feat: add direct endpoint overrides for auxiliary and delegation

Add base_url/api_key overrides for auxiliary tasks and delegation so users can
route those flows straight to a custom OpenAI-compatible endpoint without
having to rely on provider=main or named custom providers.

Also clear gateway session env vars in test isolation so the full suite stays
deterministic when run from a messaging-backed agent session.
This commit is contained in:
teknium1 2026-03-14 20:48:29 -07:00
parent 6c24d76533
commit 9f6bccd76a
12 changed files with 526 additions and 99 deletions

View file

@ -24,9 +24,11 @@ def _clean_env(monkeypatch):
for key in (
"OPENROUTER_API_KEY", "OPENAI_BASE_URL", "OPENAI_API_KEY",
"OPENAI_MODEL", "LLM_MODEL", "NOUS_INFERENCE_BASE_URL",
# Per-task provider/model overrides
# Per-task provider/model/direct-endpoint overrides
"AUXILIARY_VISION_PROVIDER", "AUXILIARY_VISION_MODEL",
"AUXILIARY_VISION_BASE_URL", "AUXILIARY_VISION_API_KEY",
"AUXILIARY_WEB_EXTRACT_PROVIDER", "AUXILIARY_WEB_EXTRACT_MODEL",
"AUXILIARY_WEB_EXTRACT_BASE_URL", "AUXILIARY_WEB_EXTRACT_API_KEY",
"CONTEXT_COMPRESSION_PROVIDER", "CONTEXT_COMPRESSION_MODEL",
):
monkeypatch.delenv(key, raising=False)
@ -142,6 +144,27 @@ class TestGetTextAuxiliaryClient:
call_kwargs = mock_openai.call_args
assert call_kwargs.kwargs["base_url"] == "http://localhost:1234/v1"
def test_task_direct_endpoint_override(self, monkeypatch):
monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
monkeypatch.setenv("AUXILIARY_WEB_EXTRACT_BASE_URL", "http://localhost:2345/v1")
monkeypatch.setenv("AUXILIARY_WEB_EXTRACT_API_KEY", "task-key")
monkeypatch.setenv("AUXILIARY_WEB_EXTRACT_MODEL", "task-model")
with patch("agent.auxiliary_client.OpenAI") as mock_openai:
client, model = get_text_auxiliary_client("web_extract")
assert model == "task-model"
assert mock_openai.call_args.kwargs["base_url"] == "http://localhost:2345/v1"
assert mock_openai.call_args.kwargs["api_key"] == "task-key"
def test_task_direct_endpoint_without_openai_key_does_not_fall_back(self, monkeypatch):
monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
monkeypatch.setenv("AUXILIARY_WEB_EXTRACT_BASE_URL", "http://localhost:2345/v1")
monkeypatch.setenv("AUXILIARY_WEB_EXTRACT_MODEL", "task-model")
with patch("agent.auxiliary_client.OpenAI") as mock_openai:
client, model = get_text_auxiliary_client("web_extract")
assert client is None
assert model is None
mock_openai.assert_not_called()
def test_codex_fallback_when_nothing_else(self, codex_auth_dir):
with patch("agent.auxiliary_client._read_nous_auth", return_value=None), \
patch("agent.auxiliary_client.OpenAI") as mock_openai:
@ -194,6 +217,27 @@ class TestVisionClientFallback:
client, model = get_vision_auxiliary_client()
assert client is not None # Custom endpoint picked up as fallback
def test_vision_direct_endpoint_override(self, monkeypatch):
monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
monkeypatch.setenv("AUXILIARY_VISION_BASE_URL", "http://localhost:4567/v1")
monkeypatch.setenv("AUXILIARY_VISION_API_KEY", "vision-key")
monkeypatch.setenv("AUXILIARY_VISION_MODEL", "vision-model")
with patch("agent.auxiliary_client.OpenAI") as mock_openai:
client, model = get_vision_auxiliary_client()
assert model == "vision-model"
assert mock_openai.call_args.kwargs["base_url"] == "http://localhost:4567/v1"
assert mock_openai.call_args.kwargs["api_key"] == "vision-key"
def test_vision_direct_endpoint_requires_openai_api_key(self, monkeypatch):
monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
monkeypatch.setenv("AUXILIARY_VISION_BASE_URL", "http://localhost:4567/v1")
monkeypatch.setenv("AUXILIARY_VISION_MODEL", "vision-model")
with patch("agent.auxiliary_client.OpenAI") as mock_openai:
client, model = get_vision_auxiliary_client()
assert client is None
assert model is None
mock_openai.assert_not_called()
def test_vision_uses_openrouter_when_available(self, monkeypatch):
monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
with patch("agent.auxiliary_client.OpenAI") as mock_openai:
@ -390,6 +434,24 @@ class TestTaskSpecificOverrides:
client, model = get_text_auxiliary_client("web_extract")
assert model == "google/gemini-3-flash-preview"
def test_task_direct_endpoint_from_config(self, monkeypatch, tmp_path):
hermes_home = tmp_path / "hermes"
hermes_home.mkdir(parents=True, exist_ok=True)
(hermes_home / "config.yaml").write_text(
"""auxiliary:
web_extract:
base_url: http://localhost:3456/v1
api_key: config-key
model: config-model
"""
)
monkeypatch.setenv("HERMES_HOME", str(hermes_home))
with patch("agent.auxiliary_client.OpenAI") as mock_openai:
client, model = get_text_auxiliary_client("web_extract")
assert model == "config-model"
assert mock_openai.call_args.kwargs["base_url"] == "http://localhost:3456/v1"
assert mock_openai.call_args.kwargs["api_key"] == "config-key"
def test_task_without_override_uses_auto(self, monkeypatch):
"""A task with no provider env var falls through to auto chain."""
monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")