Commit Graph

254 Commits

Author SHA1 Message Date
Towseef Altaf
d27fb0c432 feat(langchain,openai): add strict flag to ProviderStrategy structured output (#34149) 2025-12-10 15:35:23 -05:00
Mason Daugherty
7542278997 feat(core,anthropic): extras on BaseTool (#34120) 2025-12-10 09:37:14 -05:00
ccurme
aa0f4fb927 release(langchain): 1.1.3 (#34255) 2025-12-08 14:29:40 -05:00
Sydney Runkle
d18cdc6f32 feat: add agent name to AIMessage (#34254) 2025-12-08 14:23:12 -05:00
Christophe Bornet
a64aee310c chore(core): improve typing of messages utils functions (#34225)
With this we get the correct types for `_runnable_support` annotated
functions.
* return list[BaseMessage] when messages is not None
* return Runnable when messages is None
* typing of function args
2025-12-08 09:59:43 -05:00
Mason Daugherty
4a42158e6c feat(anthropic): add effort support (#34116) 2025-12-05 13:44:42 -05:00
김주호
50e27a447b feat(langchain): add support for Upstage (Solar) in init_chat_model (#34220) 2025-12-05 09:37:37 -05:00
Sydney Runkle
f67af34ea0 release: langchain 1.1.2 (#34210) 2025-12-04 12:57:59 -05:00
Sydney Runkle
3030ffc248 fix: simplify summarization cutoff logic (#34195)
This PR changes how we find the cutoff for summarization, summarizing
content more eagerly if the initial cutoff point isn't safe (ie, would
break apart AI + tool message pairs)

This new algorithm is quite simple - it looks at the initial cutoff
point, if it's not safe, moves forward through the message list until it
finds the first non tool message.

For example:

```
H
AI
TM
--- theoretical cutoff based keep=('messages', 3)
TM
AI
TM
```

```
H
AI
TM
TM
--- actual cutoff, more aggressive summarization
AI
TM
```
2025-12-04 12:44:50 -05:00
Sydney Runkle
1ad9de4b45 release: langchain 1.1.1 (#34206) 2025-12-04 10:46:30 -05:00
William FH
1867521d1a feat: Use uuid7 for run ids (#34172)
Co-authored-by: Sydney Runkle <54324534+sydney-runkle@users.noreply.github.com>
Co-authored-by: Sydney Runkle <sydneymarierunkle@gmail.com>
2025-12-03 10:09:10 -08:00
ccurme
7a2952210e fix(langchain): (SummarizationMiddleware) adjust token counts based on model (#34161) 2025-12-01 16:22:44 +00:00
Mason Daugherty
878f033ed7 docs(langchain): docstrings for summariziation middleware types (#34158)
improving devx :)
2025-12-01 09:39:33 -05:00
Steffen Hausmann
4065106c2e fix(langchain): add types to human_in_the_loop middleware (#34137)
The `HumanInTheLoopMiddleware` is missing a type annotation for the
context schema. Without the fix in this PR, the following code does not
type check:

```
graph = create_agent(
    "gpt-5",
    tools=[send_email_tool, read_email_tool],
    middleware=[
        HumanInTheLoopMiddleware(
            interrupt_on={
                # Require approval or rejection for sending emails
                "send_email_tool": {
                    "allowed_decisions": ["approve", "reject"],
                },
                # Auto-approve reading emails
                "read_email_tool": False,
            }
        ),
    ],
    context_schema=ContextSchema,
)
```

```
Argument of type "list[HumanInTheLoopMiddleware]" cannot be assigned to parameter "middleware" of type "Sequence[AgentMiddleware[StateT_co@create_agent, ContextT@create_agent]]" in function "create_agent"
  "HumanInTheLoopMiddleware" is not assignable to "AgentMiddleware[AgentState[Unknown], ContextSchema | None]"
    Type parameter "ContextT@AgentMiddleware" is invariant, but "None" is not the same as "ContextSchema | None"
```
2025-12-01 08:46:38 -05:00
Mason Daugherty
0a6d01e61d docs(anthropic,core,langchain): updates (#34106) 2025-11-25 17:58:09 -05:00
ccurme
9c21f83e82 release(langchain): 1.1 (#34090) 2025-11-24 10:27:13 -05:00
Sydney Runkle
4ab94579ad feat(langchain): support SystemMessage in create_agent's system_prompt (#34055)
* `create_agent`'s `system_prompt` allows `str | SystemMessage`
* added `system_message: SystemMessage` on `ModelRequest`
* `ModelRequest.system_prompt` is a function of `system_message.text`,
now deprecated
* disallow setting `system_prompt` and `system_message`
* `ModelRequest.system_prompt` can still be set (w/ custom setattr) for
custom backwards compat, but the updates just get propogated to the
`ModelRequest.system_message`

---------

Co-authored-by: Chester Curme <chester.curme@gmail.com>
2025-11-24 14:53:57 +00:00
Sydney Runkle
68ab9a1e56 fix: don't reorder tool calls in HITL middleware (#34023) 2025-11-22 05:10:32 -05:00
Mason Daugherty
47b79c30c0 chore(docs): fix a few refs syntax errors (#34044)
missing whitespace for some admonitions
2025-11-22 00:58:21 -05:00
ccurme
33e5d01f7c feat(model-profiles): distribute data across packages (#34024) 2025-11-21 15:47:05 -05:00
Sydney Runkle
ee3373afc2 chore: add more robust test for runtime injection w/ explicit args_schema (#34051) 2025-11-20 16:51:37 +00:00
Sydney Runkle
b296f103a9 feat: ModelRetryMiddleware (#34027)
Closes https://github.com/langchain-ai/langchain/issues/33983

* Adds `ModelRetryMiddleware` modeled after `ToolRetryMiddleware`
* Uses `on_failure` modes of `error` and `continue` to match the
`exit_behavior` modes of model + tool call limit middleware
* In a backwards compatible manner, aligns the API of
`ToolRetryMiddleware`'s `on_failure` with the above
* Centralize common "retry" utils across these middlewares
2025-11-20 11:42:33 -05:00
Sydney Runkle
b7d1831f9d fix: deprecate setattr on ModelCallRequest (#34022)
* one alternative considered was setting `frozen=True` on the dataclass,
but this is breaking, so a deprecation is a nicer approach
2025-11-19 11:08:55 -05:00
Mason Daugherty
52b1516d44 style(langchain): fix some middleware ref syntax (#33988) 2025-11-16 00:33:17 -05:00
Sydney Runkle
9bd401a6d4 fix: resumable shell, works w/ interrupts (#33978)
fixes https://github.com/langchain-ai/langchain/issues/33684

Now able to run this minimal snippet successfully

```py
import os

from langchain.agents import create_agent
from langchain.agents.middleware import (
    HostExecutionPolicy,
    HumanInTheLoopMiddleware,
    ShellToolMiddleware,
)
from langgraph.checkpoint.memory import InMemorySaver
from langgraph.types import Command


shell_middleware = ShellToolMiddleware(
    workspace_root=os.getcwd(),
    env=os.environ,  # danger
    execution_policy=HostExecutionPolicy()
)

hil_middleware = HumanInTheLoopMiddleware(interrupt_on={"shell": True})

checkpointer = InMemorySaver()

agent = create_agent(
    "openai:gpt-4.1-mini",
    middleware=[shell_middleware, hil_middleware],
    checkpointer=checkpointer,
)

input_message = {"role": "user", "content": "run `which python`"}

config = {"configurable": {"thread_id": "1"}}

result = agent.invoke(
    {"messages": [input_message]},
    config=config,
    durability="exit",
)
```
2025-11-14 15:32:25 -05:00
ccurme
6aa3794b74 feat(langchain): reference model profiles for provider strategy (#33974) 2025-11-14 19:24:18 +00:00
Sydney Runkle
189dcf7295 chore: increase coverage for shell, filesystem, and summarization middleware (#33928)
cc generated, just a start here but wanted to bump things up from 70%
ish
2025-11-14 13:30:36 -05:00
Sydney Runkle
1bc88028e6 fix(anthropic): execute bash + file tools via tool node (#33960)
* use `override` instead of directly patching things on `ModelRequest`
* rely on `ToolNode` for execution of tools related to said middleware,
using `wrap_model_call` to inject the relevant claude tool specs +
allowing tool node to forward them along to corresponding langchain tool
implementations
* making the same change for the native shell tool middleware
* allowing shell tool middleware to specify a name for the shell tool
(negative diff then for claude bash middleware)


long term I think the solution might be to attach metadata to a tool to
map the provider spec to a langchain implementation, which we could also
take some lessons from on the MCP front.
2025-11-14 13:17:01 -05:00
Sydney Runkle
83c078f363 fix: adding missing async hooks (#33957)
* filling in missing async gaps
* using recommended tool runtime injection instead of injected state
  * updating tests to use helper function as well
2025-11-14 09:13:39 -05:00
ZhangShenao
26d39ffc4a docs: Fix doc links (#33964) 2025-11-14 09:07:32 -05:00
Sydney Runkle
9f87b27a5b fix: add filesystem middleware in init (#33955) 2025-11-13 15:07:33 -05:00
Sydney Runkle
2dc1396380 chore(langchain): update deps (#33951) 2025-11-13 14:21:25 -05:00
ccurme
3d415441e8 fix(langchain, openai): backward compat for response_format (#33945) 2025-11-13 11:11:35 -05:00
ccurme
74385e0ebd fix(langchain, openai): fix create_agent / response_format for Responses API (#33939) 2025-11-13 10:18:15 -05:00
Sydney Runkle
637bb1cbbc feat: refactor tests coverage (#33927)
middleware tests have gotten quite unwieldy, major restructuring, sets
the stage for coverage increase

this is super hard to review -- as a proof that we've retained important
tests, I ran coverage on `master` and this branch and confirmed
identical coverage.

* moving all middleware related tests to `agents/middleware` folder
* consolidating related test files
* adding coverage utility to makefile
2025-11-11 10:40:12 -05:00
Mason Daugherty
3dfea96ec1 chore: update README.md files (#33919) 2025-11-10 22:51:35 -05:00
ccurme
68643153e5 feat(langchain): support async summarization in SummarizationMiddleware (#33918) 2025-11-10 15:48:51 -05:00
ccurme
e1adf781c6 feat(langchain): (SummarizationMiddleware) support use of model context windows when triggering summarization (#33825) 2025-11-10 10:08:52 -05:00
Mason Daugherty
1996d81d72 chore(langchain): pass on reference docstrings (middleware) (#33904) 2025-11-09 21:18:28 -05:00
ccurme
e499db4266 release(langchain): 1.0.5 (#33893) 2025-11-07 17:54:43 -05:00
ccurme
f2410f7ea7 revert: Support for SystemMessage in create_agent (#33889)
Reverts langchain-ai/langchain#33640

Introduces lint errors into langchain-anthropic

Should incorporate into 1.1 instead of patch release.
2025-11-07 16:44:11 -05:00
Sydney Runkle
ee630b4539 fix: bump up default recursion limit (#33881)
Fixes https://github.com/langchain-ai/langchain/issues/33740

We don't want to depend on recursion limit here, model call limit
middleware is more appropriate
2025-11-07 13:49:12 -06:00
Mohammad Mohtashim
cf595dcc38 chore(langchain): Support for SystemMessage in create_agent (#33640)
- **Description:** Updated Function Signature of `create_agent`, the
system prompt can be both a list and string. I see no harm in doing
this, since SystemMessage accepts both.
- **Issue:** #33630

---------

Co-authored-by: Sydney Runkle <54324534+sydney-runkle@users.noreply.github.com>
2025-11-07 13:00:38 -06:00
Mahmut CAVDAR
00cc9d421f fix(langchain): Update langchain-core dependency version (#33775) 2025-11-07 10:31:06 -05:00
Mason Daugherty
e023201d42 style: some cleanup (#33857) 2025-11-06 23:50:46 -05:00
Mason Daugherty
d40e340479 chore: attribute package change versions (#33854)
Needed to disambiguate for within inherited docs
2025-11-06 16:57:30 -05:00
Sydney Runkle
9a09ed0659 fix: don't trace conditional edges and no todos in input state (#33842)
while experimenting w/ todo middleware

| Before | After |
|--------|-------|
| ![Screenshot 2025-11-05 at 1 56 21
PM](https://github.com/user-attachments/assets/63195ae4-8122-4662-8246-0fbc16cb1e22)
| ![Screenshot 2025-11-05 at 1 56 03
PM](https://github.com/user-attachments/assets/255e2fa8-e52d-4d1a-949a-33df52ee6668)
|
| Tracing conditional edges (verbose) | Not tracing conditional edges
(cleaner) |
| ![Screenshot 2025-11-05 at 1 57 56
PM](https://github.com/user-attachments/assets/449ccfe9-4c21-4c87-8e0e-6e89d7a97611)
| ![Screenshot 2025-11-05 at 1 56 58
PM](https://github.com/user-attachments/assets/c5c28d0e-2153-4572-af29-b2528761fec6)
|
| Todos in input state (cluttered) | No todos in input state (cleaner) |
2025-11-05 14:25:57 -05:00
Sydney Runkle
7946a8f64e release: langchain v1.0.4 (#33839) 2025-11-05 12:37:58 -05:00
Sydney Runkle
7af79039fc fix: only increment thread count on successful executions (#33837)
* for run count + thread count overflow we should warn model not to call
again
* don't tally mocked tool calls in thread limit -- consider the
following
  * run limit is 1 
  * thread limit is 3
  * first run calls the tool 2 times, 1 executes, 1 is blocked
* we should only count the successful execution above towards the total
thread count
* raise more helpful warnings on invalid config
2025-11-05 10:00:07 -05:00
Sydney Runkle
1755750ca1 fix: more robust tool call limit middleware (#33817)
* improving typing (covariance)
* adding in support for continuing w/ tool calls not yet at threshold,
switching default to continue
* moving all logic into after model

```py
ExitBehavior = Literal["continue", "error", "end"]
"""How to handle execution when tool call limits are exceeded.
- `"continue"`: Block exceeded tools with error messages, let other tools continue (default)
- `"error"`: Raise a `ToolCallLimitExceededError` exception
- `"end"`: Stop execution immediately, injecting a ToolMessage and an AI message
    for the single tool call that exceeded the limit. Raises `NotImplementedError`
    if there are multiple tool calls
"""
```
2025-11-05 09:18:21 -05:00