Langsmith `include_model` broken due to deserialization allowlist

[gabrielrfg]

Package

• langchain-core
• langchain-aws

Example Code (Python)

from langsmith import Client

client = Client(api_key=LANGSMITH_API_KEY)
prompt = client.pull_prompt(PROMPT_ID, include_model=True) #Model must be any ChatBedrockConverse in Langsmith

print(prompt.last.bound.invoke("hi"))

Error Message and Stack Trace (if applicable)

ValueError: Deserialization of ('langchain', 'chat_models', 'openai', 'ChatOpenAI') is not allowed. The default (allowed_objects='core') only permits core langchain-core classes. To allow trusted partner integrations, use allowed_objects='all'. Alternatively, pass an explicit list of allowed classes via allowed_objects=[...]. See langchain_core.load.mapping for the full allowlist.

Description

Irrespective of langchain-aws and langsmith versions, pulling ChatBedrockConverse models through langsmith's pull_prompt method with include_model=True is broken from langchain-core>=1.2.5, works perfectly in 1.2.4.

I traced it to #34455. In this PR it seems there was a larger enforcement of the whitelist for partner models/packages. The issue I'm facing is also referenced in langchain-ai/langchain-aws#816, where allowed_objects='all' is used to pass unit tests, however, when pulling using the langsmith client this is not possible to set, resulting in the broken integration.

Looked into the allow list in langchain_core/load/mapping.py and I found the issue: ChatBedrock is allowed, BedrockLLM is allowed, but ChatBedrockConverse is not in the list. I quickly patched this locally and the issue seems to be fully fixed. I'll be creating a PR for this, but would appreciate if anyone could point out anything I might've missed.

[Winds-AI]

Environment

• OS: Ubuntu (x86_64)
• Python: 3.12
• langchain: 1.2.7
• langchain-core: 1.2.7
• langsmith: 0.6.5
• langchain-openai: 1.1.7

Repro

from langsmith import Client

client = Client(api_key=LANGSMITH_API_KEY)
prompt = client.pull_prompt("insight_upsell", include_model=True, secrets_from_env=True)

Expected

- pull_prompt(..., include_model=True) returns a RunnableSequence (prompt + model) without error.

Actual

- Raises KeyError: 'parameters' from langchain_core/load.

Error

KeyError: 'parameters'
  at langchain_core/load/load.py (deserialization)

Prompt creation

- Prompt was created today in the LangSmith UI (not pushed via SDK).
- Model attached in UI: AzureChatOpenAI, deployment gpt-5.2
- extra_headers is empty {} (warning: "extra_headers is not default parameter").

Manifest (from pull_prompt_commit(..., include_model=True) fallback)

{
  "lc": 1,
  "type": "constructor",
  "id": ["langchain", "schema", "runnable", "RunnableSequence"],
  "kwargs": {
    "first": {
      "id": ["langchain_core", "prompts", "structured", "StructuredPrompt"],
      "kwargs": {
        "messages": [
          {
            "id": ["langchain", "prompts", "chat", "SystemMessagePromptTemplate"],
            "kwargs": {
              "prompt": {
                "id": ["langchain", "prompts", "prompt", "PromptTemplate"],
                "kwargs": {
                  "input_variables": ["rep_id", "max_items"],
                  "template_format": "f-string",
                  "template": "...redacted..."
                }
              }
            }
          }
        ],
        "input_variables": ["rep_id", "max_items"],
        "schema_": { ... }
      }
    },
    "last": {
      "id": ["langchain", "schema", "runnable", "RunnableBinding"],
      "kwargs": {
        "bound": {
          "id": ["langchain", "chat_models", "azure_openai", "AzureChatOpenAI"],
          "kwargs": {
            "deployment_name": "gpt-5.2",
            "azure_endpoint": "https://genai.heineken.com",
            "openai_api_version": "2025-03-01-preview",
            "extra_headers": {},
            "openai_api_key": { "id": ["AZURE_OPENAI_API_KEY"], "type": "secret" }
          }
        }
      }
    }
  }
}

Notes / Workaround

- client.pull_prompt(..., include_model=False) works (returns prompt template).
- client.pull_prompt_commit(..., include_model=True) works and returns the manifest (so API is OK).
- The failure appears to be in deserialization of the UI-created manifest via langchain_core.load.

[hshgogogo]

Hi, I reproduced this locally and the root cause looks clear: include_model=True ends up deserializing a model class that is missing from the current allowlist after the stricter load changes. I am working on a small fix plus regression coverage for the LangSmith pull path. If this direction looks good, I can open a PR once the patch is ready.

[gabrielrfg]

Wanted to resume this conversation as this is still not fixed almost 4 months later.
3 different people have opened a PR and it's always marked as duplicate with #34510, which does contain the necessary fix, but comes bundled with SSRF protections, which I'm assuming is why it's taking so long to approve.

My org is still stuck using 1.2.4, which limits the langchain-aws versions we can use, which then limits us to not being able to stream new bedrock llms as these need to be whitelisted in the code.

I don't think this is just a niche case, anyone using any Bedrock model with langsmith is stuck on 1.2.4, and that's if they managed to find this issue and reverted to an old version.

At this point it seems the langsmith / Bedrock combination is completely unsupported, all because we can't get 5 lines of code merged, because someone bundled a simple fix with a more complex validation, which I'm sure is really cool and all, but completely irrelevant if you're not pulling community prompts, which I'm not.

Mason Daugherty (@mdrxy) Would it be somehow possible to get one of the PR's merged (e.g.: #34646 ) and limit the scope of #34510 to just the SSRF?

[talhaanwarch]

I also getting this error with latest langchain version. I am using langsmith to fetch prompt and litellm as proxy

[fairchildadrian9-create]

Hi, +1 on this — still hitting the same limitation. Would it be possible to merge a minimal fix (like #34646) separately and keep #34510 focused on SSRF? This seems to be blocking Bedrock + LangSmith usage for multiple users

…

On Wed, Apr 1, 2026, 2:51 AM Talha Anwar ***@***.***> wrote: *talhaanwarch* left a comment (langchain-ai/langchain#34645) <#34645 (comment)> I also getting this error with latest langchain version. I am using langsmith to fetch prompt and litellm as proxy — Reply to this email directly, view it on GitHub <#34645?email_source=notifications&email_token=B7ZXUCKEYYRIDSCYLA2J4UT4TTRDHA5CNFSNUABFM5UWIORPF5TWS5BNNB2WEL2JONZXKZKDN5WW2ZLOOQXTIMJWHA4DSMJTG422M4TFMFZW63VKON2WE43DOJUWEZLEUVSXMZLOOSWGM33PORSXEX3DNRUWG2Y#issuecomment-4168891375>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/B7ZXUCNUQMIN6ECVFSBFQDT4TTRDHAVCNFSM6AAAAACRAFOLPSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DCNRYHA4TCMZXGU> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>