ouyangbo/hermes-agent
1
0
Fork 0
Code Issues Pull Requests Actions 9 Packages Projects Releases Wiki Activity

fix(providers): preserve Actual routing during startup and key reload (E-1047)

Browse Source
This commit is contained in:
Justin Bennington
2026-09-10 15:01:17 -04:00
parent 8b2b83a906
commit 4135933ec9
4 changed files with 90 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
agent/agent_init.py
+9
View File
@@ -845,6 +845,10 @@ def _routed_client_kwargs(agent, fallback_model, _provider_timeout) -> Dict[str,
_routed_client, _ = resolve_provider_client(
agent.provider or "auto", model=agent.model, raw_codex=True)
if _routed_client is not None:
from hermes_cli.providers import is_actual_route, normalize_provider
effective_provider = getattr(_routed_client, "_hermes_aux_effective_provider", "")
if is_actual_route(effective_provider):
agent.provider = normalize_provider(effective_provider)
return _client_kwargs_from_routed(_routed_client, _provider_timeout)
# No credentials: try the fallback chain BEFORE failing (an exhausted single-entry pool
# must not die with a misleading "No LLM provider configured"); only explicitly named
@@ -929,6 +933,11 @@ def _init_openai_client(agent, api_key, base_url, fallback_model, _provider_time
client_kwargs = _explicit_client_kwargs(agent, api_key, base_url, _provider_timeout)
else:
client_kwargs = _routed_client_kwargs(agent, fallback_model, _provider_timeout)
from hermes_cli.providers import is_actual_route
if is_actual_route(agent.provider, client_kwargs.get("base_url", "")):
agent.api_mode = "chat_completions"
if hasattr(agent, "_transport_cache"):
agent._transport_cache.clear()
try:
from agent.bedrock_adapter import configure_bedrock_openai_client_kwargs
configure_bedrock_openai_client_kwargs(client_kwargs, timeout=_provider_timeout)
agent/client_lifecycle.py
+6 -1
View File
@@ -677,7 +677,12 @@ class ClientLifecycleMixin:
env_url = get_env_prefer_dotenv(url_var).strip().rstrip("/") if url_var else ""
default_base = (pconfig.inference_base_url or "").strip().rstrip("/")
base_url = env_url or default_base
if self.provider in ("kimi-coding", "zai"):
if self.provider == "actual":
from hermes_cli.auth import normalize_actual_base_url
from hermes_cli.runtime_provider import _config_base_url_for_provider, _get_model_config
configured_base = _config_base_url_for_provider(_get_model_config(), "actual")
base_url = normalize_actual_base_url(configured_base or base_url)
elif self.provider in ("kimi-coding", "zai"):
from hermes_cli import auth as _auth
resolver = _auth._resolve_kimi_base_url if self.provider == "kimi-coding" else _auth._resolve_zai_base_url
base_url = resolver(api_key, pconfig.inference_base_url, env_url).rstrip("/")
hermes_cli/model_setup_flows.py
+2 -1
View File
@@ -933,8 +933,9 @@ def _model_flow_api_key_provider(config, provider_id, current_model=""):
effective_base = current_base or pconfig.inference_base_url
if provider_id == "actual":
from hermes_cli.providers import normalize_provider
model_cfg = config.get("model") or {}
if isinstance(model_cfg, dict) and model_cfg.get("provider") == provider_id:
if isinstance(model_cfg, dict) and normalize_provider(str(model_cfg.get("provider") or "")) == provider_id:
effective_base = str(model_cfg.get("base_url") or "").strip() or effective_base
if provider_id == "zai":
tests/agent/test_actual_auxiliary_routing.py
+73 -8
View File
@@ -232,7 +232,17 @@ def test_actual_background_tasks_reach_chat_completions(
],
)
@pytest.mark.parametrize(
"entrypoint", ["init", "auto", "switch", "fallback", "restore", "rotation"]
"entrypoint",
[
"init",
"auto",
"switch",
"fallback",
"restore",
"rotation",
"init_fallback",
"init_auto",
],
)
def test_actual_runtime_transitions_reach_chat_completions(
tmp_path, monkeypatch, actual_endpoint, provider, hosted, entrypoint
@@ -266,12 +276,20 @@ def test_actual_runtime_transitions_reach_chat_completions(
(tmp_path / "config.yaml").write_text(yaml.safe_dump(config), encoding="utf-8")
runtime = resolve_runtime_provider(requested=provider)
assert runtime["api_mode"] == "chat_completions"
initial_base = base_url.removesuffix("/v1")
initial_provider = provider
initial_key = "actual-test-key"
if entrypoint in {"init_fallback", "init_auto"}:
initial_base = initial_key = None
initial_provider = (
"missing-provider" if entrypoint == "init_fallback" else "auto"
)
elif entrypoint == "rotation":
initial_base = base_url.replace("api.actual.inc", "127.0.0.1")
agent = AIAgent(
provider=provider,
base_url=base_url.replace("api.actual.inc", "127.0.0.1")
if entrypoint == "rotation"
else base_url.removesuffix("/v1"),
api_key="actual-test-key",
provider=initial_provider,
base_url=initial_base,
api_key=initial_key,
api_mode=None if entrypoint == "auto" else "codex_responses",
model="primary-model" if entrypoint == "fallback" else "gpt-5.4",
enabled_toolsets=[],
@@ -386,7 +404,10 @@ def test_actual_auxiliary_fallback_reaches_chat_completions(
@pytest.mark.parametrize("override", ["", "http://127.0.0.1:8081", "invalid-url"])
def test_actual_setup_keeps_provider_settings_in_yaml(tmp_path, monkeypatch, override):
@pytest.mark.parametrize("configured_provider", ["actual", "aci"])
def test_actual_setup_keeps_provider_settings_in_yaml(
tmp_path, monkeypatch, override, configured_provider
):
from hermes_cli import config as config_module
from hermes_cli import model_setup_flows as setup
from hermes_cli.auth import resolve_api_key_provider_credentials
@@ -399,7 +420,7 @@ def test_actual_setup_keeps_provider_settings_in_yaml(tmp_path, monkeypatch, ove
configured_url = "http://127.0.0.1:8080"
raw = {
"model": {
"provider": "actual",
"provider": configured_provider,
"default": "old-model",
"base_url": configured_url,
}
@@ -435,3 +456,47 @@ def test_actual_setup_keeps_provider_settings_in_yaml(tmp_path, monkeypatch, ove
runtime = resolve_runtime_provider(requested="actual", **kwargs)
assert runtime["base_url"] == expected_url + "/v1"
assert runtime["api_mode"] == "chat_completions"
def test_actual_key_reload_keeps_yaml_endpoint(tmp_path, monkeypatch, actual_endpoint):
from run_agent import AIAgent
base_url, requests = actual_endpoint
monkeypatch.setenv("HERMES_HOME", str(tmp_path))
monkeypatch.setenv("ACTUAL_API_KEY", "actual-test-key")
monkeypatch.setenv("ACTUAL_BASE_URL", "http://127.0.0.1:1")
(tmp_path / "config.yaml").write_text(
yaml.safe_dump({
"model": {"provider": "aci", "default": "test-model", "base_url": base_url},
}),
encoding="utf-8",
)
env_path = tmp_path / ".env"
env_path.write_text("ACTUAL_API_KEY=actual-test-key\n", encoding="utf-8")
agent = AIAgent(
provider="actual",
base_url=base_url,
api_key="actual-test-key",
model="test-model",
enabled_toolsets=[],
quiet_mode=True,
skip_context_files=True,
skip_memory=True,
save_trajectories=False,
)
try:
assert not agent._try_refresh_env_client_credentials()
env_path.write_text("ACTUAL_API_KEY=actual-rotated-key\n", encoding="utf-8")
assert agent._try_refresh_env_client_credentials()
assert agent.api_key == "actual-rotated-key"
assert agent.base_url == base_url + "/v1"
assert agent.api_mode == "chat_completions"
response = agent._interruptible_api_call(
agent._build_api_kwargs([{"role": "user", "content": "Reply briefly."}], [])
)
assert response.choices[0].message.content == "The task is complete."
assert [
path for path, body in requests if body.get("model") == "test-model"
] == ["/v1/chat/completions"]
finally:
agent.client.close()