Summary

• Fixes [Feature]: openai client cannot get the right model info from local Ollama service #14189.
• Preserves model.default/model.model on plain provider: custom runtime resolution, including credential-pool custom endpoints.
• Carries the runtime model through the gateway bridge and uses it when the session/config model is empty, while removing it from runtime kwargs before constructing AIAgent so callers do not double-pass model.

Root cause

Remote/custom Ollama-compatible endpoints could resolve provider credentials as provider=custom but drop the configured model name before gateway-created agents were constructed. The gateway then fell back to the provider catalog default for custom, which is empty, causing OpenAI-compatible local servers to reject requests with model is required.

Tests

• /Users/stephenyu/Documents/hermes-agent/.venv/bin/python -m pytest tests/hermes_cli/test_runtime_provider_resolution.py::test_plain_custom_runtime_propagates_config_model tests/hermes_cli/test_runtime_provider_resolution.py::test_plain_custom_runtime_propagates_config_model_pool_path tests/gateway/test_session_model_override_routing.py::test_runtime_agent_kwargs_preserves_runtime_model tests/gateway/test_session_model_override_routing.py::test_session_runtime_inherits_custom_model_from_runtime -q -> 4 passed
• /Users/stephenyu/Documents/hermes-agent/.venv/bin/python -m pytest tests/hermes_cli/test_runtime_provider_resolution.py tests/gateway/test_session_model_override_routing.py -q -> 77 passed
• /Users/stephenyu/Documents/hermes-agent/.venv/bin/python -m pytest tests/gateway/test_compress_command.py tests/gateway/test_compress_focus.py tests/gateway/test_session_model_override_routing.py -q -> 8 passed
• git diff --check

Notes

• I also tried /Users/stephenyu/Documents/hermes-agent/.venv/bin/python -m pytest tests/gateway/test_agent_cache.py tests/gateway/test_model_switch_persistence.py -q; the only failure was the existing macOS-local terminal cleanup assertion in tests/gateway/test_agent_cache.py::TestAgentCacheIdleResume::test_close_vs_release_full_teardown_difference, unrelated to this model routing change.