[jonahgabriel]

OMN-12664: Preserve full Gemini endpoint path in bifrost gateway (omnibase_infra side)

Workstream A of docs/plans/2026-06-03-late-day-contractor-live-path-stabilization.md. Cross-repo with the omnimarket side.

Problem

The cloud-tier Gemini call dropped the registered /v1beta/openai path and posted to /v1/chat/completions, producing a 404. The bifrost gateway fed every backend URL through ModelLlmInferenceRequest.base_url, so the OpenAI-compatible handler (_build_url) appended /v1/chat/completions even when the configured URL was already a complete chat endpoint. Two 404 variants resulted:

• bare-origin base -> https://generativelanguage.googleapis.com/v1/chat/completions
• /v1beta/openai base -> https://generativelanguage.googleapis.com/v1beta/openai/v1/chat/completions

The registered endpoint is https://generativelanguage.googleapis.com/v1beta/openai/chat/completions.

Changes

1. handler_bifrost_gateway.py::_build_inference_request — when a backend's configured base_url is already a complete chat endpoint (ends in /chat/completions or /completions), route it through endpoint_url so _build_url posts it verbatim with no second path append. Structural URL-shape check, not a provider sniff. _build_url itself is unchanged (it was already correct, OMN-10489).
2. config_loader_bifrost.py — the Gemini default now resolves to the full https://generativelanguage.googleapis.com/v1beta/openai/chat/completions endpoint instead of the bare origin. Gemini-only; local and GLM backends are untouched and keep their legacy base_url append path.

No new router/adapter/module that selects behavior by provider (WS-A acceptance + ONEX_CANONICAL_ARCHITECTURE).

dod_evidence

What changed: 2 source files (handler_bifrost_gateway.py, config_loader_bifrost.py) + 2 test files.

Tests (new + updated):

• tests/unit/.../bifrost/test_bifrost_endpoint_preservation.py (new):
  • test_full_gemini_endpoint_routes_through_endpoint_url — full endpoint posts verbatim
  • test_no_double_append_for_full_chat_endpoint — the /v1beta/openai/v1/chat/completions 404 variant is NOT produced
  • test_gemini_openai_base_is_not_a_full_endpoint — documents the 2nd 404 variant (why the default must be the complete endpoint)
  • test_gemini_bare_origin_does_not_truncate_path — 1st 404 variant via legacy path
  • test_local_openai_backend_unaffected, test_local_full_endpoint_routes_through_endpoint_url, test_trailing_slash_full_endpoint_detected — non-Gemini / structural-detection coverage
• tests/unit/.../bifrost/test_config_loader_bifrost.py (updated): Gemini default now carries /v1beta/openai/chat/completions; explicit override preserved; non-Gemini backends unchanged.

Local gates (run in worktree, no -k filter):

• uv run pytest tests/unit/nodes/node_llm_inference_effect/ -> 466 passed
• new + config-loader regression tests -> 19 passed
• uv run ruff format / ruff check --fix src/ tests/ -> clean (warnings are pre-existing in unrelated files)
• uv run mypy <two touched src files> --strict -> Success, no issues
• pre-commit run --files <changed files> -> all hooks Passed
• Full suite uv run pytest tests/ -m "not slow and not integration and not kafka and not postgres and not consul and not chaos and not performance" -> 20998 passed, 2 failed. Both failures are pre-existing and unrelated to this change (neither test file is in this PR's diff): test_node_migration_discovery::test_sync_check_reports_in_sync (local omnimarket migration vendor drift) and test_receipt_gate_install_guard (CI install-script content check).

Transport-mode / URL proof: regression tests assert the resolved POST URL via HandlerLlmOpenaiCompatible._build_url. For the Gemini full endpoint the resolved URL is https://generativelanguage.googleapis.com/v1beta/openai/chat/completions (posted as-is). No live .201/runtime mutation was performed (code+PR only per plan).

Cross-repo

Sibling omnimarket PR: OmniNode-ai/omnimarket#1031

Evidence-Source: OCC#2141
Evidence-Ticket: OMN-12664

[coderabbitai]

Warning

Review limit reached

@jonahgabriel, we couldn't start this review because you've reached your PR review rate limit.

More reviews will be available in 19 minutes and 6 seconds. Learn how PR review limits work.

Your organization has run out of usage credits. Purchase more in the billing tab.

⌛ How to resolve this issue?

After more reviews become available, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

We recommend that you space out your commits to avoid hitting the rate limit.

🚦 How do rate limits work?

CodeRabbit enforces hourly rate limits for each developer per organization.

Our paid plans include higher PR review limits than trial, open-source, and free plans. In all cases, reviews become available again over time. During sustained high-volume PR review activity, CodeRabbit may temporarily slow when the next review becomes available.

Please see our Fair Usage Limits Policy for further information.

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: d2dcc891-956d-491e-ab3b-84b3465a28ef

📥 Commits

Reviewing files that changed from the base of the PR and between d30f05d and b60665c.

📒 Files selected for processing (4)

• src/omnibase_infra/nodes/node_llm_inference_effect/handlers/bifrost/config_loader_bifrost.py
• src/omnibase_infra/nodes/node_llm_inference_effect/handlers/bifrost/handler_bifrost_gateway.py
• tests/unit/nodes/node_llm_inference_effect/handlers/bifrost/test_bifrost_endpoint_preservation.py
• tests/unit/nodes/node_llm_inference_effect/handlers/bifrost/test_config_loader_bifrost.py

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch jonah/omn-12664-bifrost-gemini-endpoint

---

_{Comment @coderabbitai help to get the list of available commands and usage tips.}