Bug type

Behavior bug (incorrect output/state without crash)

Beta release blocker

No

Summary

When calling the /v1/responses API, I pass a stable user value (for example: user: "chat_12_1").

According to the documentation, requests that include a user string should derive a stable session key so that repeated calls can share the same agent session.

However, in my testing, every API call appears to start a new conversation. The model does not retain previous context, even when the same user value is used.

Additionally, it seems that a new /v1/responses request may clear or replace the previous conversation associated with that user. As a result, follow-up questions cannot access earlier messages.

Interestingly, conversations made directly through the web panel continue to preserve context correctly and are not affected.

Steps to reproduce

Steps to Reproduce

1. Call /v1/responses with:

   {
     "user": "chat_12_1",
     "input": "Remember the number 123456."
   }

2. Call /v1/responses again with the same user:

   {
     "user": "chat_12_1",
     "input": "What number did I ask you to remember?"
   }

3. The model does not remember the previous message and behaves as if it is a new conversation.

Expected behavior

Requests using the same user value should share the same agent session and retain conversation context, as described in the documentation.

Actual behavior

Each /v1/responses API call appears to behave like an independent session, while conversations in the web panel retain context correctly.

OpenClaw version

2026.6.8

Operating system

Ubuntu 24

Install method

No response

Model

gpt5.5

Provider / routing chain

openclaw

Additional provider/model setup details

No response

Logs, screenshots, and evidence

Impact and severity

No response

Additional information

No response