langchain - 💡(How to fix) Fix langchain-openrouter: `rs_*` reasoning IDs cause 404 on `/responses` replay across upstreams

• I am a LangChain maintainer, or was asked directly by a LangChain maintainer to create an issue here.

Issue Content

Filed from langchain-ai/deepagents#3174 (closing that as upstream — this is the canonical tracker).

ChatOpenRouter forwards reasoning_content / reasoning_details from assistant additional_kwargs on subsequent turns. Those payloads include opaque rs_* reasoning item IDs. OpenRouter's /responses beta does not propagate store=true / previous_response_id across upstreams, so replayed rs_* IDs 404 regardless of which upstream OpenRouter selects.

Symptoms users hit:

Provider returned error 404
{
  "error": {
    "message": "Item with id 'rs_053a05e24b0da75e0169fa358ea9fc81908b18aff8157798c1' not found. Items are not persisted when `store` is set to false. Try again with `store` set to true, or remove this item from your input.",
    "type": "invalid_request_error",
    "param": "input"
  }
}
provider_name: OpenAI
is_byok: false

The deepagents-side mitigation in langchain-ai/deepagents#3157 set openrouter_provider={"ignore": ["azure"]} on the assumption that Azure's stateless /responses was the offender. Reproduction with that mitigation applied (deepagents 0.5.7, langchain-openrouter 0.2.3) shows the same 404 from OpenAI direct — confirming the root cause is OpenRouter's /responses plumbing, not any single upstream.

Options

1. Strip rs_* reasoning item IDs from outbound messages in langchain-openrouter. Preserve textual reasoning content, drop opaque IDs that the upstream can't resolve. Smallest blast radius; safe for non-OpenRouter callers because the change is package-local.
2. Force chat-completions semantics for OpenRouter requests so the /responses replay path is never taken. Bigger behavior change but sidesteps the problem entirely.
3. Pure docs: confirm with OpenRouter whether state propagation has shipped, and document the constraint. Not a fix.

Option 1 looks like the right starting point. Happy to send a PR if a maintainer wants to weigh in on the approach first.

Cross-refs

• langchain-ai/deepagents#3157 — Azure-specific workaround that masked the broader bug
• langchain-ai/deepagents#3174 — original tracker, closed pointing here

Package (Required)

• langchain-openrouter