[Souravrajvi0]

Summary

Fixes #3865

Adds an extra_headers model parameter to the OpenAI API Compatible LLM provider so callers can send per-invocation HTTP headers with Dify-resolved variables.

Use case

OpenCode Go requires a stable session header per conversation:

{
  "x-opencode-session": "{{#sys.conversation_id#}}"
}

Dify resolves workflow/system variables in model parameters at invocation time; this change forwards the resolved headers to the underlying OpenAI-compatible request.

Changes

• Expose extra_headers (JSON string) in the LLM model parameter schema
• Parse and merge model-parameter headers with any credential-level extra_headers
• Forward merged headers on both Chat Completions (requests) and Responses API (OpenAI client) paths
• Add regression tests (8 passing)

Test plan

• uv run pytest models/openai_api_compatible/tests/test_extra_headers.py
• Configure OpenAI API Compatible model with extra_headers using {{#sys.conversation_id#}}

[crazywoola]

Hi @Souravrajvi0, thanks for adding the dynamic extra_headers model parameter!

Heads-up: #3832 also targets models/openai_api_compatible and was closed just now because it has merge conflicts with main. To keep the release queue clean, could you please fold the reasoning_effort forwarding change from #3832 into this PR if it's still needed, and reference its issue #3827 in the description alongside #3865?

One more thing about manifest.yaml: the consolidated dependency refresh (#4106–#4111) bumped models/openai_api_compatible to 0.0.69 on main, and there are several open PRs for this plugin from different authors (#3795, #4060, #4066, #4068, #4079). Please rebase on latest main, add the version bump, and use the next unused version so the releases don't collide.

Thanks again for your contributions! 🙏