Bug: LM Studio provider sends incompatible response_format JSON Schema (HTTP 400)

[MachineFm]

LM Studio provider always sends response_format, causing HTTP 400 JSON Schema parser error

Environment

• OpenClaw: 2026.7.1-2 (0790d9f)
• OS: Windows 10 Pro 22H2 (19045.6937)
• Node.js: v24.18.0
• Provider: Native LM Studio provider
• LM Studio: 0.4.20
• API Endpoint: http://127.0.0.1:1234/v1

Model:

• lmstudio-community/Qwen3.5-4B-GGUF/Qwen3.5-4B-Q4_K_M.gguf

---

Description

OpenClaw sends a response_format JSON Schema to LM Studio for chat completions.

LM Studio rejects the request with HTTP 400 before inference begins.

The problem disappears immediately if response_format is not sent.

---

Steps to reproduce

Start LM Studio server:

lms server start --port 1234

Configure OpenClaw:

openclaw onboard \
  --non-interactive \
  --accept-risk \
  --auth-choice lmstudio \
  --custom-base-url http://127.0.0.1:1234/v1 \
  --lmstudio-api-key lmstudio \
  --custom-model-id qwen3.5-4b

Send any prompt.

---

Actual result

LM Studio returns:

HTTP 400

Unable to generate parser for this template.
Automatic parser generation failed:

JSON schema conversion failed:
Pattern must start with '^' and end with '$'

LM Studio log:

Engine protocol predict request returned 400:

{
  "error": {
    "code": 400,
    "message": "Unable to generate parser for this template. Automatic parser generation failed: JSON schema conversion failed: Pattern must start with '^' and end with '$'",
    "type": "invalid_request_error"
  }
}

---

Investigation

The request contains response_format.

Inside OpenClaw:

dist/openai-transport-stream-*.js

contains:

if (options?.responseFormat !== void 0)
    params.response_format = options.responseFormat;

Commenting out this line fixes the problem immediately.

After restarting the gateway:

• Chat works normally
• Tool calling still works
• LM Studio no longer returns HTTP 400

---

Expected behavior

The LM Studio provider should either:

• detect that the provider cannot consume the generated JSON Schema and omit response_format, or
• gracefully fall back to plain text, or
• only send response_format when the target provider explicitly supports the generated schema.

---

Additional information

LM Studio also prints:

Reasoning setting 'medium' is not supported.
Supported settings: 'on', 'off'.

This warning is unrelated.

The request only fails because of the transmitted response_format JSON Schema.

---

Suggested fix

The LM Studio provider should avoid attaching response_format by default, or perform capability detection before sending structured outputs.

A provider-specific compatibility flag similar to other provider capability checks would prevent this issue.