[knowhycodata]

Description

Adds LLMTR (https://llmtr.com) as a built-in OpenAI-compatible gateway provider, following the same pattern as Onomeo and Runcrate. Disclosure: I maintain LLMTR.

• deeptutor/services/provider_registry.py: new llmtr ProviderSpec (openai_compat, gateway, LLMTR_API_KEY, key prefix llmtr-, base keyword llmtr, default base https://llmtr.com/v1). Model ids are not stripped, so prefixed ids like llmtr/qwen3-8-27b work as-is.
• web/lib/provider-branding.ts + web/public/provider-icons/llmtr.png: icon for the Settings provider picker, with its source recorded in sources.json.
• Tests: provider choices, runtime resolution for the llmtr binding, base-URL detection from an openai binding, and inclusion in the native-tools provider list.

Related Issues

• Closes [Feature Request]: Add LLMTR as an OpenAI-compatible LLM provider #2155

Module(s) Affected

• agents
• api
• config
• core
• knowledge
• logging
• services
• tools
• utils
• web (Frontend)
• docs (Documentation)
• scripts
• tests
• Other: ...

Checklist

• This PR targets dev (not main).
• I have read and followed the contribution guidelines.
• My code follows the project's coding standards.
• I have run pre-commit run --all-files and fixed any issues. (Ran ruff check and ruff format --check on the changed files instead; both clean.)
• I have added relevant tests for my changes.
• I have updated the documentation (if necessary). (No docs change needed; the provider shows up in Settings.)
• My changes do not introduce any new security vulnerabilities.

Additional Notes

Tested from a fresh clone of this branch (Python 3.13, requirements/full.txt, editable install):

• pytest tests/api/test_settings_router.py tests/core/test_agentic_client_provider_kwargs.py tests/services/config/test_provider_runtime.py: 176 passed.
• Live check with a real key: deeptutor init --non-interactive, then deeptutor config apply with {"llm": {"provider": "llmtr", "model": "llmtr/qwen3-8-27b", "api_key_env": "LLMTR_API_KEY"}} (validated with --check first), then deeptutor run chat "Reply with exactly: LLMTR_OK" --format json. The chat streamed back LLMTR_OK and completed in one round, about 2.4 s.
• For Qwen model ids the existing heuristic sends extra_body={"enable_thinking": true}; LLMTR accepts it, so no special handling is needed.