Skip to content

The agent engineering platform.

chore(langchain_v1): add runtime and context to model request (#33365)

Eugene Yurtsev982a950ccfa2
diff --git a/libs/langchain_v1/langchain/agents/factory.py b/libs/langchain_v1/langchain/agents/factory.py
index e9214313d1..0a0f53fbf3 100644
--- a/libs/langchain_v1/langchain/agents/factory.py
+++ b/libs/langchain_v1/langchain/agents/factory.py
@@ -87,13 +87,13 @@ class _InternalModelResponse:
def _chain_model_call_handlers(
handlers: Sequence[
Callable[
- [ModelRequest, Any, Any, Callable[[ModelRequest], AIMessage]],
+ [ModelRequest, Callable[[ModelRequest], AIMessage]],
AIMessage,
]
],
) -> (
Callable[
- [ModelRequest, Any, Any, Callable[[ModelRequest], AIMessage]],
+ [ModelRequest, Callable[[ModelRequest], AIMessage]],
AIMessage,
]
| None
@@ -141,31 +141,29 @@ def _chain_model_call_handlers(
def compose_two(
outer: Callable[
- [ModelRequest, Any, Any, Callable[[ModelRequest], AIMessage]],
+ [ModelRequest, Callable[[ModelRequest], AIMessage]],
AIMessage,
],
inner: Callable[
- [ModelRequest, Any, Any, Callable[[ModelRequest], AIMessage]],
+ [ModelRequest, Callable[[ModelRequest], AIMessage]],
AIMessage,
],
) -> Callable[
- [ModelRequest, Any, Any, Callable[[ModelRequest], AIMessage]],
+ [ModelRequest, Callable[[ModelRequest], AIMessage]],
AIMessage,
]:
"""Compose two handlers where outer wraps inner."""
def composed(
request: ModelRequest,
- state: Any,
- runtime: Any,
handler: Callable[[ModelRequest], AIMessage],
) -> AIMessage:
# Create a wrapper that calls inner with the base handler
def inner_handler(req: ModelRequest) -> AIMessage:
- return inner(req, state, runtime, handler)
+ return inner(req, handler)
# Call outer with the wrapped inner as its handler
- return outer(request, state, runtime, inner_handler)
+ return outer(request, inner_handler)
return composed
@@ -180,13 +178,13 @@ def _chain_model_call_handlers(
def _chain_async_model_call_handlers(
handlers: Sequence[
Callable[
- [ModelRequest, Any, Any, Callable[[ModelRequest], Awaitable[AIMessage]]],
+ [ModelRequest, Callable[[ModelRequest], Awaitable[AIMessage]]],
Awaitable[AIMessage],
]
],
) -> (
Callable[
- [ModelRequest, Any, Any, Callable[[ModelRequest], Awaitable[AIMessage]]],
+ [ModelRequest, Callable[[ModelRequest], Awaitable[AIMessage]]],
Awaitable[AIMessage],
]
| None
@@ -207,31 +205,29 @@ def _chain_async_model_call_handlers(
def compose_two(
outer: Callable[
- [ModelRequest, Any, Any, Callable[[ModelRequest], Awaitable[AIMessage]]],
+ [ModelRequest, Callable[[ModelRequest], Awaitable[AIMessage]]],
Awaitable[AIMessage],
],
inner: Callable[
- [ModelRequest, Any, Any, Callable[[ModelRequest], Awaitable[AIMessage]]],
+ [ModelRequest, Callable[[ModelRequest], Awaitable[AIMessage]]],
Awaitable[AIMessage],
],
) -> Callable[
- [ModelRequest, Any, Any, Callable[[ModelRequest], Awaitable[AIMessage]]],
+ [ModelRequest, Callable[[ModelRequest], Awaitable[AIMessage]]],
Awaitable[AIMessage],
]:
"""Compose two async handlers where outer wraps inner."""
async def composed(
request: ModelRequest,
- state: Any,
- runtime: Any,
handler: Callable[[ModelRequest], Awaitable[AIMessage]],
) -> AIMessage:
# Create a wrapper that calls inner with the base handler
async def inner_handler(req: ModelRequest) -> AIMessage:
- return await inner(req, state, runtime, handler)
+ return await inner(req, handler)
# Call outer with the wrapped inner as its handler
- return await outer(request, state, runtime, inner_handler)
+ return await outer(request, inner_handler)
return composed
@@ -900,12 +896,14 @@ def create_agent( # noqa: PLR0915
response_format=initial_response_format,
messages=state["messages"],
tool_choice=None,
+ state=state,
+ runtime=runtime,
)
# Apply modify_model_request middleware in sequence
for m in middleware_w_modify_model_request:
if m.__class__.modify_model_request is not AgentMiddleware.modify_model_request:
- request = m.modify_model_request(request, state, runtime)
+ request = m.modify_model_request(request)
else:
msg = (
f"No synchronous function provided for "
@@ -935,7 +933,7 @@ def create_agent( # noqa: PLR0915
output = base_handler(request)
else:
# Call composed handler with base handler
- output = on_model_call_handler(request, state, runtime, base_handler)
+ output = on_model_call_handler(request, base_handler)
return {
"thread_model_call_count": state.get("thread_model_call_count", 0) + 1,
"run_model_call_count": state.get("run_model_call_count", 0) + 1,
@@ -977,11 +975,13 @@ def create_agent( # noqa: PLR0915
response_format=initial_response_format,
messages=state["messages"],
tool_choice=None,
+ state=state,
+ runtime=runtime,
)
# Apply modify_model_request middleware in sequence
for m in middleware_w_modify_model_request:
- request = await m.amodify_model_request(request, state, runtime)
+ request = await m.amodify_model_request(request)
# Execute with or without handler
effective_response_format: Any = None
@@ -1003,7 +1003,7 @@ def create_agent( # noqa: PLR0915
output = await base_handler(request)
else:
# Call composed async handler with base handler
- output = await aon_model_call_handler(request, state, runtime, base_handler)
+ output = await aon_model_call_handler(request, base_handler)
return {
"thread_model_call_count": state.get("thread_model_call_count", 0) + 1,
"run_model_call_count": state.get("run_model_call_count", 0) + 1,
diff --git a/libs/langchain_v1/langchain/agents/middleware/context_editing.py b/libs/langchain_v1/langchain/agents/middleware/context_editing.py
index 017c460083..fcd244bce6 100644
--- a/libs/langchain_v1/langchain/agents/middleware/context_editing.py
+++ b/libs/langchain_v1/langchain/agents/middleware/context_editing.py
@@ -10,7 +10,7 @@ from __future__ import annotations
from collections.abc import Callable, Iterable, Sequence
from dataclasses import dataclass
-from typing import TYPE_CHECKING, Literal
+from typing import Literal
from langchain_core.messages import (
AIMessage,
@@ -22,10 +22,7 @@ from langchain_core.messages import (
from langchain_core.messages.utils import count_tokens_approximately
from typing_extensions import Protocol
-from langchain.agents.middleware.types import AgentMiddleware, AgentState, ModelRequest
-
-if TYPE_CHECKING:
- from langgraph.runtime import Runtime
+from langchain.agents.middleware.types import AgentMiddleware, ModelRequest
DEFAULT_TOOL_PLACEHOLDER = "[cleared]"
@@ -212,8 +209,6 @@ class ContextEditingMiddleware(AgentMiddleware):
def modify_model_request(
self,
request: ModelRequest,
- state: AgentState, # noqa: ARG002
- runtime: Runtime, # noqa: ARG002
) -> ModelRequest:
"""Modify the model request by applying context edits before invocation."""
if not request.messages:
diff --git a/libs/langchain_v1/langchain/agents/middleware/model_fallback.py b/libs/langchain_v1/langchain/agents/middleware/model_fallback.py
index a0b8a2d1c2..69b8fdccb3 100644
--- a/libs/langchain_v1/langchain/agents/middleware/model_fallback.py
+++ b/libs/langchain_v1/langchain/agents/middleware/model_fallback.py
@@ -2,7 +2,7 @@
from __future__ import annotations
-from typing import TYPE_CHECKING, Any
+from typing import TYPE_CHECKING
from langchain.agents.middleware.types import (
AgentMiddleware,
@@ -15,8 +15,6 @@ if TYPE_CHECKING:
from langchain_core.language_models.chat_models import BaseChatModel
from langchain_core.messages import AIMessage
- from langgraph.runtime import Runtime
- from langgraph.typing import ContextT
class ModelFallbackMiddleware(AgentMiddleware):
@@ -70,8 +68,6 @@ class ModelFallbackMiddleware(AgentMiddleware):
def on_model_call(
self,
request: ModelRequest,
- state: Any, # noqa: ARG002
- runtime: Runtime[ContextT], # noqa: ARG002
handler: Callable[[ModelRequest], AIMessage],
) -> AIMessage:
"""Try fallback models in sequence on errors.
diff --git a/libs/langchain_v1/langchain/agents/middleware/planning.py b/libs/langchain_v1/langchain/agents/middleware/planning.py
index 98e76b62fd..a8873cb9da 100644
--- a/libs/langchain_v1/langchain/agents/middleware/planning.py
+++ b/libs/langchain_v1/langchain/agents/middleware/planning.py
@@ -3,7 +3,7 @@
from __future__ import annotations
-from typing import TYPE_CHECKING, Annotated, Literal
+from typing import Annotated, Literal
from langchain_core.messages import ToolMessage
from langchain_core.tools import tool
@@ -13,9 +13,6 @@ from typing_extensions import NotRequired, TypedDict
from langchain.agents.middleware.types import AgentMiddleware, AgentState, ModelRequest
from langchain.tools import InjectedToolCallId
-if TYPE_CHECKING:
- from langgraph.runtime import Runtime
-
class Todo(TypedDict):
"""A single todo item with content and status."""
@@ -189,8 +186,6 @@ class PlanningMiddleware(AgentMiddleware):
def modify_model_request(
self,
request: ModelRequest,
- state: AgentState, # noqa: ARG002
- runtime: Runtime, # noqa: ARG002
) -> ModelRequest:
"""Update the system prompt to include the todo system prompt."""
request.system_prompt = (
diff --git a/libs/langchain_v1/langchain/agents/middleware/prompt_caching.py b/libs/langchain_v1/langchain/agents/middleware/prompt_caching.py
index 3613263188..dfbd0a4978 100644
--- a/libs/langchain_v1/langchain/agents/middleware/prompt_caching.py
+++ b/libs/langchain_v1/langchain/agents/middleware/prompt_caching.py
@@ -3,9 +3,7 @@
from typing import Literal
from warnings import warn
-from langgraph.runtime import Runtime
-
-from langchain.agents.middleware.types import AgentMiddleware, AgentState, ModelRequest
+from langchain.agents.middleware.types import AgentMiddleware, ModelRequest
class AnthropicPromptCachingMiddleware(AgentMiddleware):
@@ -44,8 +42,6 @@ class AnthropicPromptCachingMiddleware(AgentMiddleware):
def modify_model_request(
self,
request: ModelRequest,
- state: AgentState, # noqa: ARG002
- runtime: Runtime, # noqa: ARG002
) -> ModelRequest:
"""Modify the model request to add cache control blocks."""
try:
diff --git a/libs/langchain_v1/langchain/agents/middleware/tool_selection.py b/libs/langchain_v1/langchain/agents/middleware/tool_selection.py
index fe17f1c1b7..839cfff379 100644
--- a/libs/langchain_v1/langchain/agents/middleware/tool_selection.py
+++ b/libs/langchain_v1/langchain/agents/middleware/tool_selection.py
@@ -11,13 +11,10 @@ from langchain_core.messages import HumanMessage
from pydantic import Field, TypeAdapter
from typing_extensions import TypedDict
-from langchain.agents.middleware.types import AgentMiddleware, AgentState, ModelRequest, StateT
+from langchain.agents.middleware.types import AgentMiddleware, ModelRequest
from langchain.chat_models.base import init_chat_model
if TYPE_CHECKING:
- from langgraph.runtime import Runtime
- from langgraph.typing import ContextT
-
from langchain.tools import BaseTool
logger = logging.getLogger(__name__)
@@ -246,8 +243,6 @@ class LLMToolSelectorMiddleware(AgentMiddleware):
def modify_model_request(
self,
request: ModelRequest,
- state: StateT, # noqa: ARG002
- runtime: Runtime[ContextT], # noqa: ARG002
) -> ModelRequest:
"""Modify the model request to filter tools based on LLM selection."""
selection_request = self._prepare_selection_request(request)
@@ -277,8 +272,6 @@ class LLMToolSelectorMiddleware(AgentMiddleware):
async def amodify_model_request(
self,
request: ModelRequest,
- state: AgentState, # noqa: ARG002
- runtime: Runtime, # noqa: ARG002
) -> ModelRequest:
"""Modify the model request to filter tools based on LLM selection."""
selection_request = self._prepare_selection_request(request)
diff --git a/libs/langchain_v1/langchain/agents/middleware/types.py b/libs/langchain_v1/langchain/agents/middleware/types.py
index 0db65ac7e7..7a9f2ca4e0 100644
--- a/libs/langchain_v1/langchain/agents/middleware/types.py
+++ b/libs/langchain_v1/langchain/agents/middleware/types.py
@@ -72,6 +72,8 @@ class ModelRequest:
tool_choice: Any | None
tools: list[BaseTool | dict]
response_format: ResponseFormat | None
+ state: AgentState
+ runtime: Runtime[ContextT] # type: ignore[valid-type]
model_settings: dict[str, Any] = field(default_factory=dict)
@@ -160,8 +162,6 @@ class AgentMiddleware(Generic[StateT, ContextT]):
def modify_model_request(
self,
request: ModelRequest,
- state: StateT, # noqa: ARG002
- runtime: Runtime[ContextT], # noqa: ARG002
) -> ModelRequest:
"""Logic to modify request kwargs before the model is called."""
return request
@@ -169,11 +169,9 @@ class AgentMiddleware(Generic[StateT, ContextT]):
async def amodify_model_request(
self,
request: ModelRequest,
- state: StateT,
- runtime: Runtime[ContextT],
) -> ModelRequest:
"""Async logic to modify request kwargs before the model is called."""
- return await run_in_executor(None, self.modify_model_request, request, state, runtime)
+ return await run_in_executor(None, self.modify_model_request, request)
def after_model(self, state: StateT, runtime: Runtime[ContextT]) -> dict[str, Any] | None:
"""Logic to run after the model is called."""
@@ -186,8 +184,6 @@ class AgentMiddleware(Generic[StateT, ContextT]):
def on_model_call(
self,
request: ModelRequest,
- state: StateT,
- runtime: Runtime[ContextT],
handler: Callable[[ModelRequest], AIMessage],
) -> AIMessage:
"""Intercept and control model execution via handler callback.
@@ -198,9 +194,7 @@ class AgentMiddleware(Generic[StateT, ContextT]):
compose with first in list as outermost layer.
Args:
- request: Initial model request to execute.
- state: Current agent state.
- runtime: LangGraph runtime context.
+ request: Model request to execute (includes state and runtime).
handler: Callback that executes the model request and returns AIMessage.
Call this to execute the model. Can be called multiple times
for retry logic. Can skip calling it to short-circuit.
@@ -211,7 +205,7 @@ class AgentMiddleware(Generic[StateT, ContextT]):
Examples:
Retry on error:
```python
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
for attempt in range(3):
try:
return handler(request)
@@ -222,14 +216,14 @@ class AgentMiddleware(Generic[StateT, ContextT]):
Rewrite response:
```python
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
result = handler(request)
return AIMessage(content=f"[{result.content}]")
```
Error to fallback:
```python
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
try:
return handler(request)
except Exception:
@@ -238,7 +232,7 @@ class AgentMiddleware(Generic[StateT, ContextT]):
Cache/short-circuit:
```python
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
if cached := get_cache(request):
return cached # Short-circuit with cached result
result = handler(request)
@@ -251,16 +245,12 @@ class AgentMiddleware(Generic[StateT, ContextT]):
async def aon_model_call(
self,
request: ModelRequest,
- state: StateT,
- runtime: Runtime[ContextT],
handler: Callable[[ModelRequest], Awaitable[AIMessage]],
) -> AIMessage:
"""Async version of on_model_call.
Args:
- request: Initial model request to execute.
- state: Current agent state.
- runtime: LangGraph runtime context.
+ request: Model request to execute (includes state and runtime).
handler: Async callback that executes the model request.
Returns:
@@ -269,7 +259,7 @@ class AgentMiddleware(Generic[StateT, ContextT]):
Examples:
Retry on error:
```python
- async def aon_model_call(self, request, state, runtime, handler):
+ async def aon_model_call(self, request, handler):
for attempt in range(3):
try:
return await handler(request)
@@ -368,27 +358,23 @@ class _CallableWithStateAndRuntime(Protocol[StateT_contra, ContextT]):
...
-class _CallableWithModelRequestAndStateAndRuntime(Protocol[StateT_contra, ContextT]):
- """Callable with ModelRequest, AgentState, and Runtime as arguments."""
+class _CallableWithModelRequestAndStateAndRuntime(Protocol[StateT_contra, ContextT]): # type: ignore[misc]
+ """Callable with ModelRequest as argument (contains state and runtime)."""
- def __call__(
- self, request: ModelRequest, state: StateT_contra, runtime: Runtime[ContextT]
- ) -> ModelRequest | Awaitable[ModelRequest]:
- """Perform some logic with the model request, state, and runtime."""
+ def __call__(self, request: ModelRequest) -> ModelRequest | Awaitable[ModelRequest]:
+ """Perform some logic with the model request."""
...
-class _CallableReturningPromptString(Protocol[StateT_contra, ContextT]):
- """Callable that returns a prompt string given ModelRequest, AgentState, and Runtime."""
+class _CallableReturningPromptString(Protocol[StateT_contra, ContextT]): # type: ignore[misc]
+ """Callable that returns a prompt string given ModelRequest (contains state and runtime)."""
- def __call__(
- self, request: ModelRequest, state: StateT_contra, runtime: Runtime[ContextT]
- ) -> str | Awaitable[str]:
- """Generate a system prompt string based on the request, state, and runtime."""
+ def __call__(self, request: ModelRequest) -> str | Awaitable[str]:
+ """Generate a system prompt string based on the request."""
...
-class _CallableReturningModelResponse(Protocol[StateT_contra, ContextT]):
+class _CallableReturningModelResponse(Protocol[StateT_contra, ContextT]): # type: ignore[misc]
"""Callable for model call interception with handler callback.
Receives handler callback to execute model and returns final AIMessage.
@@ -397,8 +383,6 @@ class _CallableReturningModelResponse(Protocol[StateT_contra, ContextT]):
def __call__(
self,
request: ModelRequest,
- state: StateT_contra,
- runtime: Runtime[ContextT],
handler: Callable[[ModelRequest], AIMessage],
) -> AIMessage:
"""Intercept model execution via handler callback."""
@@ -634,8 +618,7 @@ def modify_model_request(
Args:
func: The function to be decorated. Must accept:
- `request: ModelRequest, state: StateT, runtime: Runtime[ContextT]` -
- Model request, state, and runtime context
+ `request: ModelRequest` - Model request (contains state and runtime)
state_schema: Optional custom state schema type. If not provided, uses the default
AgentState schema.
tools: Optional list of additional tools to register with this middleware.
@@ -653,9 +636,7 @@ def modify_model_request(
Basic usage to modify system prompt:
```python
@modify_model_request
- def add_context_to_prompt(
- request: ModelRequest, state: AgentState, runtime: Runtime
- ) -> ModelRequest:
+ def add_context_to_prompt(request: ModelRequest) -> ModelRequest:
if request.system_prompt:
request.system_prompt += "\n\nAdditional context: ..."
else:
@@ -666,11 +647,9 @@ def modify_model_request(
Usage with runtime and custom model settings:
```python
@modify_model_request
- def dynamic_model_settings(
- request: ModelRequest, state: AgentState, runtime: Runtime
- ) -> ModelRequest:
+ def dynamic_model_settings(request: ModelRequest) -> ModelRequest:
# Use a different model based on user subscription tier
- if runtime.context.get("subscription_tier") == "premium":
+ if request.runtime.context.get("subscription_tier") == "premium":
request.model = "gpt-4o"
else:
request.model = "gpt-4o-mini"
@@ -689,10 +668,8 @@ def modify_model_request(
async def async_wrapped(
self: AgentMiddleware[StateT, ContextT], # noqa: ARG001
request: ModelRequest,
- state: StateT,
- runtime: Runtime[ContextT],
) -> ModelRequest:
- return await func(request, state, runtime) # type: ignore[misc]
+ return await func(request) # type: ignore[misc]
middleware_name = name or cast(
"str", getattr(func, "__name__", "ModifyModelRequestMiddleware")
@@ -711,10 +688,8 @@ def modify_model_request(
def wrapped(
self: AgentMiddleware[StateT, ContextT], # noqa: ARG001
request: ModelRequest,
- state: StateT,
- runtime: Runtime[ContextT],
) -> ModelRequest:
- return func(request, state, runtime) # type: ignore[return-value]
+ return func(request) # type: ignore[return-value]
middleware_name = name or cast(
"str", getattr(func, "__name__", "ModifyModelRequestMiddleware")
@@ -1171,8 +1146,7 @@ def dynamic_prompt(
Args:
func: The function to be decorated. Must accept:
- `request: ModelRequest, state: StateT, runtime: Runtime[ContextT]` -
- Model request, state, and runtime context
+ `request: ModelRequest` - Model request (contains state and runtime)
Returns:
Either an AgentMiddleware instance (if func is provided) or a decorator function
@@ -1185,16 +1159,16 @@ def dynamic_prompt(
Basic usage with dynamic content:
```python
@dynamic_prompt
- def my_prompt(request: ModelRequest, state: AgentState, runtime: Runtime) -> str:
- user_name = runtime.context.get("user_name", "User")
+ def my_prompt(request: ModelRequest) -> str:
+ user_name = request.runtime.context.get("user_name", "User")
return f"You are a helpful assistant helping {user_name}."
```
Using state to customize the prompt:
```python
@dynamic_prompt
- def context_aware_prompt(request: ModelRequest, state: AgentState, runtime: Runtime) -> str:
- msg_count = len(state["messages"])
+ def context_aware_prompt(request: ModelRequest) -> str:
+ msg_count = len(request.state["messages"])
if msg_count > 10:
return "You are in a long conversation. Be concise."
return "You are a helpful assistant."
@@ -1216,10 +1190,8 @@ def dynamic_prompt(
async def async_wrapped(
self: AgentMiddleware[StateT, ContextT], # noqa: ARG001
request: ModelRequest,
- state: StateT,
- runtime: Runtime[ContextT],
) -> ModelRequest:
- prompt = await func(request, state, runtime) # type: ignore[misc]
+ prompt = await func(request) # type: ignore[misc]
request.system_prompt = prompt
return request
@@ -1238,10 +1210,8 @@ def dynamic_prompt(
def wrapped(
self: AgentMiddleware[StateT, ContextT], # noqa: ARG001
request: ModelRequest,
- state: StateT,
- runtime: Runtime[ContextT],
) -> ModelRequest:
- prompt = cast("str", func(request, state, runtime))
+ prompt = cast("str", func(request))
request.system_prompt = prompt
return request
@@ -1300,8 +1270,8 @@ def on_model_call(
model calls, implement retry logic, handle errors, and rewrite responses.
Args:
- func: Function accepting (request, state, runtime, handler) that calls
- handler(request) to execute the model and returns final AIMessage.
+ func: Function accepting (request, handler) that calls handler(request)
+ to execute the model and returns final AIMessage. Request contains state and runtime.
state_schema: Custom state schema. Defaults to AgentState.
tools: Additional tools to register with this middleware.
name: Middleware class name. Defaults to function name.
@@ -1313,7 +1283,7 @@ def on_model_call(
Basic retry logic:
```python
@on_model_call
- def retry_on_error(request, state, runtime, handler):
+ def retry_on_error(request, handler):
max_retries = 3
for attempt in range(max_retries):
try:
@@ -1326,7 +1296,7 @@ def on_model_call(
Model fallback:
```python
@on_model_call
- def fallback_model(request, state, runtime, handler):
+ def fallback_model(request, handler):
# Try primary model
try:
return handler(request)
@@ -1341,7 +1311,7 @@ def on_model_call(
Rewrite response content:
```python
@on_model_call
- def uppercase_responses(request, state, runtime, handler):
+ def uppercase_responses(request, handler):
result = handler(request)
return AIMessage(content=result.content.upper())
```
@@ -1357,11 +1327,9 @@ def on_model_call(
async def async_wrapped(
self: AgentMiddleware[StateT, ContextT], # noqa: ARG001
request: ModelRequest,
- state: StateT,
- runtime: Runtime[ContextT],
handler: Callable[[ModelRequest], Awaitable[AIMessage]],
) -> AIMessage:
- return await func(request, state, runtime, handler) # type: ignore[misc, arg-type]
+ return await func(request, handler) # type: ignore[misc, arg-type]
middleware_name = name or cast(
"str", getattr(func, "__name__", "OnModelCallMiddleware")
@@ -1380,11 +1348,9 @@ def on_model_call(
def wrapped(
self: AgentMiddleware[StateT, ContextT], # noqa: ARG001
request: ModelRequest,
- state: StateT,
- runtime: Runtime[ContextT],
handler: Callable[[ModelRequest], AIMessage],
) -> AIMessage:
- return func(request, state, runtime, handler)
+ return func(request, handler)
middleware_name = name or cast("str", getattr(func, "__name__", "OnModelCallMiddleware"))
diff --git a/libs/langchain_v1/tests/unit_tests/agents/middleware/test_llm_tool_selection.py b/libs/langchain_v1/tests/unit_tests/agents/middleware/test_llm_tool_selection.py
index 6098cf433f..a6039d1cfe 100644
--- a/libs/langchain_v1/tests/unit_tests/agents/middleware/test_llm_tool_selection.py
+++ b/libs/langchain_v1/tests/unit_tests/agents/middleware/test_llm_tool_selection.py
@@ -111,7 +111,7 @@ class TestLLMToolSelectorBasic:
model_requests = []
@modify_model_request
- def trace_model_requests(request: ModelRequest, state: AgentState, runtime) -> ModelRequest:
+ def trace_model_requests(request: ModelRequest) -> ModelRequest:
"""Middleware to select relevant tools based on state/context."""
# Select a small, relevant subset of tools based on state/context
model_requests.append(request)
@@ -216,7 +216,7 @@ class TestMaxToolsLimiting:
model_requests = []
@modify_model_request
- def trace_model_requests(request: ModelRequest, state: AgentState, runtime) -> ModelRequest:
+ def trace_model_requests(request: ModelRequest) -> ModelRequest:
model_requests.append(request)
return request
@@ -271,7 +271,7 @@ class TestMaxToolsLimiting:
model_requests = []
@modify_model_request
- def trace_model_requests(request: ModelRequest, state: AgentState, runtime) -> ModelRequest:
+ def trace_model_requests(request: ModelRequest) -> ModelRequest:
model_requests.append(request)
return request
@@ -333,7 +333,7 @@ class TestAlwaysInclude:
model_requests = []
@modify_model_request
- def trace_model_requests(request: ModelRequest, state: AgentState, runtime) -> ModelRequest:
+ def trace_model_requests(request: ModelRequest) -> ModelRequest:
model_requests.append(request)
return request
@@ -383,7 +383,7 @@ class TestAlwaysInclude:
model_requests = []
@modify_model_request
- def trace_model_requests(request: ModelRequest, state: AgentState, runtime) -> ModelRequest:
+ def trace_model_requests(request: ModelRequest) -> ModelRequest:
model_requests.append(request)
return request
@@ -437,7 +437,7 @@ class TestAlwaysInclude:
model_requests = []
@modify_model_request
- def trace_model_requests(request: ModelRequest, state: AgentState, runtime) -> ModelRequest:
+ def trace_model_requests(request: ModelRequest) -> ModelRequest:
model_requests.append(request)
return request
@@ -494,7 +494,7 @@ class TestDuplicateAndInvalidTools:
model_requests = []
@modify_model_request
- def trace_model_requests(request: ModelRequest, state: AgentState, runtime) -> ModelRequest:
+ def trace_model_requests(request: ModelRequest) -> ModelRequest:
model_requests.append(request)
return request
@@ -547,7 +547,7 @@ class TestDuplicateAndInvalidTools:
model_requests = []
@modify_model_request
- def trace_model_requests(request: ModelRequest, state: AgentState, runtime) -> ModelRequest:
+ def trace_model_requests(request: ModelRequest) -> ModelRequest:
model_requests.append(request)
return request
diff --git a/libs/langchain_v1/tests/unit_tests/agents/middleware/test_on_model_call_decorator.py b/libs/langchain_v1/tests/unit_tests/agents/middleware/test_on_model_call_decorator.py
index 7c0c582f1a..ba4d85c399 100644
--- a/libs/langchain_v1/tests/unit_tests/agents/middleware/test_on_model_call_decorator.py
+++ b/libs/langchain_v1/tests/unit_tests/agents/middleware/test_on_model_call_decorator.py
@@ -20,7 +20,7 @@ class TestOnModelCallDecorator:
"""Test basic decorator usage without parameters."""
@on_model_call
- def passthrough_middleware(request, state, runtime, handler):
+ def passthrough_middleware(request, handler):
return handler(request)
# Should return an AgentMiddleware instance
@@ -38,7 +38,7 @@ class TestOnModelCallDecorator:
"""Test decorator with custom middleware name."""
@on_model_call(name="CustomMiddleware")
- def my_middleware(request, state, runtime, handler):
+ def my_middleware(request, handler):
return handler(request)
assert isinstance(my_middleware, AgentMiddleware)
@@ -56,7 +56,7 @@ class TestOnModelCallDecorator:
return super()._generate(messages, **kwargs)
@on_model_call
- def retry_once(request, state, runtime, handler):
+ def retry_once(request, handler):
try:
return handler(request)
except Exception:
@@ -75,7 +75,7 @@ class TestOnModelCallDecorator:
"""Test decorator for rewriting responses."""
@on_model_call
- def uppercase_responses(request, state, runtime, handler):
+ def uppercase_responses(request, handler):
result = handler(request)
return AIMessage(content=result.content.upper())
@@ -94,7 +94,7 @@ class TestOnModelCallDecorator:
raise ValueError("Model error")
@on_model_call
- def error_to_fallback(request, state, runtime, handler):
+ def error_to_fallback(request, handler):
try:
return handler(request)
except Exception:
@@ -112,8 +112,8 @@ class TestOnModelCallDecorator:
state_values = []
@on_model_call
- def log_state(request, state, runtime, handler):
- state_values.append(state.get("messages"))
+ def log_state(request, handler):
+ state_values.append(request.state.get("messages"))
return handler(request)
model = GenericFakeChatModel(messages=iter([AIMessage(content="Response")]))
@@ -131,14 +131,14 @@ class TestOnModelCallDecorator:
execution_order = []
@on_model_call
- def outer_middleware(request, state, runtime, handler):
+ def outer_middleware(request, handler):
execution_order.append("outer-before")
result = handler(request)
execution_order.append("outer-after")
return result
@on_model_call
- def inner_middleware(request, state, runtime, handler):
+ def inner_middleware(request, handler):
execution_order.append("inner-before")
result = handler(request)
execution_order.append("inner-after")
@@ -165,7 +165,7 @@ class TestOnModelCallDecorator:
custom_field: str
@on_model_call(state_schema=CustomState)
- def middleware_with_schema(request, state, runtime, handler):
+ def middleware_with_schema(request, handler):
return handler(request)
assert isinstance(middleware_with_schema, AgentMiddleware)
@@ -182,7 +182,7 @@ class TestOnModelCallDecorator:
return f"Result: {query}"
@on_model_call(tools=[test_tool])
- def middleware_with_tools(request, state, runtime, handler):
+ def middleware_with_tools(request, handler):
return handler(request)
assert isinstance(middleware_with_tools, AgentMiddleware)
@@ -194,12 +194,12 @@ class TestOnModelCallDecorator:
# Without parentheses
@on_model_call
- def middleware_no_parens(request, state, runtime, handler):
+ def middleware_no_parens(request, handler):
return handler(request)
# With parentheses
@on_model_call()
- def middleware_with_parens(request, state, runtime, handler):
+ def middleware_with_parens(request, handler):
return handler(request)
assert isinstance(middleware_no_parens, AgentMiddleware)
@@ -209,7 +209,7 @@ class TestOnModelCallDecorator:
"""Test that decorator uses function name for class name."""
@on_model_call
- def my_custom_middleware(request, state, runtime, handler):
+ def my_custom_middleware(request, handler):
return handler(request)
assert my_custom_middleware.__class__.__name__ == "my_custom_middleware"
@@ -219,14 +219,14 @@ class TestOnModelCallDecorator:
execution_order = []
@on_model_call
- def decorated_middleware(request, state, runtime, handler):
+ def decorated_middleware(request, handler):
execution_order.append("decorated-before")
result = handler(request)
execution_order.append("decorated-after")
return result
class ClassMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
execution_order.append("class-before")
result = handler(request)
execution_order.append("class-after")
@@ -261,7 +261,7 @@ class TestOnModelCallDecorator:
return super()._generate(messages, **kwargs)
@on_model_call
- def retry_with_tracking(request, state, runtime, handler):
+ def retry_with_tracking(request, handler):
max_retries = 3
last_exception = None
for attempt in range(max_retries):
@@ -289,7 +289,7 @@ class TestOnModelCallDecorator:
call_log = []
@on_model_call
- async def logging_middleware(request, state, runtime, handler):
+ async def logging_middleware(request, handler):
call_log.append("before")
result = await handler(request)
call_log.append("after")
@@ -308,7 +308,7 @@ class TestOnModelCallDecorator:
modified_prompts = []
@on_model_call
- def add_system_prompt(request, state, runtime, handler):
+ def add_system_prompt(request, handler):
# Modify request to add system prompt
modified_request = ModelRequest(
messages=request.messages,
@@ -317,6 +317,8 @@ class TestOnModelCallDecorator:
tool_choice=request.tool_choice,
tools=request.tools,
response_format=request.response_format,
+ state={},
+ runtime=None,
)
modified_prompts.append(modified_request.system_prompt)
return handler(modified_request)
@@ -332,7 +334,7 @@ class TestOnModelCallDecorator:
"""Test decorator applying multiple transformations."""
@on_model_call
- def multi_transform(request, state, runtime, handler):
+ def multi_transform(request, handler):
result = handler(request)
# First transformation: uppercase
diff --git a/libs/langchain_v1/tests/unit_tests/agents/middleware/test_on_model_call_middleware.py b/libs/langchain_v1/tests/unit_tests/agents/middleware/test_on_model_call_middleware.py
index ac0375e537..7a211c4375 100644
--- a/libs/langchain_v1/tests/unit_tests/agents/middleware/test_on_model_call_middleware.py
+++ b/libs/langchain_v1/tests/unit_tests/agents/middleware/test_on_model_call_middleware.py
@@ -19,7 +19,7 @@ class TestBasicOnModelCall:
"""Test middleware that simply passes through without modification."""
class PassthroughMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
return handler(request)
model = GenericFakeChatModel(messages=iter([AIMessage(content="Hello")]))
@@ -35,7 +35,7 @@ class TestBasicOnModelCall:
call_log = []
class LoggingMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
call_log.append("before")
result = handler(request)
call_log.append("after")
@@ -57,7 +57,7 @@ class TestBasicOnModelCall:
super().__init__()
self.call_count = 0
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
self.call_count += 1
return handler(request)
@@ -89,7 +89,7 @@ class TestRetryMiddleware:
super().__init__()
self.retry_count = 0
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
try:
result = handler(request)
return result
@@ -120,7 +120,7 @@ class TestRetryMiddleware:
self.max_retries = max_retries
self.attempts = []
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
last_exception = None
for attempt in range(self.max_retries):
self.attempts.append(attempt + 1)
@@ -151,7 +151,7 @@ class TestResponseRewriting:
"""Test middleware that transforms response to uppercase."""
class UppercaseMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
result = handler(request)
return AIMessage(content=result.content.upper())
@@ -170,7 +170,7 @@ class TestResponseRewriting:
super().__init__()
self.prefix = prefix
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
result = handler(request)
return AIMessage(content=f"{self.prefix}{result.content}")
@@ -193,7 +193,7 @@ class TestErrorHandling:
raise ValueError("Model error")
class ErrorToSuccessMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
try:
return handler(request)
except Exception:
@@ -216,7 +216,7 @@ class TestErrorHandling:
raise ConnectionError("Network error")
class SelectiveErrorMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
try:
return handler(request)
except ConnectionError:
@@ -235,7 +235,7 @@ class TestErrorHandling:
call_log = []
class ErrorRecoveryMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
try:
call_log.append("before-yield")
result = handler(request)
@@ -279,7 +279,7 @@ class TestShortCircuit:
model_calls = []
class CachingMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
# Simple cache key based on last message
cache_key = str(request.messages[-1].content) if request.messages else ""
@@ -335,7 +335,7 @@ class TestRequestModification:
super().__init__()
self.system_prompt = system_prompt
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
# Modify request to add system prompt
modified_request = ModelRequest(
model=request.model,
@@ -345,6 +345,8 @@ class TestRequestModification:
tool_choice=request.tool_choice,
response_format=request.response_format,
model_settings=request.model_settings,
+ state=request.state,
+ runtime=request.runtime,
)
received_requests.append(modified_request)
return handler(modified_request)
@@ -370,12 +372,12 @@ class TestStateAndRuntime:
state_values = []
class StateAwareMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
- # Access state
+ def on_model_call(self, request, handler):
+ # Access state from request
state_values.append(
{
- "thread_model_call_count": state.get("thread_model_call_count", 0),
- "messages_count": len(state.get("messages", [])),
+ "thread_model_call_count": request.state.get("thread_model_call_count", 0),
+ "messages_count": len(request.state.get("messages", [])),
}
)
return handler(request)
@@ -393,7 +395,7 @@ class TestStateAndRuntime:
"""Test middleware that tracks retry count in state."""
class StateTrackingRetryMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
max_retries = 2
for attempt in range(max_retries):
try:
@@ -429,14 +431,14 @@ class TestMiddlewareComposition:
execution_order = []
class OuterMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
execution_order.append("outer-before")
response = handler(request)
execution_order.append("outer-after")
return response
class InnerMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
execution_order.append("inner-before")
response = handler(request)
execution_order.append("inner-after")
@@ -468,14 +470,14 @@ class TestMiddlewareComposition:
return super()._generate(messages, **kwargs)
class LoggingMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
log.append("logging-before")
result = handler(request)
log.append("logging-after")
return result
class RetryMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
log.append("retry-before")
try:
result = handler(request)
@@ -507,12 +509,12 @@ class TestMiddlewareComposition:
"""Test multiple middleware that each transform the response."""
class PrefixMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
result = handler(request)
return AIMessage(content=f"[PREFIX] {result.content}")
class SuffixMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
result = handler(request)
return AIMessage(content=f"{result.content} [SUFFIX]")
@@ -538,7 +540,7 @@ class TestMiddlewareComposition:
return super()._generate(messages, **kwargs)
class RetryMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
try:
result = handler(request)
return result
@@ -547,7 +549,7 @@ class TestMiddlewareComposition:
return result
class UppercaseMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
result = handler(request)
return AIMessage(content=result.content.upper())
@@ -565,21 +567,21 @@ class TestMiddlewareComposition:
execution_order = []
class FirstMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
execution_order.append("first-before")
response = handler(request)
execution_order.append("first-after")
return response
class SecondMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
execution_order.append("second-before")
response = handler(request)
execution_order.append("second-after")
return response
class ThirdMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
execution_order.append("third-before")
response = handler(request)
execution_order.append("third-after")
@@ -609,14 +611,14 @@ class TestMiddlewareComposition:
model_calls = []
class OuterMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
execution_order.append("outer-before")
result = handler(request)
execution_order.append("outer-after")
return result
class MiddleRetryMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
execution_order.append("middle-before")
# Always retry once (call handler twice)
result = handler(request)
@@ -626,7 +628,7 @@ class TestMiddlewareComposition:
return result
class InnerMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
execution_order.append("inner-before")
result = handler(request)
execution_order.append("inner-after")
@@ -671,7 +673,7 @@ class TestAsyncOnModelCall:
log = []
class LoggingMiddleware(AgentMiddleware):
- async def aon_model_call(self, request, state, runtime, handler):
+ async def aon_model_call(self, request, handler):
log.append("before")
result = await handler(request)
log.append("after")
@@ -698,7 +700,7 @@ class TestAsyncOnModelCall:
return await super()._agenerate(messages, **kwargs)
class RetryMiddleware(AgentMiddleware):
- async def aon_model_call(self, request, state, runtime, handler):
+ async def aon_model_call(self, request, handler):
try:
return await handler(request)
except Exception:
@@ -721,7 +723,7 @@ class TestEdgeCases:
modified_messages = []
class RequestModifyingMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
# Add a system message to the request
modified_request = request
modified_messages.append(len(modified_request.messages))
@@ -739,7 +741,7 @@ class TestEdgeCases:
attempts = []
class MultiModelRetryMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
attempts.append("first-attempt")
try:
result = handler(request)
diff --git a/libs/langchain_v1/tests/unit_tests/agents/test_context_editing_middleware.py b/libs/langchain_v1/tests/unit_tests/agents/test_context_editing_middleware.py
index ca612fe572..78af9a7f6e 100644
--- a/libs/langchain_v1/tests/unit_tests/agents/test_context_editing_middleware.py
+++ b/libs/langchain_v1/tests/unit_tests/agents/test_context_editing_middleware.py
@@ -62,6 +62,8 @@ def _make_state_and_request(
tool_choice=None,
tools=[],
response_format=None,
+ state=state,
+ runtime=_fake_runtime(),
model_settings={},
)
return state, request
@@ -80,7 +82,7 @@ def test_no_edit_when_below_trigger() -> None:
edits=[ClearToolUsesEdit(trigger=50)],
)
- result = middleware.modify_model_request(request, state, _fake_runtime())
+ result = middleware.modify_model_request(request)
assert result.messages[0].content == ""
assert result.messages[1].content == "12345"
@@ -108,7 +110,7 @@ def test_clear_tool_outputs_and_inputs() -> None:
)
middleware = ContextEditingMiddleware(edits=[edit])
- result = middleware.modify_model_request(request, state, _fake_runtime())
+ result = middleware.modify_model_request(request)
cleared_ai = result.messages[0]
cleared_tool = result.messages[1]
@@ -156,7 +158,7 @@ def test_respects_keep_last_tool_results() -> None:
token_count_method="model",
)
- result = middleware.modify_model_request(request, state, _fake_runtime())
+ result = middleware.modify_model_request(request)
cleared_messages = [
msg
@@ -200,7 +202,7 @@ def test_exclude_tools_prevents_clearing() -> None:
],
)
- result = middleware.modify_model_request(request, state, _fake_runtime())
+ result = middleware.modify_model_request(request)
search_tool = result.messages[1]
calc_tool = result.messages[3]
diff --git a/libs/langchain_v1/tests/unit_tests/agents/test_handler_composition.py b/libs/langchain_v1/tests/unit_tests/agents/test_handler_composition.py
index 2b7713fcb4..d8835a90be 100644
--- a/libs/langchain_v1/tests/unit_tests/agents/test_handler_composition.py
+++ b/libs/langchain_v1/tests/unit_tests/agents/test_handler_composition.py
@@ -6,9 +6,13 @@ from langchain_core.messages import AIMessage
from langchain.agents.factory import _chain_model_call_handlers
from langchain.agents.middleware.types import ModelRequest
+from typing import cast
+from langgraph.runtime import Runtime
+
def create_test_request(**kwargs):
"""Helper to create a ModelRequest with sensible defaults."""
+
defaults = {
"messages": [],
"model": None,
@@ -16,6 +20,8 @@ def create_test_request(**kwargs):
"tool_choice": None,
"tools": [],
"response_format": None,
+ "state": {},
+ "runtime": cast(Runtime, object()),
}
defaults.update(kwargs)
return ModelRequest(**defaults)
@@ -42,13 +48,13 @@ class TestChainModelCallHandlers:
"""Test basic composition of two handlers."""
execution_order = []
- def outer(request, state, runtime, handler):
+ def outer(request, handler):
execution_order.append("outer-before")
result = handler(request)
execution_order.append("outer-after")
return result
- def inner(request, state, runtime, handler):
+ def inner(request, handler):
execution_order.append("inner-before")
result = handler(request)
execution_order.append("inner-after")
@@ -61,7 +67,7 @@ class TestChainModelCallHandlers:
def mock_base_handler(req):
return AIMessage(content="test")
- result = composed(create_test_request(), None, None, mock_base_handler)
+ result = composed(create_test_request(), mock_base_handler)
assert execution_order == [
"outer-before",
@@ -75,19 +81,19 @@ class TestChainModelCallHandlers:
"""Test composition of three handlers."""
execution_order = []
- def first(request, state, runtime, handler):
+ def first(request, handler):
execution_order.append("first-before")
result = handler(request)
execution_order.append("first-after")
return result
- def second(request, state, runtime, handler):
+ def second(request, handler):
execution_order.append("second-before")
result = handler(request)
execution_order.append("second-after")
return result
- def third(request, state, runtime, handler):
+ def third(request, handler):
execution_order.append("third-before")
result = handler(request)
execution_order.append("third-after")
@@ -99,7 +105,7 @@ class TestChainModelCallHandlers:
def mock_base_handler(req):
return AIMessage(content="test")
- result = composed(create_test_request(), None, None, mock_base_handler)
+ result = composed(create_test_request(), mock_base_handler)
# First wraps second wraps third
assert execution_order == [
@@ -116,10 +122,10 @@ class TestChainModelCallHandlers:
"""Test inner handler retrying before outer sees response."""
inner_attempts = []
- def outer_passthrough(request, state, runtime, handler):
+ def outer_passthrough(request, handler):
return handler(request)
- def inner_with_retry(request, state, runtime, handler):
+ def inner_with_retry(request, handler):
for attempt in range(3):
inner_attempts.append(attempt)
try:
@@ -140,7 +146,7 @@ class TestChainModelCallHandlers:
raise ValueError("fail")
return AIMessage(content="success")
- result = composed(create_test_request(), None, None, mock_base_handler)
+ result = composed(create_test_request(), mock_base_handler)
assert inner_attempts == [0, 1, 2]
assert result.content == "success"
@@ -148,13 +154,13 @@ class TestChainModelCallHandlers:
def test_error_to_success_conversion(self) -> None:
"""Test handler converting error to success response."""
- def outer_error_handler(request, state, runtime, handler):
+ def outer_error_handler(request, handler):
try:
return handler(request)
except Exception:
return AIMessage(content="Fallback response")
- def inner_passthrough(request, state, runtime, handler):
+ def inner_passthrough(request, handler):
return handler(request)
composed = _chain_model_call_handlers([outer_error_handler, inner_passthrough])
@@ -163,7 +169,7 @@ class TestChainModelCallHandlers:
def mock_base_handler(req):
raise ValueError("Model failed")
- result = composed(create_test_request(), None, None, mock_base_handler)
+ result = composed(create_test_request(), mock_base_handler)
assert result.content == "Fallback response"
@@ -171,13 +177,13 @@ class TestChainModelCallHandlers:
"""Test handlers modifying the request."""
requests_seen = []
- def outer_add_context(request, state, runtime, handler):
+ def outer_add_context(request, handler):
modified_request = create_test_request(
messages=[*request.messages], system_prompt="Added by outer"
)
return handler(modified_request)
- def inner_track_request(request, state, runtime, handler):
+ def inner_track_request(request, handler):
requests_seen.append(request.system_prompt)
return handler(request)
@@ -187,7 +193,7 @@ class TestChainModelCallHandlers:
def mock_base_handler(req):
return AIMessage(content="response")
- result = composed(create_test_request(), None, None, mock_base_handler)
+ result = composed(create_test_request(), mock_base_handler)
assert requests_seen == ["Added by outer"]
assert result.content == "response"
@@ -197,14 +203,14 @@ class TestChainModelCallHandlers:
state_values = []
runtime_values = []
- def outer(request, state, runtime, handler):
- state_values.append(("outer", state))
- runtime_values.append(("outer", runtime))
+ def outer(request, handler):
+ state_values.append(("outer", request.state))
+ runtime_values.append(("outer", request.runtime))
return handler(request)
- def inner(request, state, runtime, handler):
- state_values.append(("inner", state))
- runtime_values.append(("inner", runtime))
+ def inner(request, handler):
+ state_values.append(("inner", request.state))
+ runtime_values.append(("inner", request.runtime))
return handler(request)
composed = _chain_model_call_handlers([outer, inner])
@@ -216,7 +222,11 @@ class TestChainModelCallHandlers:
def mock_base_handler(req):
return AIMessage(content="test")
- result = composed(create_test_request(), test_state, test_runtime, mock_base_handler)
+ # Create request with state and runtime
+ test_request = create_test_request()
+ test_request.state = test_state
+ test_request.runtime = test_runtime
+ result = composed(test_request, mock_base_handler)
# Both handlers should see same state and runtime
assert state_values == [("outer", test_state), ("inner", test_state)]
@@ -227,11 +237,11 @@ class TestChainModelCallHandlers:
"""Test handler that retries multiple times."""
call_count = {"value": 0}
- def outer_counts_calls(request, state, runtime, handler):
+ def outer_counts_calls(request, handler):
call_count["value"] += 1
return handler(request)
- def inner_retries(request, state, runtime, handler):
+ def inner_retries(request, handler):
try:
return handler(request)
except ValueError:
@@ -249,7 +259,7 @@ class TestChainModelCallHandlers:
raise ValueError("fail")
return AIMessage(content="ok")
- result = composed(create_test_request(), None, None, mock_base_handler)
+ result = composed(create_test_request(), mock_base_handler)
# Outer called once, inner retried so base handler called twice
assert call_count["value"] == 1
diff --git a/libs/langchain_v1/tests/unit_tests/agents/test_middleware_agent.py b/libs/langchain_v1/tests/unit_tests/agents/test_middleware_agent.py
index 39e62456c3..a22c6be326 100644
--- a/libs/langchain_v1/tests/unit_tests/agents/test_middleware_agent.py
+++ b/libs/langchain_v1/tests/unit_tests/agents/test_middleware_agent.py
@@ -13,6 +13,8 @@ from langchain_core.messages import (
ToolCall,
ToolMessage,
)
+from typing import cast
+
from langchain_core.outputs import ChatGeneration, ChatResult
from langchain_core.tools import InjectedToolCallId, tool
from langgraph.checkpoint.base import BaseCheckpointSaver
@@ -112,7 +114,7 @@ def test_create_agent_diagram(
def before_model(self, state, runtime):
pass
- def modify_model_request(self, request, state, runtime):
+ def modify_model_request(self, request):
pass
def after_model(self, state, runtime):
@@ -122,7 +124,7 @@ def test_create_agent_diagram(
def before_model(self, state, runtime):
pass
- def modify_model_request(self, request, state, runtime):
+ def modify_model_request(self, request):
pass
def after_model(self, state, runtime):
@@ -246,7 +248,7 @@ def test_create_agent_invoke(
def before_model(self, state, runtime):
calls.append("NoopSeven.before_model")
- def modify_model_request(self, request, state, runtime):
+ def modify_model_request(self, request):
calls.append("NoopSeven.modify_model_request")
return request
@@ -257,7 +259,7 @@ def test_create_agent_invoke(
def before_model(self, state, runtime):
calls.append("NoopEight.before_model")
- def modify_model_request(self, request, state, runtime):
+ def modify_model_request(self, request):
calls.append("NoopEight.modify_model_request")
return request
@@ -339,7 +341,7 @@ def test_create_agent_jump(
def before_model(self, state, runtime):
calls.append("NoopSeven.before_model")
- def modify_model_request(self, request, state, runtime):
+ def modify_model_request(self, request):
calls.append("NoopSeven.modify_model_request")
return request
@@ -352,7 +354,7 @@ def test_create_agent_jump(
calls.append("NoopEight.before_model")
return {"jump_to": "end"}
- def modify_model_request(self, request, state, runtime) -> ModelRequest:
+ def modify_model_request(self, request) -> ModelRequest:
calls.append("NoopEight.modify_model_request")
return request
@@ -1012,16 +1014,20 @@ def test_anthropic_prompt_caching_middleware_initialization() -> None:
tool_choice=None,
tools=[],
response_format=None,
+ state={"messages": [HumanMessage("Hello")]},
+ runtime=cast(Runtime, object()),
model_settings={},
)
- assert middleware.modify_model_request(fake_request, {}, None).model_settings == {
+ assert middleware.modify_model_request(fake_request).model_settings == {
"cache_control": {"type": "ephemeral", "ttl": "5m"}
}
def test_anthropic_prompt_caching_middleware_unsupported_model() -> None:
"""Test AnthropicPromptCachingMiddleware with unsupported model."""
+ from typing import cast
+
fake_request = ModelRequest(
model=FakeToolCallingModel(),
messages=[HumanMessage("Hello")],
@@ -1029,6 +1035,8 @@ def test_anthropic_prompt_caching_middleware_unsupported_model() -> None:
tool_choice=None,
tools=[],
response_format=None,
+ state={"messages": [HumanMessage("Hello")]},
+ runtime=cast(Runtime, object()),
model_settings={},
)
@@ -1038,7 +1046,7 @@ def test_anthropic_prompt_caching_middleware_unsupported_model() -> None:
ValueError,
match="AnthropicPromptCachingMiddleware caching middleware only supports Anthropic models. Please install langchain-anthropic.",
):
- middleware.modify_model_request(fake_request, {}, None)
+ middleware.modify_model_request(fake_request)
langchain_anthropic = ModuleType("langchain_anthropic")
@@ -1052,12 +1060,12 @@ def test_anthropic_prompt_caching_middleware_unsupported_model() -> None:
ValueError,
match="AnthropicPromptCachingMiddleware caching middleware only supports Anthropic models, not instances of",
):
- middleware.modify_model_request(fake_request, {}, None)
+ middleware.modify_model_request(fake_request)
middleware = AnthropicPromptCachingMiddleware(unsupported_model_behavior="warn")
with warnings.catch_warnings(record=True) as w:
- result = middleware.modify_model_request(fake_request, {}, None)
+ result = middleware.modify_model_request(fake_request)
assert len(w) == 1
assert (
"AnthropicPromptCachingMiddleware caching middleware only supports Anthropic models. Please install langchain-anthropic."
@@ -1067,7 +1075,7 @@ def test_anthropic_prompt_caching_middleware_unsupported_model() -> None:
with warnings.catch_warnings(record=True) as w:
with patch.dict("sys.modules", {"langchain_anthropic": langchain_anthropic}):
- result = middleware.modify_model_request(fake_request, {}, None)
+ result = middleware.modify_model_request(fake_request)
assert result is fake_request
assert len(w) == 1
assert (
@@ -1077,11 +1085,11 @@ def test_anthropic_prompt_caching_middleware_unsupported_model() -> None:
middleware = AnthropicPromptCachingMiddleware(unsupported_model_behavior="ignore")
- result = middleware.modify_model_request(fake_request, {}, None)
+ result = middleware.modify_model_request(fake_request)
assert result is fake_request
with patch.dict("sys.modules", {"langchain_anthropic": {"ChatAnthropic": object()}}):
- result = middleware.modify_model_request(fake_request, {}, None)
+ result = middleware.modify_model_request(fake_request)
assert result is fake_request
@@ -1304,9 +1312,7 @@ def test_summarization_middleware_full_workflow() -> None:
def test_modify_model_request() -> None:
class ModifyMiddleware(AgentMiddleware):
- def modify_model_request(
- self, request: ModelRequest, state: AgentState, runtime: Runtime
- ) -> ModelRequest:
+ def modify_model_request(self, request: ModelRequest) -> ModelRequest:
request.messages.append(HumanMessage("remember to be nice!"))
return request
@@ -1438,10 +1444,8 @@ def test_runtime_injected_into_middleware() -> None:
assert runtime is not None
return None
- def modify_model_request(
- self, request: ModelRequest, state: AgentState, runtime: Runtime
- ) -> ModelRequest:
- assert runtime is not None
+ def modify_model_request(self, request: ModelRequest) -> ModelRequest:
+ assert request.runtime is not None
return request
def after_model(self, state: AgentState, runtime: Runtime) -> None:
@@ -1536,6 +1540,8 @@ def test_planning_middleware_modify_model_request(original_prompt, expected_prom
middleware = PlanningMiddleware()
model = FakeToolCallingModel()
+ state: PlanningState = {"messages": [HumanMessage(content="Hello")]}
+
request = ModelRequest(
model=model,
system_prompt=original_prompt,
@@ -1543,11 +1549,12 @@ def test_planning_middleware_modify_model_request(original_prompt, expected_prom
tool_choice=None,
tools=[],
response_format=None,
+ state=state,
+ runtime=cast(Runtime, object()),
model_settings={},
)
- state: PlanningState = {"messages": [HumanMessage(content="Hello")]}
- modified_request = middleware.modify_model_request(request, state, None)
+ modified_request = middleware.modify_model_request(request)
assert modified_request.system_prompt.startswith(expected_prompt_prefix)
@@ -1673,7 +1680,7 @@ def test_planning_middleware_custom_system_prompt() -> None:
)
state: PlanningState = {"messages": [HumanMessage(content="Hello")]}
- modified_request = middleware.modify_model_request(request, state, None)
+ modified_request = middleware.modify_model_request(request)
assert modified_request.system_prompt == f"Original prompt\n\n{custom_system_prompt}"
@@ -1698,6 +1705,8 @@ def test_planning_middleware_custom_system_prompt_and_tool_description() -> None
# Verify system prompt
model = FakeToolCallingModel()
+ state: PlanningState = {"messages": [HumanMessage(content="Hello")]}
+
request = ModelRequest(
model=model,
system_prompt=None,
@@ -1705,11 +1714,12 @@ def test_planning_middleware_custom_system_prompt_and_tool_description() -> None
tool_choice=None,
tools=[],
response_format=None,
+ state=state,
+ runtime=cast(Runtime, object()),
model_settings={},
)
- state: PlanningState = {"messages": [HumanMessage(content="Hello")]}
- modified_request = middleware.modify_model_request(request, state, None)
+ modified_request = middleware.modify_model_request(request)
assert modified_request.system_prompt == custom_system_prompt
# Verify tool description
@@ -1979,7 +1989,7 @@ async def test_create_agent_async_invoke() -> None:
async def abefore_model(self, state, runtime) -> None:
calls.append("AsyncMiddleware.abefore_model")
- async def amodify_model_request(self, request, state, runtime) -> ModelRequest:
+ async def amodify_model_request(self, request) -> ModelRequest:
calls.append("AsyncMiddleware.amodify_model_request")
request.messages.append(HumanMessage("async middleware message"))
return request
@@ -2036,7 +2046,7 @@ async def test_create_agent_async_invoke_multiple_middleware() -> None:
async def abefore_model(self, state, runtime) -> None:
calls.append("AsyncMiddlewareOne.abefore_model")
- async def amodify_model_request(self, request, state, runtime) -> ModelRequest:
+ async def amodify_model_request(self, request) -> ModelRequest:
calls.append("AsyncMiddlewareOne.amodify_model_request")
return request
@@ -2047,7 +2057,7 @@ async def test_create_agent_async_invoke_multiple_middleware() -> None:
async def abefore_model(self, state, runtime) -> None:
calls.append("AsyncMiddlewareTwo.abefore_model")
- async def amodify_model_request(self, request, state, runtime) -> ModelRequest:
+ async def amodify_model_request(self, request) -> ModelRequest:
calls.append("AsyncMiddlewareTwo.amodify_model_request")
return request
@@ -2116,7 +2126,7 @@ async def test_create_agent_mixed_sync_async_middleware() -> None:
def before_model(self, state, runtime) -> None:
calls.append("SyncMiddleware.before_model")
- def modify_model_request(self, request, state, runtime) -> ModelRequest:
+ def modify_model_request(self, request) -> ModelRequest:
calls.append("SyncMiddleware.modify_model_request")
return request
@@ -2127,7 +2137,7 @@ async def test_create_agent_mixed_sync_async_middleware() -> None:
async def abefore_model(self, state, runtime) -> None:
calls.append("AsyncMiddleware.abefore_model")
- async def amodify_model_request(self, request, state, runtime) -> ModelRequest:
+ async def amodify_model_request(self, request) -> ModelRequest:
calls.append("AsyncMiddleware.amodify_model_request")
return request
@@ -2179,7 +2189,7 @@ def test_on_model_call_hook() -> None:
super().__init__()
self.retry_count = 0
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
try:
return handler(request)
except Exception:
@@ -2219,7 +2229,7 @@ def test_on_model_call_retry_count() -> None:
super().__init__()
self.attempts = []
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
max_retries = 3
last_exception = None
for attempt in range(max_retries):
@@ -2261,7 +2271,7 @@ def test_on_model_call_no_retry() -> None:
return "failing"
class NoRetryMiddleware(AgentMiddleware):
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
return handler(request)
agent = create_agent(model=FailingModel(), middleware=[NoRetryMiddleware()])
@@ -2378,7 +2388,7 @@ def test_on_model_call_max_attempts() -> None:
self.max_retries = max_retries
self.attempt_count = 0
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
last_exception = None
for attempt in range(self.max_retries):
self.attempt_count += 1
@@ -2432,7 +2442,7 @@ async def test_on_model_call_async() -> None:
super().__init__()
self.retry_count = 0
- async def aon_model_call(self, request, state, runtime, handler):
+ async def aon_model_call(self, request, handler):
try:
return await handler(request)
except Exception:
@@ -2472,7 +2482,7 @@ def test_on_model_call_rewrite_response() -> None:
class ResponseRewriteMiddleware(AgentMiddleware):
"""Middleware that rewrites the response."""
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
result = handler(request)
# Rewrite the response
@@ -2505,7 +2515,7 @@ def test_on_model_call_convert_error_to_response() -> None:
class ErrorToResponseMiddleware(AgentMiddleware):
"""Middleware that converts errors to success responses."""
- def on_model_call(self, request, state, runtime, handler):
+ def on_model_call(self, request, handler):
try:
return handler(request)
except Exception as e:
@@ -2529,7 +2539,7 @@ def test_create_agent_sync_invoke_with_only_async_middleware_raises_error() -> N
"""Test that sync invoke with only async middleware works via run_in_executor."""
class AsyncOnlyMiddleware(AgentMiddleware):
- async def amodify_model_request(self, request, state, runtime) -> ModelRequest:
+ async def amodify_model_request(self, request) -> ModelRequest:
return request
agent = create_agent(
@@ -2557,11 +2567,11 @@ def test_create_agent_sync_invoke_with_mixed_middleware() -> None:
async def abefore_model(self, state, runtime) -> None:
calls.append("MixedMiddleware.abefore_model")
- def modify_model_request(self, request, state, runtime) -> ModelRequest:
+ def modify_model_request(self, request) -> ModelRequest:
calls.append("MixedMiddleware.modify_model_request")
return request
- async def amodify_model_request(self, request, state, runtime) -> ModelRequest:
+ async def amodify_model_request(self, request) -> ModelRequest:
calls.append("MixedMiddleware.amodify_model_request")
return request
diff --git a/libs/langchain_v1/tests/unit_tests/agents/test_middleware_decorators.py b/libs/langchain_v1/tests/unit_tests/agents/test_middleware_decorators.py
index 700f3939cf..ef6a10c1f4 100644
--- a/libs/langchain_v1/tests/unit_tests/agents/test_middleware_decorators.py
+++ b/libs/langchain_v1/tests/unit_tests/agents/test_middleware_decorators.py
@@ -88,9 +88,7 @@ def test_modify_model_request_decorator() -> None:
"""Test modify_model_request decorator with all configuration options."""
@modify_model_request(state_schema=CustomState, tools=[test_tool], name="CustomModifyRequest")
- def custom_modify_request(
- request: ModelRequest, state: CustomState, runtime: Runtime
- ) -> ModelRequest:
+ def custom_modify_request(request: ModelRequest) -> ModelRequest:
request.system_prompt = "Modified"
return request
@@ -108,10 +106,10 @@ def test_modify_model_request_decorator() -> None:
tool_choice=None,
tools=[],
response_format=None,
+ state={"messages": [HumanMessage("Hello")]},
+ runtime=None,
)
- result = custom_modify_request.modify_model_request(
- original_request, {"messages": [HumanMessage("Hello")]}, None
- )
+ result = custom_modify_request.modify_model_request(original_request)
assert result.system_prompt == "Modified"
@@ -125,7 +123,7 @@ def test_all_decorators_integration() -> None:
return None
@modify_model_request
- def track_modify(request: ModelRequest, state: AgentState, runtime: Runtime) -> ModelRequest:
+ def track_modify(request: ModelRequest) -> ModelRequest:
call_order.append("modify")
return request
@@ -151,7 +149,7 @@ def test_decorators_use_function_names_as_default() -> None:
return None
@modify_model_request
- def my_modify_hook(request: ModelRequest, state: AgentState, runtime: Runtime) -> ModelRequest:
+ def my_modify_hook(request: ModelRequest) -> ModelRequest:
return request
@after_model
@@ -273,9 +271,7 @@ def test_async_modify_model_request_decorator() -> None:
"""Test modify_model_request decorator with async function."""
@modify_model_request(state_schema=CustomState, tools=[test_tool], name="AsyncModifyRequest")
- async def async_modify_request(
- request: ModelRequest, state: CustomState, runtime: Runtime
- ) -> ModelRequest:
+ async def async_modify_request(request: ModelRequest) -> ModelRequest:
request.system_prompt = "Modified async"
return request
@@ -297,13 +293,11 @@ def test_mixed_sync_async_decorators() -> None:
return None
@modify_model_request(name="MixedModifyRequest")
- def sync_modify(request: ModelRequest, state: AgentState, runtime: Runtime) -> ModelRequest:
+ def sync_modify(request: ModelRequest) -> ModelRequest:
return request
@modify_model_request(name="MixedModifyRequest")
- async def async_modify(
- request: ModelRequest, state: AgentState, runtime: Runtime
- ) -> ModelRequest:
+ async def async_modify(request: ModelRequest) -> ModelRequest:
return request
# Both should create valid middleware instances
@@ -324,9 +318,7 @@ async def test_async_decorators_integration() -> None:
return None
@modify_model_request
- async def track_async_modify(
- request: ModelRequest, state: AgentState, runtime: Runtime
- ) -> ModelRequest:
+ async def track_async_modify(request: ModelRequest) -> ModelRequest:
call_order.append("async_modify")
return request
@@ -361,16 +353,12 @@ async def test_mixed_sync_async_decorators_integration() -> None:
return None
@modify_model_request
- def track_sync_modify(
- request: ModelRequest, state: AgentState, runtime: Runtime
- ) -> ModelRequest:
+ def track_sync_modify(request: ModelRequest) -> ModelRequest:
call_order.append("sync_modify")
return request
@modify_model_request
- async def track_async_modify(
- request: ModelRequest, state: AgentState, runtime: Runtime
- ) -> ModelRequest:
+ async def track_async_modify(request: ModelRequest) -> ModelRequest:
call_order.append("async_modify")
return request
@@ -579,7 +567,7 @@ def test_dynamic_prompt_decorator() -> None:
"""Test dynamic_prompt decorator with basic usage."""
@dynamic_prompt
- def my_prompt(request: ModelRequest, state: AgentState, runtime: Runtime) -> str:
+ def my_prompt(request: ModelRequest) -> str:
return "Dynamic test prompt"
assert isinstance(my_prompt, AgentMiddleware)
@@ -595,10 +583,10 @@ def test_dynamic_prompt_decorator() -> None:
tool_choice=None,
tools=[],
response_format=None,
+ state={"messages": [HumanMessage("Hello")]},
+ runtime=None,
)
- result = my_prompt.modify_model_request(
- original_request, {"messages": [HumanMessage("Hello")]}, None
- )
+ result = my_prompt.modify_model_request(original_request)
assert result.system_prompt == "Dynamic test prompt"
@@ -606,8 +594,8 @@ def test_dynamic_prompt_uses_state() -> None:
"""Test that dynamic_prompt can use state information."""
@dynamic_prompt
- def custom_prompt(request: ModelRequest, state: AgentState, runtime: Runtime) -> str:
- msg_count = len(state["messages"])
+ def custom_prompt(request: ModelRequest) -> str:
+ msg_count = len(request.state["messages"])
return f"Prompt with {msg_count} messages"
# Verify it uses state correctly
@@ -618,10 +606,10 @@ def test_dynamic_prompt_uses_state() -> None:
tool_choice=None,
tools=[],
response_format=None,
+ state={"messages": [HumanMessage("Hello"), HumanMessage("World")]},
+ runtime=None,
)
- result = custom_prompt.modify_model_request(
- original_request, {"messages": [HumanMessage("Hello"), HumanMessage("World")]}, None
- )
+ result = custom_prompt.modify_model_request(original_request)
assert result.system_prompt == "Prompt with 2 messages"
@@ -631,7 +619,7 @@ def test_dynamic_prompt_integration() -> None:
prompt_calls = 0
@dynamic_prompt
- def context_aware_prompt(request: ModelRequest, state: AgentState, runtime: Runtime) -> str:
+ def context_aware_prompt(request: ModelRequest) -> str:
nonlocal prompt_calls
prompt_calls += 1
return f"you are a helpful assistant."
@@ -649,7 +637,7 @@ async def test_async_dynamic_prompt_decorator() -> None:
"""Test dynamic_prompt decorator with async function."""
@dynamic_prompt
- async def async_prompt(request: ModelRequest, state: AgentState, runtime: Runtime) -> str:
+ async def async_prompt(request: ModelRequest) -> str:
return "Async dynamic prompt"
assert isinstance(async_prompt, AgentMiddleware)
@@ -664,9 +652,7 @@ async def test_async_dynamic_prompt_integration() -> None:
prompt_calls = 0
@dynamic_prompt
- async def async_context_prompt(
- request: ModelRequest, state: AgentState, runtime: Runtime
- ) -> str:
+ async def async_context_prompt(request: ModelRequest) -> str:
nonlocal prompt_calls
prompt_calls += 1
return f"Async assistant."
@@ -683,7 +669,7 @@ def test_dynamic_prompt_overwrites_system_prompt() -> None:
"""Test that dynamic_prompt overwrites the original system_prompt."""
@dynamic_prompt
- def override_prompt(request: ModelRequest, state: AgentState, runtime: Runtime) -> str:
+ def override_prompt(request: ModelRequest) -> str:
return "Overridden prompt."
agent = create_agent(
@@ -701,11 +687,11 @@ def test_dynamic_prompt_multiple_in_sequence() -> None:
"""Test multiple dynamic_prompt decorators in sequence (last wins)."""
@dynamic_prompt
- def first_prompt(request: ModelRequest, state: AgentState, runtime: Runtime) -> str:
+ def first_prompt(request: ModelRequest) -> str:
return "First prompt."
@dynamic_prompt
- def second_prompt(request: ModelRequest, state: AgentState, runtime: Runtime) -> str:
+ def second_prompt(request: ModelRequest) -> str:
return "Second prompt."
# When used together, the last middleware in the list should win
diff --git a/libs/langchain_v1/tests/unit_tests/agents/test_middleware_tools.py b/libs/langchain_v1/tests/unit_tests/agents/test_middleware_tools.py
index d52927e3f4..a1b9f0eb00 100644
--- a/libs/langchain_v1/tests/unit_tests/agents/test_middleware_tools.py
+++ b/libs/langchain_v1/tests/unit_tests/agents/test_middleware_tools.py
@@ -26,9 +26,7 @@ def test_model_request_tools_are_base_tools() -> None:
return f"Result: {expression}"
class RequestCapturingMiddleware(AgentMiddleware):
- def modify_model_request(
- self, request: ModelRequest, state: AgentState, runtime: Runtime
- ) -> ModelRequest:
+ def modify_model_request(self, request: ModelRequest) -> ModelRequest:
captured_requests.append(request)
return request
@@ -70,9 +68,7 @@ def test_middleware_can_modify_tools() -> None:
return "C"
class ToolFilteringMiddleware(AgentMiddleware):
- def modify_model_request(
- self, request: ModelRequest, state: AgentState, runtime: Runtime
- ) -> ModelRequest:
+ def modify_model_request(self, request: ModelRequest) -> ModelRequest:
# Only allow tool_a and tool_b
request.tools = [t for t in request.tools if t.name in ["tool_a", "tool_b"]]
return request
@@ -113,9 +109,7 @@ def test_unknown_tool_raises_error() -> None:
return "unknown"
class BadMiddleware(AgentMiddleware):
- def modify_model_request(
- self, request: ModelRequest, state: AgentState, runtime: Runtime
- ) -> ModelRequest:
+ def modify_model_request(self, request: ModelRequest) -> ModelRequest:
# Add an unknown tool
request.tools = request.tools + [unknown_tool]
return request
@@ -150,11 +144,9 @@ def test_middleware_can_add_and_remove_tools() -> None:
class ConditionalToolMiddleware(AgentMiddleware[AdminState]):
state_schema = AdminState
- def modify_model_request(
- self, request: ModelRequest, state: AdminState, runtime: Runtime
- ) -> ModelRequest:
+ def modify_model_request(self, request: ModelRequest) -> ModelRequest:
# Remove admin_tool if not admin
- if not state.get("is_admin", False):
+ if not request.state.get("is_admin", False):
request.tools = [t for t in request.tools if t.name != "admin_tool"]
return request
@@ -186,9 +178,7 @@ def test_empty_tools_list_is_valid() -> None:
return "result"
class NoToolsMiddleware(AgentMiddleware):
- def modify_model_request(
- self, request: ModelRequest, state: AgentState, runtime: Runtime
- ) -> ModelRequest:
+ def modify_model_request(self, request: ModelRequest) -> ModelRequest:
# Remove all tools
request.tools = []
return request
@@ -227,18 +217,14 @@ def test_tools_preserved_across_multiple_middleware() -> None:
return "C"
class FirstMiddleware(AgentMiddleware):
- def modify_model_request(
- self, request: ModelRequest, state: AgentState, runtime: Runtime
- ) -> ModelRequest:
+ def modify_model_request(self, request: ModelRequest) -> ModelRequest:
modification_order.append([t.name for t in request.tools])
# Remove tool_c
request.tools = [t for t in request.tools if t.name != "tool_c"]
return request
class SecondMiddleware(AgentMiddleware):
- def modify_model_request(
- self, request: ModelRequest, state: AgentState, runtime: Runtime
- ) -> ModelRequest:
+ def modify_model_request(self, request: ModelRequest) -> ModelRequest:
modification_order.append([t.name for t in request.tools])
# Should not see tool_c here
assert all(t.name != "tool_c" for t in request.tools)
diff --git a/libs/langchain_v1/tests/unit_tests/agents/test_response_format.py b/libs/langchain_v1/tests/unit_tests/agents/test_response_format.py
index 4101a03e77..361bdc153a 100644
--- a/libs/langchain_v1/tests/unit_tests/agents/test_response_format.py
+++ b/libs/langchain_v1/tests/unit_tests/agents/test_response_format.py
@@ -726,7 +726,7 @@ class TestDynamicModelWithResponseFormat:
# Create middleware that swaps the model in the request
class ModelSwappingMiddleware(AgentMiddleware):
- def modify_model_request(self, request: ModelRequest, state, runtime) -> ModelRequest:
+ def modify_model_request(self, request: ModelRequest) -> ModelRequest:
# Replace the model with our custom test model
request.model = model
return request