-Dist: langchain-core<2,>=1.4
+Requires-Dist: langchain-deepseek<2,>=1.0
+Requires-Dist: langchain-google-genai<5,>=4.2
+Requires-Dist: langchain-openai<2,>=1.2
+Requires-Dist: langchain-openrouter<1,>=0.2.3
+Requires-Dist: langchain-text-splitters<2,>=1.1
+Requires-Dist: langchain-xai<2,>=1.2
+Requires-Dist: langchain<2,>=1.3
+Requires-Dist: langgraph-checkpoint-postgres<4,>=3.1
+Requires-Dist: langgraph-checkpoint<5,>=4.1
+Requires-Dist: langgraph-prebuilt<2,>=1.1
+Requires-Dist: langgraph<2,>=1.2
 Requires-Dist: mcp>=1.0.0
 Requires-Dist: mypy-boto3-s3<2.0.0,>=1.37.24
 Requires-Dist: openai>=1.59.6

{intentkit-1.2.11 → intentkit-1.2.13}/intentkit/__init__.py

@@ -3,7 +3,7 @@
 A powerful platform for building AI agents with blockchain and cryptocurrency capabilities.
 """
 
-__version__ = "1.2.11"
+__version__ = "1.2.13"
 __author__ = "Ruihua"
 __email__ = "ruihua@crestal.network"
 

intentkit-1.2.13/intentkit/abstracts/graph.py

@@ -0,0 +1,108 @@
+from collections.abc import Callable, Sequence
+from enum import Enum
+from typing import Annotated, Any, NotRequired
+
+from langchain.agents import AgentState as BaseAgentState
+from langchain_core.messages import AnyMessage, RemoveMessage
+from langgraph.channels.delta import DeltaChannel
+
+# `_messages_delta_reducer` is upstream's batching-invariant counterpart of
+# `add_messages` for use with `DeltaChannel`. It is private (leading underscore)
+# but is documented in its own docstring with a `DeltaChannel` usage example —
+# treat as semi-public within the `langgraph>=1.2,<2` pin.
+from langgraph.graph.message import (
+    REMOVE_ALL_MESSAGES,
+    _messages_delta_reducer,  # pyright: ignore[reportPrivateUsage]
+)
+from pydantic import BaseModel
+
+from intentkit.models.agent import Agent
+from intentkit.models.chat import AuthorType, ChatMessageAttachment
+
+
+class AgentError(str, Enum):
+    """The error types that can be raised by the agent."""
+
+    INSUFFICIENT_CREDITS = "insufficient_credits"
+
+
+def _messages_reducer(
+    state: list[AnyMessage], writes: Sequence[Any]
+) -> list[AnyMessage]:
+    """Batching-invariant messages reducer for `DeltaChannel`.
+
+    Wraps `_messages_delta_reducer` to add `REMOVE_ALL_MESSAGES` handling,
+    which the upstream batched reducer documents as out-of-scope. Without
+    this wrapper, `SummarizationMiddleware` (which emits
+    `[RemoveMessage(REMOVE_ALL_MESSAGES), summary, ...preserved]`) silently
+    no-ops the reset and the message list grows unbounded.
+
+    Strategy: flatten writes once, find the *last* `REMOVE_ALL_MESSAGES`
+    sentinel. If present, reset state to empty and delegate to the upstream
+    reducer with only the post-sentinel writes — preserves batching
+    invariance (`reduce(reduce(s, A), B) == reduce(s, A+B)`) across all
+    combinations of resets and normal writes.
+    """
+    flat: list[Any] = []
+    for w in writes:
+        if isinstance(w, list):
+            flat.extend(w)
+        else:
+            flat.append(w)
+    last_reset = -1
+    for i, m in enumerate(flat):
+        if isinstance(m, RemoveMessage) and m.id == REMOVE_ALL_MESSAGES:
+            last_reset = i
+    if last_reset >= 0:
+        return _messages_delta_reducer([], [flat[last_reset + 1 :]])
+    return _messages_delta_reducer(state, writes)  # pyright: ignore[reportArgumentType]
+
+
+class AgentState(BaseAgentState[Any]):
+    """The state of the agent.
+
+    `messages` is overridden to use `DeltaChannel` so checkpointing stores
+    incremental writes rather than a full snapshot per step — for long chat
+    threads this turns O(N^2) checkpoint growth into O(N). `_messages_reducer`
+    delegates to the upstream batching-invariant `_messages_delta_reducer`
+    but adds `REMOVE_ALL_MESSAGES` handling so the bundled
+    `SummarizationMiddleware` reset path still works.
+
+    `snapshot_frequency=50` is tuned for chat workloads (a few-to-dozens of
+    steps per thread) rather than the upstream default of 1000 which targets
+    long workflow graphs. 50 caps resume-replay cost at one short reducer
+    pass while keeping snapshot writes rare for typical conversations.
+    """
+
+    messages: Annotated[
+        list[AnyMessage],
+        DeltaChannel(_messages_reducer, snapshot_frequency=50),
+    ]
+    context: dict[str, Any]
+    error: NotRequired[AgentError]
+    step_count: NotRequired[int]
+    __extra__: NotRequired[dict[str, Any]]
+
+
+class AgentContext(BaseModel):
+    agent_id: str
+    get_agent: Callable[[], Agent]
+    chat_id: str
+    user_id: str | None = None
+    team_id: str | None = None
+    app_id: str | None = None
+    entrypoint: AuthorType
+    is_private: bool
+    thinking: bool = False
+    payer: str | None = None
+    start_message_id: str = ""
+    start_message_attachments: list[ChatMessageAttachment] | None = None
+    call_depth: int = 0
+
+    @property
+    def agent(self) -> Agent:
+        return self.get_agent()
+
+    @property
+    def thread_id(self) -> str:
+        return f"{self.agent_id}-{self.chat_id}"

{intentkit-1.2.11 → intentkit-1.2.13}/intentkit/core/executor.py

@@ -88,6 +88,7 @@ async def build_executor(
         TodoListMiddleware,
         ToolRetryMiddleware,
     )
+    from langchain_anthropic.middleware import AnthropicPromptCachingMiddleware
 
     from intentkit.core.middleware import (
         DynamicPromptMiddleware,
@@ -257,6 +258,11 @@ async def build_executor(
         ModelRetryMiddleware(),
     ]
 
+    # Anthropic prompt caching: 5m ephemeral cache slashes input-token cost on
+    # long system prompts + repeated history. No-op for non-Anthropic providers.
+    if model_provider == LLMProvider.ANTHROPIC_COMPATIBLE:
+        middleware.append(AnthropicPromptCachingMiddleware())
+
     if agent.enable_todo:
         middleware.append(TodoListMiddleware())
 

{intentkit-1.2.11 → intentkit-1.2.13}/intentkit/models/llm.csv

@@ -20,15 +20,17 @@ gemini-3.1-pro-preview-customtools,Gemini 3.1 Pro,google,TRUE,2,0.2,12,4,1048576
 google/gemini-3.1-pro-preview-customtools,Gemini 3.1 Pro,openrouter,TRUE,2,0.2,12,4,1048576,65536,5,3,TRUE,TRUE,TRUE,TRUE,high,TRUE,TRUE,TRUE,300
 gemini-3.1-flash-lite-preview,Gemini 3.1 Flash Lite,google,TRUE,0.25,0.025,1.5,2,1048576,65536,3,5,TRUE,TRUE,TRUE,TRUE,high,TRUE,FALSE,FALSE,300
 google/gemini-3.1-flash-lite-preview,Gemini 3.1 Flash Lite,openrouter,TRUE,0.25,0.025,1.5,2,1048576,65536,3,5,TRUE,TRUE,TRUE,TRUE,high,TRUE,TRUE,TRUE,300
+gemini-3.5-flash,Gemini 3.5 Flash,google,TRUE,1.5,0.15,9,3,1048576,65536,5,4,TRUE,TRUE,TRUE,TRUE,high,TRUE,FALSE,FALSE,300
+google/gemini-3.5-flash,Gemini 3.5 Flash,openrouter,TRUE,1.5,0.15,9,3,1048576,65536,5,4,TRUE,TRUE,TRUE,TRUE,high,TRUE,TRUE,TRUE,300
 gpt-5.4,GPT-5.4,openai,TRUE,2.5,0.25,15,4,272000,128000,5,3,TRUE,FALSE,FALSE,TRUE,high,FALSE,FALSE,FALSE,180
 openai/gpt-5.4,GPT-5.4,openrouter,TRUE,2.5,0.25,15,4,272000,128000,5,3,TRUE,FALSE,FALSE,TRUE,high,FALSE,TRUE,TRUE,180
 gpt-5.4-nano,GPT-5.4 Nano,openai,TRUE,0.1,0.01,0.625,1,400000,128000,3,5,TRUE,FALSE,FALSE,TRUE,minimal,FALSE,FALSE,FALSE,180
 openai/gpt-5.4-nano,GPT-5.4 Nano,openrouter,TRUE,0.1,0.01,0.625,1,400000,128000,3,5,TRUE,FALSE,FALSE,TRUE,minimal,FALSE,TRUE,TRUE,180
 gpt-5.4-mini,GPT-5.4 Mini,openai,TRUE,0.375,0.0375,2.25,2,400000,128000,4,4,TRUE,FALSE,FALSE,TRUE,minimal,FALSE,FALSE,FALSE,180
 openai/gpt-5.4-mini,GPT-5.4 Mini,openrouter,TRUE,0.375,0.0375,2.25,2,400000,128000,4,4,TRUE,FALSE,FALSE,TRUE,minimal,FALSE,TRUE,TRUE,180
-deepseek-v4-flash,Deepseek V4 Flash,deepseek,TRUE,0.14,0.028,0.28,1,1048576,32000,4,4,FALSE,FALSE,FALSE,FALSE,none,TRUE,TRUE,TRUE,300
+deepseek-v4-flash,Deepseek V4 Flash,deepseek,TRUE,0.14,0.0028,0.28,1,1048576,32000,4,4,FALSE,FALSE,FALSE,FALSE,none,TRUE,TRUE,TRUE,300
 deepseek/deepseek-v4-flash,Deepseek V4 Flash,openrouter,TRUE,0.14,0.028,0.28,1,1048576,32000,4,4,FALSE,FALSE,FALSE,FALSE,none,TRUE,TRUE,TRUE,300
-deepseek-v4-pro,Deepseek V4 Pro,deepseek,TRUE,1.74,0.145,3.48,3,1048576,64000,5,2,FALSE,FALSE,FALSE,FALSE,high,TRUE,TRUE,TRUE,300
+deepseek-v4-pro,Deepseek V4 Pro,deepseek,TRUE,0.435,0.003625,0.87,2,1048576,64000,5,2,FALSE,FALSE,FALSE,FALSE,high,TRUE,TRUE,TRUE,300
 deepseek/deepseek-v4-pro,Deepseek V4 Pro,openrouter,TRUE,1.74,0.145,3.48,3,1048576,64000,5,2,FALSE,FALSE,FALSE,FALSE,high,TRUE,TRUE,TRUE,300
 grok-4.3,Grok 4.3,xai,TRUE,1.25,0.125,2.5,2,1000000,128000,5,4,TRUE,FALSE,FALSE,TRUE,high,TRUE,FALSE,FALSE,180
 x-ai/grok-4.3,Grok 4.3,openrouter,TRUE,1.25,0.125,2.5,2,1000000,128000,5,4,TRUE,FALSE,FALSE,TRUE,high,TRUE,TRUE,TRUE,180

{intentkit-1.2.11 → intentkit-1.2.13}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "intentkit"
-version = "1.2.11"
+version = "1.2.13"
 description = "Intent-based AI Agent Platform - Core Package"
 authors = [
     { name = "Ruihua", email = "ruihua@crestal.network" },
@@ -29,21 +29,21 @@ dependencies = [
     "redis>=5.2.1,<8.0.0",
     "supabase>=2.16.0",
     # LangChain & LLM
-    "langchain>=0.3.25",
-    "langchain-community>=0.3.19",
-    "langchain-core>=0.3.43",
-    "langchain-openai>=0.3.8",
-    "langchain-text-splitters>=0.3.8",
-    "langchain-xai>=0.2.1",
-    "langchain-deepseek>=0.1.4",
-    "langchain-google-genai>=3.2.0",
-    "langchain-anthropic>=0.3.12",
-    "langchain-openrouter>=0.2.1",
+    "langchain>=1.3,<2",
+    "langchain-community>=0.4,<1",
+    "langchain-core>=1.4,<2",
+    "langchain-openai>=1.2,<2",
+    "langchain-text-splitters>=1.1,<2",
+    "langchain-xai>=1.2,<2",
+    "langchain-deepseek>=1.0,<2",
+    "langchain-google-genai>=4.2,<5",
+    "langchain-anthropic>=1.4,<2",
+    "langchain-openrouter>=0.2.3,<1",
     "openrouter>=0.9.0",
-    "langgraph>=0.6.1",
-    "langgraph-checkpoint>=2.0.18",
-    "langgraph-checkpoint-postgres>=2.0.16",
-    "langgraph-prebuilt>=0.6.1",
+    "langgraph>=1.2,<2",
+    "langgraph-checkpoint>=4.1,<5",
+    "langgraph-checkpoint-postgres>=3.1,<4",
+    "langgraph-prebuilt>=1.1,<2",
     "openai>=1.59.6",
     # Blockchain / Web3
     "cdp-sdk>=1.31.1",
@@ -120,7 +120,7 @@ dev = [
     # Database migrations
     "alembic>=1.14.0",
     # Local LLM testing
-    "langchain-ollama>=0.2.0",
+    "langchain-ollama>=1.1,<2",
     "jsonschema>=4.26.0",
 ]
 

intentkit-1.2.11/intentkit/abstracts/graph.py

@@ -1,52 +0,0 @@
-from collections.abc import Callable
-from enum import Enum
-from typing import Any, NotRequired
-
-from langchain.agents import AgentState as BaseAgentState
-from pydantic import BaseModel
-
-from intentkit.models.agent import Agent
-from intentkit.models.chat import AuthorType, ChatMessageAttachment
-
-
-class AgentError(str, Enum):
-    """The error types that can be raised by the agent."""
-
-    INSUFFICIENT_CREDITS = "insufficient_credits"
-
-
-# We create the AgentState that we will pass around
-# This simply involves a list of messages
-# We want steps to return messages to append to the list
-# So we annotate the messages attribute with operator.add
-class AgentState(BaseAgentState[Any]):
-    """The state of the agent."""
-
-    context: dict[str, Any]
-    error: NotRequired[AgentError]
-    step_count: NotRequired[int]
-    __extra__: NotRequired[dict[str, Any]]
-
-
-class AgentContext(BaseModel):
-    agent_id: str
-    get_agent: Callable[[], Agent]
-    chat_id: str
-    user_id: str | None = None
-    team_id: str | None = None
-    app_id: str | None = None
-    entrypoint: AuthorType
-    is_private: bool
-    thinking: bool = False
-    payer: str | None = None
-    start_message_id: str = ""
-    start_message_attachments: list[ChatMessageAttachment] | None = None
-    call_depth: int = 0
-
-    @property
-    def agent(self) -> Agent:
-        return self.get_agent()
-
-    @property
-    def thread_id(self) -> str:
-        return f"{self.agent_id}-{self.chat_id}"