Files
hermes-agent/tests/agent/test_auxiliary_client.py

1293 lines
60 KiB
Python
Raw Normal View History

"""Tests for agent.auxiliary_client resolution chain, provider overrides, and model overrides."""
import json
import os
from pathlib import Path
from unittest.mock import patch, MagicMock
import pytest
from agent.auxiliary_client import (
get_text_auxiliary_client,
get_vision_auxiliary_client,
get_available_vision_backends,
resolve_vision_provider_client,
resolve_provider_client,
auxiliary_max_tokens_param,
call_llm,
_read_codex_access_token,
_get_auxiliary_provider,
_get_provider_chain,
_is_payment_error,
_try_payment_fallback,
_resolve_forced_provider,
_resolve_auto,
)
@pytest.fixture(autouse=True)
def _clean_env(monkeypatch):
"""Strip provider env vars so each test starts clean."""
for key in (
"OPENROUTER_API_KEY", "OPENAI_BASE_URL", "OPENAI_API_KEY",
"OPENAI_MODEL", "LLM_MODEL", "NOUS_INFERENCE_BASE_URL",
"ANTHROPIC_API_KEY", "ANTHROPIC_TOKEN", "CLAUDE_CODE_OAUTH_TOKEN",
# 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)
@pytest.fixture
def codex_auth_dir(tmp_path, monkeypatch):
"""Provide a writable ~/.codex/ directory with a valid auth.json."""
codex_dir = tmp_path / ".codex"
codex_dir.mkdir()
auth_file = codex_dir / "auth.json"
auth_file.write_text(json.dumps({
"tokens": {
"access_token": "codex-test-token-abc123",
"refresh_token": "codex-refresh-xyz",
}
}))
monkeypatch.setattr(
"agent.auxiliary_client._read_codex_access_token",
lambda: "codex-test-token-abc123",
)
return codex_dir
class TestReadCodexAccessToken:
def test_valid_auth_store(self, tmp_path, monkeypatch):
hermes_home = tmp_path / "hermes"
hermes_home.mkdir(parents=True, exist_ok=True)
(hermes_home / "auth.json").write_text(json.dumps({
"version": 1,
"providers": {
"openai-codex": {
"tokens": {"access_token": "tok-123", "refresh_token": "r-456"},
},
},
}))
monkeypatch.setenv("HERMES_HOME", str(hermes_home))
result = _read_codex_access_token()
assert result == "tok-123"
def test_missing_returns_none(self, tmp_path, monkeypatch):
hermes_home = tmp_path / "hermes"
hermes_home.mkdir(parents=True, exist_ok=True)
(hermes_home / "auth.json").write_text(json.dumps({"version": 1, "providers": {}}))
monkeypatch.setenv("HERMES_HOME", str(hermes_home))
result = _read_codex_access_token()
assert result is None
def test_empty_token_returns_none(self, tmp_path, monkeypatch):
hermes_home = tmp_path / "hermes"
hermes_home.mkdir(parents=True, exist_ok=True)
(hermes_home / "auth.json").write_text(json.dumps({
"version": 1,
"providers": {
"openai-codex": {
"tokens": {"access_token": " ", "refresh_token": "r"},
},
},
}))
monkeypatch.setenv("HERMES_HOME", str(hermes_home))
result = _read_codex_access_token()
assert result is None
def test_malformed_json_returns_none(self, tmp_path):
codex_dir = tmp_path / ".codex"
codex_dir.mkdir()
(codex_dir / "auth.json").write_text("{bad json")
with patch("agent.auxiliary_client.Path.home", return_value=tmp_path):
result = _read_codex_access_token()
assert result is None
def test_missing_tokens_key_returns_none(self, tmp_path):
codex_dir = tmp_path / ".codex"
codex_dir.mkdir()
(codex_dir / "auth.json").write_text(json.dumps({"other": "data"}))
with patch("agent.auxiliary_client.Path.home", return_value=tmp_path):
result = _read_codex_access_token()
assert result is None
def test_expired_jwt_returns_none(self, tmp_path, monkeypatch):
"""Expired JWT tokens should be skipped so auto chain continues."""
import base64
import time as _time
# Build a JWT with exp in the past
header = base64.urlsafe_b64encode(b'{"alg":"RS256","typ":"JWT"}').rstrip(b"=").decode()
payload_data = json.dumps({"exp": int(_time.time()) - 3600}).encode()
payload = base64.urlsafe_b64encode(payload_data).rstrip(b"=").decode()
expired_jwt = f"{header}.{payload}.fakesig"
hermes_home = tmp_path / "hermes"
hermes_home.mkdir(parents=True, exist_ok=True)
(hermes_home / "auth.json").write_text(json.dumps({
"version": 1,
"providers": {
"openai-codex": {
"tokens": {"access_token": expired_jwt, "refresh_token": "r"},
},
},
}))
monkeypatch.setenv("HERMES_HOME", str(hermes_home))
result = _read_codex_access_token()
assert result is None, "Expired JWT should return None"
def test_valid_jwt_returns_token(self, tmp_path, monkeypatch):
"""Non-expired JWT tokens should be returned."""
import base64
import time as _time
header = base64.urlsafe_b64encode(b'{"alg":"RS256","typ":"JWT"}').rstrip(b"=").decode()
payload_data = json.dumps({"exp": int(_time.time()) + 3600}).encode()
payload = base64.urlsafe_b64encode(payload_data).rstrip(b"=").decode()
valid_jwt = f"{header}.{payload}.fakesig"
hermes_home = tmp_path / "hermes"
hermes_home.mkdir(parents=True, exist_ok=True)
(hermes_home / "auth.json").write_text(json.dumps({
"version": 1,
"providers": {
"openai-codex": {
"tokens": {"access_token": valid_jwt, "refresh_token": "r"},
},
},
}))
monkeypatch.setenv("HERMES_HOME", str(hermes_home))
result = _read_codex_access_token()
assert result == valid_jwt
def test_non_jwt_token_passes_through(self, tmp_path, monkeypatch):
"""Non-JWT tokens (no dots) should be returned as-is."""
hermes_home = tmp_path / "hermes"
hermes_home.mkdir(parents=True, exist_ok=True)
(hermes_home / "auth.json").write_text(json.dumps({
"version": 1,
"providers": {
"openai-codex": {
"tokens": {"access_token": "plain-token-no-jwt", "refresh_token": "r"},
},
},
}))
monkeypatch.setenv("HERMES_HOME", str(hermes_home))
result = _read_codex_access_token()
assert result == "plain-token-no-jwt"
class TestAnthropicOAuthFlag:
"""Test that OAuth tokens get is_oauth=True in auxiliary Anthropic client."""
def test_oauth_token_sets_flag(self, monkeypatch):
"""OAuth tokens (sk-ant-oat01-*) should create client with is_oauth=True."""
monkeypatch.setenv("ANTHROPIC_TOKEN", "sk-ant-oat01-test-token")
with patch("agent.anthropic_adapter.build_anthropic_client") as mock_build:
mock_build.return_value = MagicMock()
from agent.auxiliary_client import _try_anthropic, AnthropicAuxiliaryClient
client, model = _try_anthropic()
assert client is not None
assert isinstance(client, AnthropicAuxiliaryClient)
# The adapter inside should have is_oauth=True
adapter = client.chat.completions
assert adapter._is_oauth is True
def test_api_key_no_oauth_flag(self, monkeypatch):
"""Regular API keys (sk-ant-api-*) should create client with is_oauth=False."""
with patch("agent.anthropic_adapter.resolve_anthropic_token", return_value="sk-ant-api03-testkey1234"), \
feat(auth): same-provider credential pools with rotation, custom endpoint support, and interactive CLI (#2647) * feat(auth): add same-provider credential pools and rotation UX Add same-provider credential pooling so Hermes can rotate across multiple credentials for a single provider, recover from exhausted credentials without jumping providers immediately, and configure that behavior directly in hermes setup. - agent/credential_pool.py: persisted per-provider credential pools - hermes auth add/list/remove/reset CLI commands - 429/402/401 recovery with pool rotation in run_agent.py - Setup wizard integration for pool strategy configuration - Auto-seeding from env vars and existing OAuth state Co-authored-by: kshitijk4poor <82637225+kshitijk4poor@users.noreply.github.com> Salvaged from PR #2647 * fix(tests): prevent pool auto-seeding from host env in credential pool tests Tests for non-pool Anthropic paths and auth remove were failing when host env vars (ANTHROPIC_API_KEY) or file-backed OAuth credentials were present. The pool auto-seeding picked these up, causing unexpected pool entries in tests. - Mock _select_pool_entry in auxiliary_client OAuth flag tests - Clear Anthropic env vars and mock _seed_from_singletons in auth remove test * feat(auth): add thread safety, least_used strategy, and request counting - Add threading.Lock to CredentialPool for gateway thread safety (concurrent requests from multiple gateway sessions could race on pool state mutations without this) - Add 'least_used' rotation strategy that selects the credential with the lowest request_count, distributing load more evenly - Add request_count field to PooledCredential for usage tracking - Add mark_used() method to increment per-credential request counts - Wrap select(), mark_exhausted_and_rotate(), and try_refresh_current() with lock acquisition - Add tests: least_used selection, mark_used counting, concurrent thread safety (4 threads × 20 selects with no corruption) * feat(auth): add interactive mode for bare 'hermes auth' command When 'hermes auth' is called without a subcommand, it now launches an interactive wizard that: 1. Shows full credential pool status across all providers 2. Offers a menu: add, remove, reset cooldowns, set strategy 3. For OAuth-capable providers (anthropic, nous, openai-codex), the add flow explicitly asks 'API key or OAuth login?' — making it clear that both auth types are supported for the same provider 4. Strategy picker shows all 4 options (fill_first, round_robin, least_used, random) with the current selection marked 5. Remove flow shows entries with indices for easy selection The subcommand paths (hermes auth add/list/remove/reset) still work exactly as before for scripted/non-interactive use. * fix(tests): update runtime_provider tests for config.yaml source of truth (#4165) Tests were using OPENAI_BASE_URL env var which is no longer consulted after #4165. Updated to use model config (provider, base_url, api_key) which is the new single source of truth for custom endpoint URLs. * feat(auth): support custom endpoint credential pools keyed by provider name Custom OpenAI-compatible endpoints all share provider='custom', making the provider-keyed pool useless. Now pools for custom endpoints are keyed by 'custom:<normalized_name>' where the name comes from the custom_providers config list (auto-generated from URL hostname). - Pool key format: 'custom:together.ai', 'custom:local-(localhost:8080)' - load_pool('custom:name') seeds from custom_providers api_key AND model.api_key when base_url matches - hermes auth add/list now shows custom endpoints alongside registry providers - _resolve_openrouter_runtime and _resolve_named_custom_runtime check pool before falling back to single config key - 6 new tests covering custom pool keying, seeding, and listing * docs: add Excalidraw diagram of full credential pool flow Comprehensive architecture diagram showing: - Credential sources (env vars, auth.json OAuth, config.yaml, CLI) - Pool storage and auto-seeding - Runtime resolution paths (registry, custom, OpenRouter) - Error recovery (429 retry-then-rotate, 402 immediate, 401 refresh) - CLI management commands and strategy configuration Open at: https://excalidraw.com/#json=2Ycqhqpi6f12E_3ITyiwh,c7u9jSt5BwrmiVzHGbm87g * fix(tests): update setup wizard pool tests for unified select_provider_and_model flow The setup wizard now delegates to select_provider_and_model() instead of using its own prompt_choice-based provider picker. Tests needed: - Mock select_provider_and_model as no-op (provider pre-written to config) - Call _stub_tts BEFORE custom prompt_choice mock (it overwrites it) - Pre-write model.provider to config so the pool step is reached * docs: add comprehensive credential pool documentation - New page: website/docs/user-guide/features/credential-pools.md Full guide covering quick start, CLI commands, rotation strategies, error recovery, custom endpoint pools, auto-discovery, thread safety, architecture, and storage format. - Updated fallback-providers.md to reference credential pools as the first layer of resilience (same-provider rotation before cross-provider) - Added hermes auth to CLI commands reference with usage examples - Added credential_pool_strategies to configuration guide * chore: remove excalidraw diagram from repo (external link only) * refactor: simplify credential pool code — extract helpers, collapse extras, dedup patterns - _load_config_safe(): replace 4 identical try/except/import blocks - _iter_custom_providers(): shared generator for custom provider iteration - PooledCredential.extra dict: collapse 11 round-trip-only fields (token_type, scope, client_id, portal_base_url, obtained_at, expires_in, agent_key_id, agent_key_expires_in, agent_key_reused, agent_key_obtained_at, tls) into a single extra dict with __getattr__ for backward-compatible access - _available_entries(): shared exhaustion-check between select and peek - Dedup anthropic OAuth seeding (hermes_pkce + claude_code identical) - SimpleNamespace replaces class _Args boilerplate in auth_commands - _try_resolve_from_custom_pool(): shared pool-check in runtime_provider Net -17 lines. All 383 targeted tests pass. --------- Co-authored-by: kshitijk4poor <82637225+kshitijk4poor@users.noreply.github.com>
2026-03-31 03:10:01 -07:00
patch("agent.anthropic_adapter.build_anthropic_client") as mock_build, \
patch("agent.auxiliary_client._select_pool_entry", return_value=(False, None)):
mock_build.return_value = MagicMock()
from agent.auxiliary_client import _try_anthropic, AnthropicAuxiliaryClient
client, model = _try_anthropic()
assert client is not None
assert isinstance(client, AnthropicAuxiliaryClient)
adapter = client.chat.completions
assert adapter._is_oauth is False
feat(auth): same-provider credential pools with rotation, custom endpoint support, and interactive CLI (#2647) * feat(auth): add same-provider credential pools and rotation UX Add same-provider credential pooling so Hermes can rotate across multiple credentials for a single provider, recover from exhausted credentials without jumping providers immediately, and configure that behavior directly in hermes setup. - agent/credential_pool.py: persisted per-provider credential pools - hermes auth add/list/remove/reset CLI commands - 429/402/401 recovery with pool rotation in run_agent.py - Setup wizard integration for pool strategy configuration - Auto-seeding from env vars and existing OAuth state Co-authored-by: kshitijk4poor <82637225+kshitijk4poor@users.noreply.github.com> Salvaged from PR #2647 * fix(tests): prevent pool auto-seeding from host env in credential pool tests Tests for non-pool Anthropic paths and auth remove were failing when host env vars (ANTHROPIC_API_KEY) or file-backed OAuth credentials were present. The pool auto-seeding picked these up, causing unexpected pool entries in tests. - Mock _select_pool_entry in auxiliary_client OAuth flag tests - Clear Anthropic env vars and mock _seed_from_singletons in auth remove test * feat(auth): add thread safety, least_used strategy, and request counting - Add threading.Lock to CredentialPool for gateway thread safety (concurrent requests from multiple gateway sessions could race on pool state mutations without this) - Add 'least_used' rotation strategy that selects the credential with the lowest request_count, distributing load more evenly - Add request_count field to PooledCredential for usage tracking - Add mark_used() method to increment per-credential request counts - Wrap select(), mark_exhausted_and_rotate(), and try_refresh_current() with lock acquisition - Add tests: least_used selection, mark_used counting, concurrent thread safety (4 threads × 20 selects with no corruption) * feat(auth): add interactive mode for bare 'hermes auth' command When 'hermes auth' is called without a subcommand, it now launches an interactive wizard that: 1. Shows full credential pool status across all providers 2. Offers a menu: add, remove, reset cooldowns, set strategy 3. For OAuth-capable providers (anthropic, nous, openai-codex), the add flow explicitly asks 'API key or OAuth login?' — making it clear that both auth types are supported for the same provider 4. Strategy picker shows all 4 options (fill_first, round_robin, least_used, random) with the current selection marked 5. Remove flow shows entries with indices for easy selection The subcommand paths (hermes auth add/list/remove/reset) still work exactly as before for scripted/non-interactive use. * fix(tests): update runtime_provider tests for config.yaml source of truth (#4165) Tests were using OPENAI_BASE_URL env var which is no longer consulted after #4165. Updated to use model config (provider, base_url, api_key) which is the new single source of truth for custom endpoint URLs. * feat(auth): support custom endpoint credential pools keyed by provider name Custom OpenAI-compatible endpoints all share provider='custom', making the provider-keyed pool useless. Now pools for custom endpoints are keyed by 'custom:<normalized_name>' where the name comes from the custom_providers config list (auto-generated from URL hostname). - Pool key format: 'custom:together.ai', 'custom:local-(localhost:8080)' - load_pool('custom:name') seeds from custom_providers api_key AND model.api_key when base_url matches - hermes auth add/list now shows custom endpoints alongside registry providers - _resolve_openrouter_runtime and _resolve_named_custom_runtime check pool before falling back to single config key - 6 new tests covering custom pool keying, seeding, and listing * docs: add Excalidraw diagram of full credential pool flow Comprehensive architecture diagram showing: - Credential sources (env vars, auth.json OAuth, config.yaml, CLI) - Pool storage and auto-seeding - Runtime resolution paths (registry, custom, OpenRouter) - Error recovery (429 retry-then-rotate, 402 immediate, 401 refresh) - CLI management commands and strategy configuration Open at: https://excalidraw.com/#json=2Ycqhqpi6f12E_3ITyiwh,c7u9jSt5BwrmiVzHGbm87g * fix(tests): update setup wizard pool tests for unified select_provider_and_model flow The setup wizard now delegates to select_provider_and_model() instead of using its own prompt_choice-based provider picker. Tests needed: - Mock select_provider_and_model as no-op (provider pre-written to config) - Call _stub_tts BEFORE custom prompt_choice mock (it overwrites it) - Pre-write model.provider to config so the pool step is reached * docs: add comprehensive credential pool documentation - New page: website/docs/user-guide/features/credential-pools.md Full guide covering quick start, CLI commands, rotation strategies, error recovery, custom endpoint pools, auto-discovery, thread safety, architecture, and storage format. - Updated fallback-providers.md to reference credential pools as the first layer of resilience (same-provider rotation before cross-provider) - Added hermes auth to CLI commands reference with usage examples - Added credential_pool_strategies to configuration guide * chore: remove excalidraw diagram from repo (external link only) * refactor: simplify credential pool code — extract helpers, collapse extras, dedup patterns - _load_config_safe(): replace 4 identical try/except/import blocks - _iter_custom_providers(): shared generator for custom provider iteration - PooledCredential.extra dict: collapse 11 round-trip-only fields (token_type, scope, client_id, portal_base_url, obtained_at, expires_in, agent_key_id, agent_key_expires_in, agent_key_reused, agent_key_obtained_at, tls) into a single extra dict with __getattr__ for backward-compatible access - _available_entries(): shared exhaustion-check between select and peek - Dedup anthropic OAuth seeding (hermes_pkce + claude_code identical) - SimpleNamespace replaces class _Args boilerplate in auth_commands - _try_resolve_from_custom_pool(): shared pool-check in runtime_provider Net -17 lines. All 383 targeted tests pass. --------- Co-authored-by: kshitijk4poor <82637225+kshitijk4poor@users.noreply.github.com>
2026-03-31 03:10:01 -07:00
def test_pool_entry_takes_priority_over_legacy_resolution(self):
class _Entry:
access_token = "sk-ant-oat01-pooled"
base_url = "https://api.anthropic.com"
class _Pool:
def has_credentials(self):
return True
def select(self):
return _Entry()
with (
patch("agent.auxiliary_client.load_pool", return_value=_Pool()),
patch("agent.anthropic_adapter.resolve_anthropic_token", side_effect=AssertionError("legacy path should not run")),
patch("agent.anthropic_adapter.build_anthropic_client", return_value=MagicMock()) as mock_build,
):
from agent.auxiliary_client import _try_anthropic
client, model = _try_anthropic()
assert client is not None
assert model == "claude-haiku-4-5-20251001"
assert mock_build.call_args.args[0] == "sk-ant-oat01-pooled"
class TestExpiredCodexFallback:
"""Test that expired Codex tokens don't block the auto chain."""
def test_expired_codex_falls_through_to_next(self, tmp_path, monkeypatch):
"""When Codex token is expired, auto chain should skip it and try next provider."""
import base64
import time as _time
# Expired Codex JWT
header = base64.urlsafe_b64encode(b'{"alg":"RS256","typ":"JWT"}').rstrip(b"=").decode()
payload_data = json.dumps({"exp": int(_time.time()) - 3600}).encode()
payload = base64.urlsafe_b64encode(payload_data).rstrip(b"=").decode()
expired_jwt = f"{header}.{payload}.fakesig"
hermes_home = tmp_path / "hermes"
hermes_home.mkdir(parents=True, exist_ok=True)
(hermes_home / "auth.json").write_text(json.dumps({
"version": 1,
"providers": {
"openai-codex": {
"tokens": {"access_token": expired_jwt, "refresh_token": "r"},
},
},
}))
monkeypatch.setenv("HERMES_HOME", str(hermes_home))
# Set up Anthropic as fallback
monkeypatch.setenv("ANTHROPIC_TOKEN", "sk-ant-oat01-test-fallback")
with patch("agent.anthropic_adapter.build_anthropic_client") as mock_build:
mock_build.return_value = MagicMock()
from agent.auxiliary_client import _resolve_auto, AnthropicAuxiliaryClient
client, model = _resolve_auto()
# Should NOT be Codex, should be Anthropic (or another available provider)
assert not isinstance(client, type(None)), "Should find a provider after expired Codex"
def test_expired_codex_openrouter_wins(self, tmp_path, monkeypatch):
"""With expired Codex + OpenRouter key, OpenRouter should win (1st in chain)."""
import base64
import time as _time
header = base64.urlsafe_b64encode(b'{"alg":"RS256","typ":"JWT"}').rstrip(b"=").decode()
payload_data = json.dumps({"exp": int(_time.time()) - 3600}).encode()
payload = base64.urlsafe_b64encode(payload_data).rstrip(b"=").decode()
expired_jwt = f"{header}.{payload}.fakesig"
hermes_home = tmp_path / "hermes"
hermes_home.mkdir(parents=True, exist_ok=True)
(hermes_home / "auth.json").write_text(json.dumps({
"version": 1,
"providers": {
"openai-codex": {
"tokens": {"access_token": expired_jwt, "refresh_token": "r"},
},
},
}))
monkeypatch.setenv("HERMES_HOME", str(hermes_home))
monkeypatch.setenv("OPENROUTER_API_KEY", "or-test-key")
with patch("agent.auxiliary_client.OpenAI") as mock_openai:
mock_openai.return_value = MagicMock()
from agent.auxiliary_client import _resolve_auto
client, model = _resolve_auto()
assert client is not None
# OpenRouter is 1st in chain, should win
mock_openai.assert_called()
def test_expired_codex_custom_endpoint_wins(self, tmp_path, monkeypatch):
"""With expired Codex + custom endpoint (Ollama), custom should win (3rd in chain)."""
import base64
import time as _time
header = base64.urlsafe_b64encode(b'{"alg":"RS256","typ":"JWT"}').rstrip(b"=").decode()
payload_data = json.dumps({"exp": int(_time.time()) - 3600}).encode()
payload = base64.urlsafe_b64encode(payload_data).rstrip(b"=").decode()
expired_jwt = f"{header}.{payload}.fakesig"
hermes_home = tmp_path / "hermes"
hermes_home.mkdir(parents=True, exist_ok=True)
(hermes_home / "auth.json").write_text(json.dumps({
"version": 1,
"providers": {
"openai-codex": {
"tokens": {"access_token": expired_jwt, "refresh_token": "r"},
},
},
}))
monkeypatch.setenv("HERMES_HOME", str(hermes_home))
# Simulate Ollama or custom endpoint
with patch("agent.auxiliary_client._resolve_custom_runtime",
return_value=("http://localhost:11434/v1", "sk-dummy")):
with patch("agent.auxiliary_client.OpenAI") as mock_openai:
mock_openai.return_value = MagicMock()
from agent.auxiliary_client import _resolve_auto
client, model = _resolve_auto()
assert client is not None
def test_hermes_oauth_file_sets_oauth_flag(self, monkeypatch):
"""OAuth-style tokens should get is_oauth=*** (token is not sk-ant-api-*)."""
# Mock resolve_anthropic_token to return an OAuth-style token
with patch("agent.anthropic_adapter.resolve_anthropic_token", return_value="hermes-oauth-jwt-token"), \
patch("agent.anthropic_adapter.build_anthropic_client") as mock_build, \
patch("agent.auxiliary_client._select_pool_entry", return_value=(False, None)):
mock_build.return_value = MagicMock()
from agent.auxiliary_client import _try_anthropic, AnthropicAuxiliaryClient
client, model = _try_anthropic()
assert client is not None, "Should resolve token"
adapter = client.chat.completions
assert adapter._is_oauth is True, "Non-sk-ant-api token should set is_oauth=True"
def test_jwt_missing_exp_passes_through(self, tmp_path, monkeypatch):
"""JWT with valid JSON but no exp claim should pass through."""
import base64
header = base64.urlsafe_b64encode(b'{"alg":"RS256","typ":"JWT"}').rstrip(b"=").decode()
payload_data = json.dumps({"sub": "user123"}).encode() # no exp
payload = base64.urlsafe_b64encode(payload_data).rstrip(b"=").decode()
no_exp_jwt = f"{header}.{payload}.fakesig"
hermes_home = tmp_path / "hermes"
hermes_home.mkdir(parents=True, exist_ok=True)
(hermes_home / "auth.json").write_text(json.dumps({
"version": 1,
"providers": {
"openai-codex": {
"tokens": {"access_token": no_exp_jwt, "refresh_token": "r"},
},
},
}))
monkeypatch.setenv("HERMES_HOME", str(hermes_home))
result = _read_codex_access_token()
assert result == no_exp_jwt, "JWT without exp should pass through"
def test_jwt_invalid_json_payload_passes_through(self, tmp_path, monkeypatch):
"""JWT with valid base64 but invalid JSON payload should pass through."""
import base64
header = base64.urlsafe_b64encode(b'{"alg":"RS256"}').rstrip(b"=").decode()
payload = base64.urlsafe_b64encode(b"not-json-content").rstrip(b"=").decode()
bad_jwt = f"{header}.{payload}.fakesig"
hermes_home = tmp_path / "hermes"
hermes_home.mkdir(parents=True, exist_ok=True)
(hermes_home / "auth.json").write_text(json.dumps({
"version": 1,
"providers": {
"openai-codex": {
"tokens": {"access_token": bad_jwt, "refresh_token": "r"},
},
},
}))
monkeypatch.setenv("HERMES_HOME", str(hermes_home))
result = _read_codex_access_token()
assert result == bad_jwt, "JWT with invalid JSON payload should pass through"
def test_claude_code_oauth_env_sets_flag(self, monkeypatch):
"""CLAUDE_CODE_OAUTH_TOKEN env var should get is_oauth=True."""
monkeypatch.setenv("CLAUDE_CODE_OAUTH_TOKEN", "cc-oauth-token-test")
monkeypatch.delenv("ANTHROPIC_TOKEN", raising=False)
with patch("agent.anthropic_adapter.build_anthropic_client") as mock_build:
mock_build.return_value = MagicMock()
from agent.auxiliary_client import _try_anthropic, AnthropicAuxiliaryClient
client, model = _try_anthropic()
assert client is not None
adapter = client.chat.completions
assert adapter._is_oauth is True
class TestExplicitProviderRouting:
"""Test explicit provider selection bypasses auto chain correctly."""
def test_explicit_anthropic_oauth(self, monkeypatch):
"""provider='anthropic' + OAuth token should work with is_oauth=True."""
monkeypatch.setenv("ANTHROPIC_TOKEN", "sk-ant-oat01-explicit-test")
with patch("agent.anthropic_adapter.build_anthropic_client") as mock_build:
mock_build.return_value = MagicMock()
client, model = resolve_provider_client("anthropic")
assert client is not None
# Verify OAuth flag propagated
adapter = client.chat.completions
assert adapter._is_oauth is True
def test_explicit_anthropic_api_key(self, monkeypatch):
"""provider='anthropic' + regular API key should work with is_oauth=False."""
with patch("agent.anthropic_adapter.resolve_anthropic_token", return_value="sk-ant-api-regular-key"), \
feat(auth): same-provider credential pools with rotation, custom endpoint support, and interactive CLI (#2647) * feat(auth): add same-provider credential pools and rotation UX Add same-provider credential pooling so Hermes can rotate across multiple credentials for a single provider, recover from exhausted credentials without jumping providers immediately, and configure that behavior directly in hermes setup. - agent/credential_pool.py: persisted per-provider credential pools - hermes auth add/list/remove/reset CLI commands - 429/402/401 recovery with pool rotation in run_agent.py - Setup wizard integration for pool strategy configuration - Auto-seeding from env vars and existing OAuth state Co-authored-by: kshitijk4poor <82637225+kshitijk4poor@users.noreply.github.com> Salvaged from PR #2647 * fix(tests): prevent pool auto-seeding from host env in credential pool tests Tests for non-pool Anthropic paths and auth remove were failing when host env vars (ANTHROPIC_API_KEY) or file-backed OAuth credentials were present. The pool auto-seeding picked these up, causing unexpected pool entries in tests. - Mock _select_pool_entry in auxiliary_client OAuth flag tests - Clear Anthropic env vars and mock _seed_from_singletons in auth remove test * feat(auth): add thread safety, least_used strategy, and request counting - Add threading.Lock to CredentialPool for gateway thread safety (concurrent requests from multiple gateway sessions could race on pool state mutations without this) - Add 'least_used' rotation strategy that selects the credential with the lowest request_count, distributing load more evenly - Add request_count field to PooledCredential for usage tracking - Add mark_used() method to increment per-credential request counts - Wrap select(), mark_exhausted_and_rotate(), and try_refresh_current() with lock acquisition - Add tests: least_used selection, mark_used counting, concurrent thread safety (4 threads × 20 selects with no corruption) * feat(auth): add interactive mode for bare 'hermes auth' command When 'hermes auth' is called without a subcommand, it now launches an interactive wizard that: 1. Shows full credential pool status across all providers 2. Offers a menu: add, remove, reset cooldowns, set strategy 3. For OAuth-capable providers (anthropic, nous, openai-codex), the add flow explicitly asks 'API key or OAuth login?' — making it clear that both auth types are supported for the same provider 4. Strategy picker shows all 4 options (fill_first, round_robin, least_used, random) with the current selection marked 5. Remove flow shows entries with indices for easy selection The subcommand paths (hermes auth add/list/remove/reset) still work exactly as before for scripted/non-interactive use. * fix(tests): update runtime_provider tests for config.yaml source of truth (#4165) Tests were using OPENAI_BASE_URL env var which is no longer consulted after #4165. Updated to use model config (provider, base_url, api_key) which is the new single source of truth for custom endpoint URLs. * feat(auth): support custom endpoint credential pools keyed by provider name Custom OpenAI-compatible endpoints all share provider='custom', making the provider-keyed pool useless. Now pools for custom endpoints are keyed by 'custom:<normalized_name>' where the name comes from the custom_providers config list (auto-generated from URL hostname). - Pool key format: 'custom:together.ai', 'custom:local-(localhost:8080)' - load_pool('custom:name') seeds from custom_providers api_key AND model.api_key when base_url matches - hermes auth add/list now shows custom endpoints alongside registry providers - _resolve_openrouter_runtime and _resolve_named_custom_runtime check pool before falling back to single config key - 6 new tests covering custom pool keying, seeding, and listing * docs: add Excalidraw diagram of full credential pool flow Comprehensive architecture diagram showing: - Credential sources (env vars, auth.json OAuth, config.yaml, CLI) - Pool storage and auto-seeding - Runtime resolution paths (registry, custom, OpenRouter) - Error recovery (429 retry-then-rotate, 402 immediate, 401 refresh) - CLI management commands and strategy configuration Open at: https://excalidraw.com/#json=2Ycqhqpi6f12E_3ITyiwh,c7u9jSt5BwrmiVzHGbm87g * fix(tests): update setup wizard pool tests for unified select_provider_and_model flow The setup wizard now delegates to select_provider_and_model() instead of using its own prompt_choice-based provider picker. Tests needed: - Mock select_provider_and_model as no-op (provider pre-written to config) - Call _stub_tts BEFORE custom prompt_choice mock (it overwrites it) - Pre-write model.provider to config so the pool step is reached * docs: add comprehensive credential pool documentation - New page: website/docs/user-guide/features/credential-pools.md Full guide covering quick start, CLI commands, rotation strategies, error recovery, custom endpoint pools, auto-discovery, thread safety, architecture, and storage format. - Updated fallback-providers.md to reference credential pools as the first layer of resilience (same-provider rotation before cross-provider) - Added hermes auth to CLI commands reference with usage examples - Added credential_pool_strategies to configuration guide * chore: remove excalidraw diagram from repo (external link only) * refactor: simplify credential pool code — extract helpers, collapse extras, dedup patterns - _load_config_safe(): replace 4 identical try/except/import blocks - _iter_custom_providers(): shared generator for custom provider iteration - PooledCredential.extra dict: collapse 11 round-trip-only fields (token_type, scope, client_id, portal_base_url, obtained_at, expires_in, agent_key_id, agent_key_expires_in, agent_key_reused, agent_key_obtained_at, tls) into a single extra dict with __getattr__ for backward-compatible access - _available_entries(): shared exhaustion-check between select and peek - Dedup anthropic OAuth seeding (hermes_pkce + claude_code identical) - SimpleNamespace replaces class _Args boilerplate in auth_commands - _try_resolve_from_custom_pool(): shared pool-check in runtime_provider Net -17 lines. All 383 targeted tests pass. --------- Co-authored-by: kshitijk4poor <82637225+kshitijk4poor@users.noreply.github.com>
2026-03-31 03:10:01 -07:00
patch("agent.anthropic_adapter.build_anthropic_client") as mock_build, \
patch("agent.auxiliary_client._select_pool_entry", return_value=(False, None)):
mock_build.return_value = MagicMock()
client, model = resolve_provider_client("anthropic")
assert client is not None
adapter = client.chat.completions
assert adapter._is_oauth is False
def test_explicit_openrouter(self, monkeypatch):
"""provider='openrouter' should use OPENROUTER_API_KEY."""
monkeypatch.setenv("OPENROUTER_API_KEY", "or-explicit")
with patch("agent.auxiliary_client.OpenAI") as mock_openai:
mock_openai.return_value = MagicMock()
client, model = resolve_provider_client("openrouter")
assert client is not None
def test_explicit_kimi(self, monkeypatch):
"""provider='kimi-coding' should use KIMI_API_KEY."""
monkeypatch.setenv("KIMI_API_KEY", "kimi-test-key")
with patch("agent.auxiliary_client.OpenAI") as mock_openai:
mock_openai.return_value = MagicMock()
client, model = resolve_provider_client("kimi-coding")
assert client is not None
def test_explicit_minimax(self, monkeypatch):
"""provider='minimax' should use MINIMAX_API_KEY."""
monkeypatch.setenv("MINIMAX_API_KEY", "mm-test-key")
with patch("agent.auxiliary_client.OpenAI") as mock_openai:
mock_openai.return_value = MagicMock()
client, model = resolve_provider_client("minimax")
assert client is not None
def test_explicit_deepseek(self, monkeypatch):
"""provider='deepseek' should use DEEPSEEK_API_KEY."""
monkeypatch.setenv("DEEPSEEK_API_KEY", "ds-test-key")
with patch("agent.auxiliary_client.OpenAI") as mock_openai:
mock_openai.return_value = MagicMock()
client, model = resolve_provider_client("deepseek")
assert client is not None
def test_explicit_zai(self, monkeypatch):
"""provider='zai' should use GLM_API_KEY."""
monkeypatch.setenv("GLM_API_KEY", "zai-test-key")
with patch("agent.auxiliary_client.OpenAI") as mock_openai:
mock_openai.return_value = MagicMock()
client, model = resolve_provider_client("zai")
assert client is not None
def test_explicit_unknown_returns_none(self, monkeypatch):
"""Unknown provider should return None."""
client, model = resolve_provider_client("nonexistent-provider")
assert client is None
class TestGetTextAuxiliaryClient:
"""Test the full resolution chain for get_text_auxiliary_client."""
def test_openrouter_takes_priority(self, monkeypatch, codex_auth_dir):
monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
with patch("agent.auxiliary_client.OpenAI") as mock_openai:
client, model = get_text_auxiliary_client()
assert model == "google/gemini-3-flash-preview"
mock_openai.assert_called_once()
call_kwargs = mock_openai.call_args
assert call_kwargs.kwargs["api_key"] == "or-key"
def test_nous_takes_priority_over_codex(self, monkeypatch, codex_auth_dir):
with patch("agent.auxiliary_client._read_nous_auth") as mock_nous, \
patch("agent.auxiliary_client.OpenAI") as mock_openai:
mock_nous.return_value = {"access_token": "nous-tok"}
client, model = get_text_auxiliary_client()
assert model == "google/gemini-3-flash-preview"
def test_custom_endpoint_over_codex(self, monkeypatch, codex_auth_dir):
config = {
"model": {
"provider": "custom",
"base_url": "http://localhost:1234/v1",
"default": "my-local-model",
}
}
monkeypatch.setenv("OPENAI_API_KEY", "lm-studio-key")
monkeypatch.setattr("hermes_cli.config.load_config", lambda: config)
monkeypatch.setattr("hermes_cli.runtime_provider.load_config", lambda: config)
# Override the autouse monkeypatch for codex
monkeypatch.setattr(
"agent.auxiliary_client._read_codex_access_token",
lambda: "codex-test-token-abc123",
)
with patch("agent.auxiliary_client._read_nous_auth", return_value=None), \
patch("agent.auxiliary_client.OpenAI") as mock_openai:
client, model = get_text_auxiliary_client()
fix: auxiliary client uses main model for custom/local endpoints instead of gpt-4o-mini (#1189) * fix: prevent model/provider mismatch when switching providers during active gateway When _update_config_for_provider() writes the new provider and base_url to config.yaml, the gateway (which re-reads config per-message) can pick up the change before model selection completes. This causes the old model name (e.g. 'anthropic/claude-opus-4.6') to be sent to the new provider's API (e.g. MiniMax), which fails. Changes: - _update_config_for_provider() now accepts an optional default_model parameter. When provided and the current model.default is empty or uses OpenRouter format (contains '/'), it sets a safe default model for the new provider. - All setup.py callers for direct-API providers (zai, kimi, minimax, minimax-cn, anthropic) now pass a provider-appropriate default model. - _setup_provider_model_selection() now validates the 'Keep current' choice: if the current model uses OpenRouter format and wouldn't work with the new provider, it warns and switches to the provider's first default model instead of silently keeping the incompatible name. Reported by a user on Home Assistant whose gateway started sending 'anthropic/claude-opus-4.6' to MiniMax's API after running hermes setup. * fix: auxiliary client uses main model for custom/local endpoints instead of gpt-4o-mini When a user runs a local server (e.g. Qwen3.5-9B via OPENAI_BASE_URL), the auxiliary client (context compression, vision, session search) would send requests for 'gpt-4o-mini' or 'google/gemini-3-flash-preview' to the local server, which only serves one model — causing 404 errors mid-task. Changes: - _try_custom_endpoint() now reads the user's configured main model via _read_main_model() (checks OPENAI_MODEL → HERMES_MODEL → LLM_MODEL → config.yaml model.default) instead of hardcoding 'gpt-4o-mini'. - resolve_provider_client() auto mode now detects when an OpenRouter- formatted model override (containing '/') would be sent to a non- OpenRouter provider (like a local server) and drops it in favor of the provider's default model. - Test isolation fixes: properly clear env vars in 'nothing available' tests to prevent host environment leakage.
2026-03-13 10:02:16 -07:00
assert model == "my-local-model"
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_uses_placeholder(self, monkeypatch):
"""Local endpoints without an API key should use 'no-key-required' placeholder."""
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 not None
assert model == "task-model"
assert mock_openai.call_args.kwargs["api_key"] == "no-key-required"
assert mock_openai.call_args.kwargs["base_url"] == "http://localhost:2345/v1"
def test_custom_endpoint_uses_config_saved_base_url(self, monkeypatch):
config = {
"model": {
"provider": "custom",
"base_url": "http://localhost:1234/v1",
"default": "my-local-model",
}
}
monkeypatch.setenv("OPENAI_API_KEY", "lm-studio-key")
monkeypatch.setattr("hermes_cli.config.load_config", lambda: config)
monkeypatch.setattr("hermes_cli.runtime_provider.load_config", lambda: config)
with patch("agent.auxiliary_client._read_nous_auth", return_value=None), \
patch("agent.auxiliary_client._read_codex_access_token", return_value=None), \
patch("agent.auxiliary_client._resolve_api_key_provider", return_value=(None, None)), \
patch("agent.auxiliary_client.OpenAI") as mock_openai:
client, model = get_text_auxiliary_client()
assert client is not None
assert model == "my-local-model"
call_kwargs = mock_openai.call_args
assert call_kwargs.kwargs["base_url"] == "http://localhost:1234/v1"
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:
client, model = get_text_auxiliary_client()
assert model == "gpt-5.2-codex"
# Returns a CodexAuxiliaryClient wrapper, not a raw OpenAI client
from agent.auxiliary_client import CodexAuxiliaryClient
assert isinstance(client, CodexAuxiliaryClient)
feat(auth): same-provider credential pools with rotation, custom endpoint support, and interactive CLI (#2647) * feat(auth): add same-provider credential pools and rotation UX Add same-provider credential pooling so Hermes can rotate across multiple credentials for a single provider, recover from exhausted credentials without jumping providers immediately, and configure that behavior directly in hermes setup. - agent/credential_pool.py: persisted per-provider credential pools - hermes auth add/list/remove/reset CLI commands - 429/402/401 recovery with pool rotation in run_agent.py - Setup wizard integration for pool strategy configuration - Auto-seeding from env vars and existing OAuth state Co-authored-by: kshitijk4poor <82637225+kshitijk4poor@users.noreply.github.com> Salvaged from PR #2647 * fix(tests): prevent pool auto-seeding from host env in credential pool tests Tests for non-pool Anthropic paths and auth remove were failing when host env vars (ANTHROPIC_API_KEY) or file-backed OAuth credentials were present. The pool auto-seeding picked these up, causing unexpected pool entries in tests. - Mock _select_pool_entry in auxiliary_client OAuth flag tests - Clear Anthropic env vars and mock _seed_from_singletons in auth remove test * feat(auth): add thread safety, least_used strategy, and request counting - Add threading.Lock to CredentialPool for gateway thread safety (concurrent requests from multiple gateway sessions could race on pool state mutations without this) - Add 'least_used' rotation strategy that selects the credential with the lowest request_count, distributing load more evenly - Add request_count field to PooledCredential for usage tracking - Add mark_used() method to increment per-credential request counts - Wrap select(), mark_exhausted_and_rotate(), and try_refresh_current() with lock acquisition - Add tests: least_used selection, mark_used counting, concurrent thread safety (4 threads × 20 selects with no corruption) * feat(auth): add interactive mode for bare 'hermes auth' command When 'hermes auth' is called without a subcommand, it now launches an interactive wizard that: 1. Shows full credential pool status across all providers 2. Offers a menu: add, remove, reset cooldowns, set strategy 3. For OAuth-capable providers (anthropic, nous, openai-codex), the add flow explicitly asks 'API key or OAuth login?' — making it clear that both auth types are supported for the same provider 4. Strategy picker shows all 4 options (fill_first, round_robin, least_used, random) with the current selection marked 5. Remove flow shows entries with indices for easy selection The subcommand paths (hermes auth add/list/remove/reset) still work exactly as before for scripted/non-interactive use. * fix(tests): update runtime_provider tests for config.yaml source of truth (#4165) Tests were using OPENAI_BASE_URL env var which is no longer consulted after #4165. Updated to use model config (provider, base_url, api_key) which is the new single source of truth for custom endpoint URLs. * feat(auth): support custom endpoint credential pools keyed by provider name Custom OpenAI-compatible endpoints all share provider='custom', making the provider-keyed pool useless. Now pools for custom endpoints are keyed by 'custom:<normalized_name>' where the name comes from the custom_providers config list (auto-generated from URL hostname). - Pool key format: 'custom:together.ai', 'custom:local-(localhost:8080)' - load_pool('custom:name') seeds from custom_providers api_key AND model.api_key when base_url matches - hermes auth add/list now shows custom endpoints alongside registry providers - _resolve_openrouter_runtime and _resolve_named_custom_runtime check pool before falling back to single config key - 6 new tests covering custom pool keying, seeding, and listing * docs: add Excalidraw diagram of full credential pool flow Comprehensive architecture diagram showing: - Credential sources (env vars, auth.json OAuth, config.yaml, CLI) - Pool storage and auto-seeding - Runtime resolution paths (registry, custom, OpenRouter) - Error recovery (429 retry-then-rotate, 402 immediate, 401 refresh) - CLI management commands and strategy configuration Open at: https://excalidraw.com/#json=2Ycqhqpi6f12E_3ITyiwh,c7u9jSt5BwrmiVzHGbm87g * fix(tests): update setup wizard pool tests for unified select_provider_and_model flow The setup wizard now delegates to select_provider_and_model() instead of using its own prompt_choice-based provider picker. Tests needed: - Mock select_provider_and_model as no-op (provider pre-written to config) - Call _stub_tts BEFORE custom prompt_choice mock (it overwrites it) - Pre-write model.provider to config so the pool step is reached * docs: add comprehensive credential pool documentation - New page: website/docs/user-guide/features/credential-pools.md Full guide covering quick start, CLI commands, rotation strategies, error recovery, custom endpoint pools, auto-discovery, thread safety, architecture, and storage format. - Updated fallback-providers.md to reference credential pools as the first layer of resilience (same-provider rotation before cross-provider) - Added hermes auth to CLI commands reference with usage examples - Added credential_pool_strategies to configuration guide * chore: remove excalidraw diagram from repo (external link only) * refactor: simplify credential pool code — extract helpers, collapse extras, dedup patterns - _load_config_safe(): replace 4 identical try/except/import blocks - _iter_custom_providers(): shared generator for custom provider iteration - PooledCredential.extra dict: collapse 11 round-trip-only fields (token_type, scope, client_id, portal_base_url, obtained_at, expires_in, agent_key_id, agent_key_expires_in, agent_key_reused, agent_key_obtained_at, tls) into a single extra dict with __getattr__ for backward-compatible access - _available_entries(): shared exhaustion-check between select and peek - Dedup anthropic OAuth seeding (hermes_pkce + claude_code identical) - SimpleNamespace replaces class _Args boilerplate in auth_commands - _try_resolve_from_custom_pool(): shared pool-check in runtime_provider Net -17 lines. All 383 targeted tests pass. --------- Co-authored-by: kshitijk4poor <82637225+kshitijk4poor@users.noreply.github.com>
2026-03-31 03:10:01 -07:00
def test_codex_pool_entry_takes_priority_over_auth_store(self):
class _Entry:
access_token = "pooled-codex-token"
base_url = "https://chatgpt.com/backend-api/codex"
class _Pool:
def has_credentials(self):
return True
def select(self):
return _Entry()
with (
patch("agent.auxiliary_client.load_pool", return_value=_Pool()),
patch("agent.auxiliary_client.OpenAI"),
patch("hermes_cli.auth._read_codex_tokens", side_effect=AssertionError("legacy codex store should not run")),
):
from agent.auxiliary_client import _try_codex
client, model = _try_codex()
from agent.auxiliary_client import CodexAuxiliaryClient
assert isinstance(client, CodexAuxiliaryClient)
assert model == "gpt-5.2-codex"
fix: auxiliary client uses main model for custom/local endpoints instead of gpt-4o-mini (#1189) * fix: prevent model/provider mismatch when switching providers during active gateway When _update_config_for_provider() writes the new provider and base_url to config.yaml, the gateway (which re-reads config per-message) can pick up the change before model selection completes. This causes the old model name (e.g. 'anthropic/claude-opus-4.6') to be sent to the new provider's API (e.g. MiniMax), which fails. Changes: - _update_config_for_provider() now accepts an optional default_model parameter. When provided and the current model.default is empty or uses OpenRouter format (contains '/'), it sets a safe default model for the new provider. - All setup.py callers for direct-API providers (zai, kimi, minimax, minimax-cn, anthropic) now pass a provider-appropriate default model. - _setup_provider_model_selection() now validates the 'Keep current' choice: if the current model uses OpenRouter format and wouldn't work with the new provider, it warns and switches to the provider's first default model instead of silently keeping the incompatible name. Reported by a user on Home Assistant whose gateway started sending 'anthropic/claude-opus-4.6' to MiniMax's API after running hermes setup. * fix: auxiliary client uses main model for custom/local endpoints instead of gpt-4o-mini When a user runs a local server (e.g. Qwen3.5-9B via OPENAI_BASE_URL), the auxiliary client (context compression, vision, session search) would send requests for 'gpt-4o-mini' or 'google/gemini-3-flash-preview' to the local server, which only serves one model — causing 404 errors mid-task. Changes: - _try_custom_endpoint() now reads the user's configured main model via _read_main_model() (checks OPENAI_MODEL → HERMES_MODEL → LLM_MODEL → config.yaml model.default) instead of hardcoding 'gpt-4o-mini'. - resolve_provider_client() auto mode now detects when an OpenRouter- formatted model override (containing '/') would be sent to a non- OpenRouter provider (like a local server) and drops it in favor of the provider's default model. - Test isolation fixes: properly clear env vars in 'nothing available' tests to prevent host environment leakage.
2026-03-13 10:02:16 -07:00
def test_returns_none_when_nothing_available(self, monkeypatch):
monkeypatch.delenv("OPENAI_BASE_URL", raising=False)
monkeypatch.delenv("OPENAI_API_KEY", raising=False)
monkeypatch.delenv("OPENROUTER_API_KEY", raising=False)
with patch("agent.auxiliary_client._read_nous_auth", return_value=None), \
fix: auxiliary client uses main model for custom/local endpoints instead of gpt-4o-mini (#1189) * fix: prevent model/provider mismatch when switching providers during active gateway When _update_config_for_provider() writes the new provider and base_url to config.yaml, the gateway (which re-reads config per-message) can pick up the change before model selection completes. This causes the old model name (e.g. 'anthropic/claude-opus-4.6') to be sent to the new provider's API (e.g. MiniMax), which fails. Changes: - _update_config_for_provider() now accepts an optional default_model parameter. When provided and the current model.default is empty or uses OpenRouter format (contains '/'), it sets a safe default model for the new provider. - All setup.py callers for direct-API providers (zai, kimi, minimax, minimax-cn, anthropic) now pass a provider-appropriate default model. - _setup_provider_model_selection() now validates the 'Keep current' choice: if the current model uses OpenRouter format and wouldn't work with the new provider, it warns and switches to the provider's first default model instead of silently keeping the incompatible name. Reported by a user on Home Assistant whose gateway started sending 'anthropic/claude-opus-4.6' to MiniMax's API after running hermes setup. * fix: auxiliary client uses main model for custom/local endpoints instead of gpt-4o-mini When a user runs a local server (e.g. Qwen3.5-9B via OPENAI_BASE_URL), the auxiliary client (context compression, vision, session search) would send requests for 'gpt-4o-mini' or 'google/gemini-3-flash-preview' to the local server, which only serves one model — causing 404 errors mid-task. Changes: - _try_custom_endpoint() now reads the user's configured main model via _read_main_model() (checks OPENAI_MODEL → HERMES_MODEL → LLM_MODEL → config.yaml model.default) instead of hardcoding 'gpt-4o-mini'. - resolve_provider_client() auto mode now detects when an OpenRouter- formatted model override (containing '/') would be sent to a non- OpenRouter provider (like a local server) and drops it in favor of the provider's default model. - Test isolation fixes: properly clear env vars in 'nothing available' tests to prevent host environment leakage.
2026-03-13 10:02:16 -07:00
patch("agent.auxiliary_client._read_codex_access_token", return_value=None), \
patch("agent.auxiliary_client._resolve_api_key_provider", return_value=(None, None)):
client, model = get_text_auxiliary_client()
assert client is None
assert model is None
class TestVisionClientFallback:
"""Vision client auto mode resolves known-good multimodal backends."""
def test_vision_returns_none_without_any_credentials(self):
with (
patch("agent.auxiliary_client._read_nous_auth", return_value=None),
patch("agent.auxiliary_client._try_anthropic", return_value=(None, None)),
):
client, model = get_vision_auxiliary_client()
assert client is None
assert model is None
def test_vision_auto_includes_anthropic_when_configured(self, monkeypatch):
monkeypatch.setenv("ANTHROPIC_API_KEY", "sk-ant-api03-key")
with (
patch("agent.auxiliary_client._read_nous_auth", return_value=None),
patch("agent.anthropic_adapter.build_anthropic_client", return_value=MagicMock()),
patch("agent.anthropic_adapter.resolve_anthropic_token", return_value="sk-ant-api03-key"),
):
backends = get_available_vision_backends()
assert "anthropic" in backends
def test_resolve_provider_client_returns_native_anthropic_wrapper(self, monkeypatch):
monkeypatch.setenv("ANTHROPIC_API_KEY", "sk-ant-api03-key")
with (
patch("agent.auxiliary_client._read_nous_auth", return_value=None),
patch("agent.anthropic_adapter.build_anthropic_client", return_value=MagicMock()),
patch("agent.anthropic_adapter.resolve_anthropic_token", return_value="sk-ant-api03-key"),
):
client, model = resolve_provider_client("anthropic")
assert client is not None
assert client.__class__.__name__ == "AnthropicAuxiliaryClient"
assert model == "claude-haiku-4-5-20251001"
feat(auth): same-provider credential pools with rotation, custom endpoint support, and interactive CLI (#2647) * feat(auth): add same-provider credential pools and rotation UX Add same-provider credential pooling so Hermes can rotate across multiple credentials for a single provider, recover from exhausted credentials without jumping providers immediately, and configure that behavior directly in hermes setup. - agent/credential_pool.py: persisted per-provider credential pools - hermes auth add/list/remove/reset CLI commands - 429/402/401 recovery with pool rotation in run_agent.py - Setup wizard integration for pool strategy configuration - Auto-seeding from env vars and existing OAuth state Co-authored-by: kshitijk4poor <82637225+kshitijk4poor@users.noreply.github.com> Salvaged from PR #2647 * fix(tests): prevent pool auto-seeding from host env in credential pool tests Tests for non-pool Anthropic paths and auth remove were failing when host env vars (ANTHROPIC_API_KEY) or file-backed OAuth credentials were present. The pool auto-seeding picked these up, causing unexpected pool entries in tests. - Mock _select_pool_entry in auxiliary_client OAuth flag tests - Clear Anthropic env vars and mock _seed_from_singletons in auth remove test * feat(auth): add thread safety, least_used strategy, and request counting - Add threading.Lock to CredentialPool for gateway thread safety (concurrent requests from multiple gateway sessions could race on pool state mutations without this) - Add 'least_used' rotation strategy that selects the credential with the lowest request_count, distributing load more evenly - Add request_count field to PooledCredential for usage tracking - Add mark_used() method to increment per-credential request counts - Wrap select(), mark_exhausted_and_rotate(), and try_refresh_current() with lock acquisition - Add tests: least_used selection, mark_used counting, concurrent thread safety (4 threads × 20 selects with no corruption) * feat(auth): add interactive mode for bare 'hermes auth' command When 'hermes auth' is called without a subcommand, it now launches an interactive wizard that: 1. Shows full credential pool status across all providers 2. Offers a menu: add, remove, reset cooldowns, set strategy 3. For OAuth-capable providers (anthropic, nous, openai-codex), the add flow explicitly asks 'API key or OAuth login?' — making it clear that both auth types are supported for the same provider 4. Strategy picker shows all 4 options (fill_first, round_robin, least_used, random) with the current selection marked 5. Remove flow shows entries with indices for easy selection The subcommand paths (hermes auth add/list/remove/reset) still work exactly as before for scripted/non-interactive use. * fix(tests): update runtime_provider tests for config.yaml source of truth (#4165) Tests were using OPENAI_BASE_URL env var which is no longer consulted after #4165. Updated to use model config (provider, base_url, api_key) which is the new single source of truth for custom endpoint URLs. * feat(auth): support custom endpoint credential pools keyed by provider name Custom OpenAI-compatible endpoints all share provider='custom', making the provider-keyed pool useless. Now pools for custom endpoints are keyed by 'custom:<normalized_name>' where the name comes from the custom_providers config list (auto-generated from URL hostname). - Pool key format: 'custom:together.ai', 'custom:local-(localhost:8080)' - load_pool('custom:name') seeds from custom_providers api_key AND model.api_key when base_url matches - hermes auth add/list now shows custom endpoints alongside registry providers - _resolve_openrouter_runtime and _resolve_named_custom_runtime check pool before falling back to single config key - 6 new tests covering custom pool keying, seeding, and listing * docs: add Excalidraw diagram of full credential pool flow Comprehensive architecture diagram showing: - Credential sources (env vars, auth.json OAuth, config.yaml, CLI) - Pool storage and auto-seeding - Runtime resolution paths (registry, custom, OpenRouter) - Error recovery (429 retry-then-rotate, 402 immediate, 401 refresh) - CLI management commands and strategy configuration Open at: https://excalidraw.com/#json=2Ycqhqpi6f12E_3ITyiwh,c7u9jSt5BwrmiVzHGbm87g * fix(tests): update setup wizard pool tests for unified select_provider_and_model flow The setup wizard now delegates to select_provider_and_model() instead of using its own prompt_choice-based provider picker. Tests needed: - Mock select_provider_and_model as no-op (provider pre-written to config) - Call _stub_tts BEFORE custom prompt_choice mock (it overwrites it) - Pre-write model.provider to config so the pool step is reached * docs: add comprehensive credential pool documentation - New page: website/docs/user-guide/features/credential-pools.md Full guide covering quick start, CLI commands, rotation strategies, error recovery, custom endpoint pools, auto-discovery, thread safety, architecture, and storage format. - Updated fallback-providers.md to reference credential pools as the first layer of resilience (same-provider rotation before cross-provider) - Added hermes auth to CLI commands reference with usage examples - Added credential_pool_strategies to configuration guide * chore: remove excalidraw diagram from repo (external link only) * refactor: simplify credential pool code — extract helpers, collapse extras, dedup patterns - _load_config_safe(): replace 4 identical try/except/import blocks - _iter_custom_providers(): shared generator for custom provider iteration - PooledCredential.extra dict: collapse 11 round-trip-only fields (token_type, scope, client_id, portal_base_url, obtained_at, expires_in, agent_key_id, agent_key_expires_in, agent_key_reused, agent_key_obtained_at, tls) into a single extra dict with __getattr__ for backward-compatible access - _available_entries(): shared exhaustion-check between select and peek - Dedup anthropic OAuth seeding (hermes_pkce + claude_code identical) - SimpleNamespace replaces class _Args boilerplate in auth_commands - _try_resolve_from_custom_pool(): shared pool-check in runtime_provider Net -17 lines. All 383 targeted tests pass. --------- Co-authored-by: kshitijk4poor <82637225+kshitijk4poor@users.noreply.github.com>
2026-03-31 03:10:01 -07:00
class TestAuxiliaryPoolAwareness:
def test_try_nous_uses_pool_entry(self):
class _Entry:
access_token = "pooled-access-token"
agent_key = "pooled-agent-key"
inference_base_url = "https://inference.pool.example/v1"
class _Pool:
def has_credentials(self):
return True
def select(self):
return _Entry()
with (
patch("agent.auxiliary_client.load_pool", return_value=_Pool()),
patch("agent.auxiliary_client.OpenAI") as mock_openai,
):
from agent.auxiliary_client import _try_nous
client, model = _try_nous()
assert client is not None
assert model == "gemini-3-flash"
call_kwargs = mock_openai.call_args.kwargs
assert call_kwargs["api_key"] == "pooled-agent-key"
assert call_kwargs["base_url"] == "https://inference.pool.example/v1"
def test_resolve_provider_client_copilot_uses_runtime_credentials(self, monkeypatch):
monkeypatch.delenv("GITHUB_TOKEN", raising=False)
monkeypatch.delenv("GH_TOKEN", raising=False)
with (
patch(
"hermes_cli.auth.resolve_api_key_provider_credentials",
return_value={
"provider": "copilot",
"api_key": "gh-cli-token",
"base_url": "https://api.githubcopilot.com",
"source": "gh auth token",
},
),
patch("agent.auxiliary_client.OpenAI") as mock_openai,
):
client, model = resolve_provider_client("copilot", model="gpt-5.4")
assert client is not None
assert model == "gpt-5.4"
call_kwargs = mock_openai.call_args.kwargs
assert call_kwargs["api_key"] == "gh-cli-token"
assert call_kwargs["base_url"] == "https://api.githubcopilot.com"
assert call_kwargs["default_headers"]["Editor-Version"]
def test_vision_auto_uses_anthropic_when_no_higher_priority_backend(self, monkeypatch):
monkeypatch.setenv("ANTHROPIC_API_KEY", "sk-ant-api03-key")
with (
patch("agent.auxiliary_client._read_nous_auth", return_value=None),
patch("agent.anthropic_adapter.build_anthropic_client", return_value=MagicMock()),
patch("agent.anthropic_adapter.resolve_anthropic_token", return_value="sk-ant-api03-key"),
):
client, model = get_vision_auxiliary_client()
assert client is not None
assert client.__class__.__name__ == "AnthropicAuxiliaryClient"
assert model == "claude-haiku-4-5-20251001"
def test_selected_anthropic_provider_is_preferred_for_vision_auto(self, monkeypatch):
monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
monkeypatch.setenv("ANTHROPIC_API_KEY", "sk-ant-api03-key")
def fake_load_config():
return {"model": {"provider": "anthropic", "default": "claude-sonnet-4-6"}}
with (
patch("agent.auxiliary_client._read_nous_auth", return_value=None),
patch("agent.anthropic_adapter.build_anthropic_client", return_value=MagicMock()),
patch("agent.anthropic_adapter.resolve_anthropic_token", return_value="sk-ant-api03-key"),
patch("agent.auxiliary_client.OpenAI") as mock_openai,
patch("hermes_cli.config.load_config", fake_load_config),
):
client, model = get_vision_auxiliary_client()
assert client is not None
assert client.__class__.__name__ == "AnthropicAuxiliaryClient"
assert model == "claude-haiku-4-5-20251001"
def test_selected_codex_provider_short_circuits_vision_auto(self, monkeypatch):
def fake_load_config():
return {"model": {"provider": "openai-codex", "default": "gpt-5.2-codex"}}
codex_client = MagicMock()
with (
patch("hermes_cli.config.load_config", fake_load_config),
patch("agent.auxiliary_client._try_codex", return_value=(codex_client, "gpt-5.2-codex")) as mock_codex,
patch("agent.auxiliary_client._try_openrouter") as mock_openrouter,
patch("agent.auxiliary_client._try_nous") as mock_nous,
patch("agent.auxiliary_client._try_anthropic") as mock_anthropic,
patch("agent.auxiliary_client._try_custom_endpoint") as mock_custom,
):
provider, client, model = resolve_vision_provider_client()
assert provider == "openai-codex"
assert client is codex_client
assert model == "gpt-5.2-codex"
mock_codex.assert_called_once()
mock_openrouter.assert_not_called()
mock_nous.assert_not_called()
mock_anthropic.assert_not_called()
mock_custom.assert_not_called()
def test_vision_auto_includes_codex(self, codex_auth_dir):
"""Codex supports vision (gpt-5.3-codex), so auto mode should use it."""
with patch("agent.auxiliary_client._read_nous_auth", return_value=None), \
patch("agent.auxiliary_client.OpenAI"):
client, model = get_vision_auxiliary_client()
from agent.auxiliary_client import CodexAuxiliaryClient
assert isinstance(client, CodexAuxiliaryClient)
assert model == "gpt-5.2-codex"
def test_vision_auto_falls_back_to_custom_endpoint(self, monkeypatch):
"""Custom endpoint is used as fallback in vision auto mode.
Many local models (Qwen-VL, LLaVA, etc.) support vision.
When no OpenRouter/Nous/Codex is available, try the custom endpoint.
"""
monkeypatch.delenv("OPENROUTER_API_KEY", raising=False)
monkeypatch.delenv("ANTHROPIC_API_KEY", raising=False)
with patch("agent.auxiliary_client._read_nous_auth", return_value=None), \
patch("agent.auxiliary_client._select_pool_entry", return_value=(False, None)), \
patch("agent.auxiliary_client._read_codex_access_token", return_value=None), \
patch("agent.auxiliary_client._resolve_custom_runtime",
return_value=("http://localhost:1234/v1", "local-key")), \
patch("agent.auxiliary_client.OpenAI") as mock_openai:
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_without_key_uses_placeholder(self, monkeypatch):
"""Vision endpoint without API key should use 'no-key-required' placeholder."""
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 not None
assert model == "vision-model"
assert mock_openai.call_args.kwargs["api_key"] == "no-key-required"
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:
client, model = get_vision_auxiliary_client()
assert model == "google/gemini-3-flash-preview"
assert client is not None
def test_vision_uses_nous_when_available(self, monkeypatch):
with patch("agent.auxiliary_client._read_nous_auth") as mock_nous, \
patch("agent.auxiliary_client.OpenAI"):
mock_nous.return_value = {"access_token": "nous-tok"}
client, model = get_vision_auxiliary_client()
assert model == "google/gemini-3-flash-preview"
assert client is not None
def test_vision_forced_main_uses_custom_endpoint(self, monkeypatch):
"""When explicitly forced to 'main', vision CAN use custom endpoint."""
config = {
"model": {
"provider": "custom",
"base_url": "http://localhost:1234/v1",
"default": "my-local-model",
}
}
monkeypatch.setenv("AUXILIARY_VISION_PROVIDER", "main")
monkeypatch.setenv("OPENAI_API_KEY", "local-key")
monkeypatch.setattr("hermes_cli.config.load_config", lambda: config)
monkeypatch.setattr("hermes_cli.runtime_provider.load_config", lambda: config)
with patch("agent.auxiliary_client._read_nous_auth", return_value=None), \
patch("agent.auxiliary_client.OpenAI") as mock_openai:
client, model = get_vision_auxiliary_client()
assert client is not None
fix: auxiliary client uses main model for custom/local endpoints instead of gpt-4o-mini (#1189) * fix: prevent model/provider mismatch when switching providers during active gateway When _update_config_for_provider() writes the new provider and base_url to config.yaml, the gateway (which re-reads config per-message) can pick up the change before model selection completes. This causes the old model name (e.g. 'anthropic/claude-opus-4.6') to be sent to the new provider's API (e.g. MiniMax), which fails. Changes: - _update_config_for_provider() now accepts an optional default_model parameter. When provided and the current model.default is empty or uses OpenRouter format (contains '/'), it sets a safe default model for the new provider. - All setup.py callers for direct-API providers (zai, kimi, minimax, minimax-cn, anthropic) now pass a provider-appropriate default model. - _setup_provider_model_selection() now validates the 'Keep current' choice: if the current model uses OpenRouter format and wouldn't work with the new provider, it warns and switches to the provider's first default model instead of silently keeping the incompatible name. Reported by a user on Home Assistant whose gateway started sending 'anthropic/claude-opus-4.6' to MiniMax's API after running hermes setup. * fix: auxiliary client uses main model for custom/local endpoints instead of gpt-4o-mini When a user runs a local server (e.g. Qwen3.5-9B via OPENAI_BASE_URL), the auxiliary client (context compression, vision, session search) would send requests for 'gpt-4o-mini' or 'google/gemini-3-flash-preview' to the local server, which only serves one model — causing 404 errors mid-task. Changes: - _try_custom_endpoint() now reads the user's configured main model via _read_main_model() (checks OPENAI_MODEL → HERMES_MODEL → LLM_MODEL → config.yaml model.default) instead of hardcoding 'gpt-4o-mini'. - resolve_provider_client() auto mode now detects when an OpenRouter- formatted model override (containing '/') would be sent to a non- OpenRouter provider (like a local server) and drops it in favor of the provider's default model. - Test isolation fixes: properly clear env vars in 'nothing available' tests to prevent host environment leakage.
2026-03-13 10:02:16 -07:00
assert model == "my-local-model"
def test_vision_forced_main_returns_none_without_creds(self, monkeypatch):
"""Forced main with no credentials still returns None."""
monkeypatch.setenv("AUXILIARY_VISION_PROVIDER", "main")
fix: auxiliary client uses main model for custom/local endpoints instead of gpt-4o-mini (#1189) * fix: prevent model/provider mismatch when switching providers during active gateway When _update_config_for_provider() writes the new provider and base_url to config.yaml, the gateway (which re-reads config per-message) can pick up the change before model selection completes. This causes the old model name (e.g. 'anthropic/claude-opus-4.6') to be sent to the new provider's API (e.g. MiniMax), which fails. Changes: - _update_config_for_provider() now accepts an optional default_model parameter. When provided and the current model.default is empty or uses OpenRouter format (contains '/'), it sets a safe default model for the new provider. - All setup.py callers for direct-API providers (zai, kimi, minimax, minimax-cn, anthropic) now pass a provider-appropriate default model. - _setup_provider_model_selection() now validates the 'Keep current' choice: if the current model uses OpenRouter format and wouldn't work with the new provider, it warns and switches to the provider's first default model instead of silently keeping the incompatible name. Reported by a user on Home Assistant whose gateway started sending 'anthropic/claude-opus-4.6' to MiniMax's API after running hermes setup. * fix: auxiliary client uses main model for custom/local endpoints instead of gpt-4o-mini When a user runs a local server (e.g. Qwen3.5-9B via OPENAI_BASE_URL), the auxiliary client (context compression, vision, session search) would send requests for 'gpt-4o-mini' or 'google/gemini-3-flash-preview' to the local server, which only serves one model — causing 404 errors mid-task. Changes: - _try_custom_endpoint() now reads the user's configured main model via _read_main_model() (checks OPENAI_MODEL → HERMES_MODEL → LLM_MODEL → config.yaml model.default) instead of hardcoding 'gpt-4o-mini'. - resolve_provider_client() auto mode now detects when an OpenRouter- formatted model override (containing '/') would be sent to a non- OpenRouter provider (like a local server) and drops it in favor of the provider's default model. - Test isolation fixes: properly clear env vars in 'nothing available' tests to prevent host environment leakage.
2026-03-13 10:02:16 -07:00
monkeypatch.delenv("OPENAI_BASE_URL", raising=False)
monkeypatch.delenv("OPENAI_API_KEY", raising=False)
with patch("agent.auxiliary_client._read_nous_auth", return_value=None), \
fix: auxiliary client uses main model for custom/local endpoints instead of gpt-4o-mini (#1189) * fix: prevent model/provider mismatch when switching providers during active gateway When _update_config_for_provider() writes the new provider and base_url to config.yaml, the gateway (which re-reads config per-message) can pick up the change before model selection completes. This causes the old model name (e.g. 'anthropic/claude-opus-4.6') to be sent to the new provider's API (e.g. MiniMax), which fails. Changes: - _update_config_for_provider() now accepts an optional default_model parameter. When provided and the current model.default is empty or uses OpenRouter format (contains '/'), it sets a safe default model for the new provider. - All setup.py callers for direct-API providers (zai, kimi, minimax, minimax-cn, anthropic) now pass a provider-appropriate default model. - _setup_provider_model_selection() now validates the 'Keep current' choice: if the current model uses OpenRouter format and wouldn't work with the new provider, it warns and switches to the provider's first default model instead of silently keeping the incompatible name. Reported by a user on Home Assistant whose gateway started sending 'anthropic/claude-opus-4.6' to MiniMax's API after running hermes setup. * fix: auxiliary client uses main model for custom/local endpoints instead of gpt-4o-mini When a user runs a local server (e.g. Qwen3.5-9B via OPENAI_BASE_URL), the auxiliary client (context compression, vision, session search) would send requests for 'gpt-4o-mini' or 'google/gemini-3-flash-preview' to the local server, which only serves one model — causing 404 errors mid-task. Changes: - _try_custom_endpoint() now reads the user's configured main model via _read_main_model() (checks OPENAI_MODEL → HERMES_MODEL → LLM_MODEL → config.yaml model.default) instead of hardcoding 'gpt-4o-mini'. - resolve_provider_client() auto mode now detects when an OpenRouter- formatted model override (containing '/') would be sent to a non- OpenRouter provider (like a local server) and drops it in favor of the provider's default model. - Test isolation fixes: properly clear env vars in 'nothing available' tests to prevent host environment leakage.
2026-03-13 10:02:16 -07:00
patch("agent.auxiliary_client._read_codex_access_token", return_value=None), \
patch("agent.auxiliary_client._resolve_api_key_provider", return_value=(None, None)):
client, model = get_vision_auxiliary_client()
assert client is None
assert model is None
def test_vision_forced_codex(self, monkeypatch, codex_auth_dir):
"""When forced to 'codex', vision uses Codex OAuth."""
monkeypatch.setenv("AUXILIARY_VISION_PROVIDER", "codex")
with patch("agent.auxiliary_client._read_nous_auth", return_value=None), \
patch("agent.auxiliary_client.OpenAI"):
client, model = get_vision_auxiliary_client()
from agent.auxiliary_client import CodexAuxiliaryClient
assert isinstance(client, CodexAuxiliaryClient)
assert model == "gpt-5.2-codex"
class TestGetAuxiliaryProvider:
"""Tests for _get_auxiliary_provider env var resolution."""
def test_no_task_returns_auto(self):
assert _get_auxiliary_provider() == "auto"
assert _get_auxiliary_provider("") == "auto"
def test_auxiliary_prefix_takes_priority(self, monkeypatch):
monkeypatch.setenv("AUXILIARY_VISION_PROVIDER", "openrouter")
assert _get_auxiliary_provider("vision") == "openrouter"
def test_context_prefix_fallback(self, monkeypatch):
monkeypatch.setenv("CONTEXT_COMPRESSION_PROVIDER", "nous")
assert _get_auxiliary_provider("compression") == "nous"
def test_auxiliary_prefix_over_context_prefix(self, monkeypatch):
monkeypatch.setenv("AUXILIARY_COMPRESSION_PROVIDER", "openrouter")
monkeypatch.setenv("CONTEXT_COMPRESSION_PROVIDER", "nous")
assert _get_auxiliary_provider("compression") == "openrouter"
def test_auto_value_treated_as_auto(self, monkeypatch):
monkeypatch.setenv("AUXILIARY_VISION_PROVIDER", "auto")
assert _get_auxiliary_provider("vision") == "auto"
def test_whitespace_stripped(self, monkeypatch):
monkeypatch.setenv("AUXILIARY_VISION_PROVIDER", " openrouter ")
assert _get_auxiliary_provider("vision") == "openrouter"
def test_case_insensitive(self, monkeypatch):
monkeypatch.setenv("AUXILIARY_VISION_PROVIDER", "OpenRouter")
assert _get_auxiliary_provider("vision") == "openrouter"
def test_main_provider(self, monkeypatch):
monkeypatch.setenv("AUXILIARY_WEB_EXTRACT_PROVIDER", "main")
assert _get_auxiliary_provider("web_extract") == "main"
class TestResolveForcedProvider:
"""Tests for _resolve_forced_provider with explicit provider selection."""
def test_forced_openrouter(self, monkeypatch):
monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
with patch("agent.auxiliary_client.OpenAI") as mock_openai:
client, model = _resolve_forced_provider("openrouter")
assert model == "google/gemini-3-flash-preview"
assert client is not None
def test_forced_openrouter_no_key(self, monkeypatch):
with patch("agent.auxiliary_client._read_nous_auth", return_value=None):
client, model = _resolve_forced_provider("openrouter")
assert client is None
assert model is None
def test_forced_nous(self, monkeypatch):
with patch("agent.auxiliary_client._read_nous_auth") as mock_nous, \
patch("agent.auxiliary_client.OpenAI"):
mock_nous.return_value = {"access_token": "nous-tok"}
client, model = _resolve_forced_provider("nous")
assert model == "google/gemini-3-flash-preview"
assert client is not None
def test_forced_nous_not_configured(self, monkeypatch):
with patch("agent.auxiliary_client._read_nous_auth", return_value=None):
client, model = _resolve_forced_provider("nous")
assert client is None
assert model is None
def test_forced_main_uses_custom(self, monkeypatch):
config = {
"model": {
"provider": "custom",
"base_url": "http://local:8080/v1",
"default": "my-local-model",
}
}
monkeypatch.setenv("OPENAI_API_KEY", "local-key")
monkeypatch.setattr("hermes_cli.config.load_config", lambda: config)
monkeypatch.setattr("hermes_cli.runtime_provider.load_config", lambda: config)
with patch("agent.auxiliary_client._read_nous_auth", return_value=None), \
patch("agent.auxiliary_client.OpenAI") as mock_openai:
client, model = _resolve_forced_provider("main")
fix: auxiliary client uses main model for custom/local endpoints instead of gpt-4o-mini (#1189) * fix: prevent model/provider mismatch when switching providers during active gateway When _update_config_for_provider() writes the new provider and base_url to config.yaml, the gateway (which re-reads config per-message) can pick up the change before model selection completes. This causes the old model name (e.g. 'anthropic/claude-opus-4.6') to be sent to the new provider's API (e.g. MiniMax), which fails. Changes: - _update_config_for_provider() now accepts an optional default_model parameter. When provided and the current model.default is empty or uses OpenRouter format (contains '/'), it sets a safe default model for the new provider. - All setup.py callers for direct-API providers (zai, kimi, minimax, minimax-cn, anthropic) now pass a provider-appropriate default model. - _setup_provider_model_selection() now validates the 'Keep current' choice: if the current model uses OpenRouter format and wouldn't work with the new provider, it warns and switches to the provider's first default model instead of silently keeping the incompatible name. Reported by a user on Home Assistant whose gateway started sending 'anthropic/claude-opus-4.6' to MiniMax's API after running hermes setup. * fix: auxiliary client uses main model for custom/local endpoints instead of gpt-4o-mini When a user runs a local server (e.g. Qwen3.5-9B via OPENAI_BASE_URL), the auxiliary client (context compression, vision, session search) would send requests for 'gpt-4o-mini' or 'google/gemini-3-flash-preview' to the local server, which only serves one model — causing 404 errors mid-task. Changes: - _try_custom_endpoint() now reads the user's configured main model via _read_main_model() (checks OPENAI_MODEL → HERMES_MODEL → LLM_MODEL → config.yaml model.default) instead of hardcoding 'gpt-4o-mini'. - resolve_provider_client() auto mode now detects when an OpenRouter- formatted model override (containing '/') would be sent to a non- OpenRouter provider (like a local server) and drops it in favor of the provider's default model. - Test isolation fixes: properly clear env vars in 'nothing available' tests to prevent host environment leakage.
2026-03-13 10:02:16 -07:00
assert model == "my-local-model"
def test_forced_main_uses_config_saved_custom_endpoint(self, monkeypatch):
config = {
"model": {
"provider": "custom",
"base_url": "http://local:8080/v1",
"default": "my-local-model",
}
}
monkeypatch.setenv("OPENAI_API_KEY", "local-key")
monkeypatch.setattr("hermes_cli.config.load_config", lambda: config)
monkeypatch.setattr("hermes_cli.runtime_provider.load_config", lambda: config)
with patch("agent.auxiliary_client._read_nous_auth", return_value=None), \
patch("agent.auxiliary_client._read_codex_access_token", return_value=None), \
patch("agent.auxiliary_client._resolve_api_key_provider", return_value=(None, None)), \
patch("agent.auxiliary_client.OpenAI") as mock_openai:
client, model = _resolve_forced_provider("main")
assert client is not None
assert model == "my-local-model"
call_kwargs = mock_openai.call_args
assert call_kwargs.kwargs["base_url"] == "http://local:8080/v1"
def test_forced_main_skips_openrouter_nous(self, monkeypatch):
"""Even if OpenRouter key is set, 'main' skips it."""
config = {
"model": {
"provider": "custom",
"base_url": "http://local:8080/v1",
"default": "my-local-model",
}
}
monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
monkeypatch.setenv("OPENAI_API_KEY", "local-key")
monkeypatch.setattr("hermes_cli.config.load_config", lambda: config)
monkeypatch.setattr("hermes_cli.runtime_provider.load_config", lambda: config)
with patch("agent.auxiliary_client._read_nous_auth", return_value=None), \
patch("agent.auxiliary_client.OpenAI") as mock_openai:
client, model = _resolve_forced_provider("main")
# Should use custom endpoint, not OpenRouter
fix: auxiliary client uses main model for custom/local endpoints instead of gpt-4o-mini (#1189) * fix: prevent model/provider mismatch when switching providers during active gateway When _update_config_for_provider() writes the new provider and base_url to config.yaml, the gateway (which re-reads config per-message) can pick up the change before model selection completes. This causes the old model name (e.g. 'anthropic/claude-opus-4.6') to be sent to the new provider's API (e.g. MiniMax), which fails. Changes: - _update_config_for_provider() now accepts an optional default_model parameter. When provided and the current model.default is empty or uses OpenRouter format (contains '/'), it sets a safe default model for the new provider. - All setup.py callers for direct-API providers (zai, kimi, minimax, minimax-cn, anthropic) now pass a provider-appropriate default model. - _setup_provider_model_selection() now validates the 'Keep current' choice: if the current model uses OpenRouter format and wouldn't work with the new provider, it warns and switches to the provider's first default model instead of silently keeping the incompatible name. Reported by a user on Home Assistant whose gateway started sending 'anthropic/claude-opus-4.6' to MiniMax's API after running hermes setup. * fix: auxiliary client uses main model for custom/local endpoints instead of gpt-4o-mini When a user runs a local server (e.g. Qwen3.5-9B via OPENAI_BASE_URL), the auxiliary client (context compression, vision, session search) would send requests for 'gpt-4o-mini' or 'google/gemini-3-flash-preview' to the local server, which only serves one model — causing 404 errors mid-task. Changes: - _try_custom_endpoint() now reads the user's configured main model via _read_main_model() (checks OPENAI_MODEL → HERMES_MODEL → LLM_MODEL → config.yaml model.default) instead of hardcoding 'gpt-4o-mini'. - resolve_provider_client() auto mode now detects when an OpenRouter- formatted model override (containing '/') would be sent to a non- OpenRouter provider (like a local server) and drops it in favor of the provider's default model. - Test isolation fixes: properly clear env vars in 'nothing available' tests to prevent host environment leakage.
2026-03-13 10:02:16 -07:00
assert model == "my-local-model"
def test_forced_main_falls_to_codex(self, codex_auth_dir, monkeypatch):
with patch("agent.auxiliary_client._read_nous_auth", return_value=None), \
patch("agent.auxiliary_client.OpenAI"):
client, model = _resolve_forced_provider("main")
from agent.auxiliary_client import CodexAuxiliaryClient
assert isinstance(client, CodexAuxiliaryClient)
assert model == "gpt-5.2-codex"
def test_forced_codex(self, codex_auth_dir, monkeypatch):
with patch("agent.auxiliary_client._read_nous_auth", return_value=None), \
patch("agent.auxiliary_client.OpenAI"):
client, model = _resolve_forced_provider("codex")
from agent.auxiliary_client import CodexAuxiliaryClient
assert isinstance(client, CodexAuxiliaryClient)
assert model == "gpt-5.2-codex"
def test_forced_codex_no_token(self, monkeypatch):
with patch("agent.auxiliary_client._read_codex_access_token", return_value=None):
client, model = _resolve_forced_provider("codex")
assert client is None
assert model is None
def test_forced_unknown_returns_none(self, monkeypatch):
with patch("agent.auxiliary_client._read_nous_auth", return_value=None), \
patch("agent.auxiliary_client._read_codex_access_token", return_value=None):
client, model = _resolve_forced_provider("invalid-provider")
assert client is None
assert model is None
class TestTaskSpecificOverrides:
"""Integration tests for per-task provider routing via get_text_auxiliary_client(task=...)."""
def test_text_with_vision_provider_override(self, monkeypatch):
"""AUXILIARY_VISION_PROVIDER should not affect text tasks."""
monkeypatch.setenv("AUXILIARY_VISION_PROVIDER", "nous")
monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
with patch("agent.auxiliary_client.OpenAI"):
client, model = get_text_auxiliary_client() # no task → auto
assert model == "google/gemini-3-flash-preview" # OpenRouter, not Nous
def test_compression_task_reads_context_prefix(self, monkeypatch):
"""Compression task should check CONTEXT_COMPRESSION_PROVIDER env var."""
monkeypatch.setenv("CONTEXT_COMPRESSION_PROVIDER", "nous")
monkeypatch.setenv("OPENROUTER_API_KEY", "or-key") # would win in auto
with patch("agent.auxiliary_client._read_nous_auth") as mock_nous, \
patch("agent.auxiliary_client.OpenAI"):
mock_nous.return_value = {"access_token": "***"}
client, model = get_text_auxiliary_client("compression")
# Config-first: model comes from config.yaml summary_model default,
# but provider is forced to Nous via env var
assert client is not None
def test_web_extract_task_override(self, monkeypatch):
monkeypatch.setenv("AUXILIARY_WEB_EXTRACT_PROVIDER", "openrouter")
monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
with patch("agent.auxiliary_client.OpenAI"):
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")
with patch("agent.auxiliary_client.OpenAI"):
client, model = get_text_auxiliary_client("compression")
assert model == "google/gemini-3-flash-preview" # auto → OpenRouter
def test_compression_summary_base_url_from_config(self, monkeypatch, tmp_path):
"""compression.summary_base_url should produce a custom-endpoint client."""
hermes_home = tmp_path / "hermes"
hermes_home.mkdir(parents=True, exist_ok=True)
(hermes_home / "config.yaml").write_text(
"""compression:
summary_provider: custom
summary_model: glm-4.7
summary_base_url: https://api.z.ai/api/coding/paas/v4
"""
)
monkeypatch.setenv("HERMES_HOME", str(hermes_home))
# Custom endpoints need an API key to build the client
monkeypatch.setenv("OPENAI_API_KEY", "test-key")
with patch("agent.auxiliary_client.OpenAI") as mock_openai:
client, model = get_text_auxiliary_client("compression")
assert model == "glm-4.7"
assert mock_openai.call_args.kwargs["base_url"] == "https://api.z.ai/api/coding/paas/v4"
class TestAuxiliaryMaxTokensParam:
def test_codex_fallback_uses_max_tokens(self, monkeypatch):
"""Codex adapter translates max_tokens internally, so we return max_tokens."""
with patch("agent.auxiliary_client._read_nous_auth", return_value=None), \
patch("agent.auxiliary_client._read_codex_access_token", return_value="tok"):
result = auxiliary_max_tokens_param(1024)
assert result == {"max_tokens": 1024}
def test_openrouter_uses_max_tokens(self, monkeypatch):
monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
result = auxiliary_max_tokens_param(1024)
assert result == {"max_tokens": 1024}
def test_no_provider_uses_max_tokens(self):
with patch("agent.auxiliary_client._read_nous_auth", return_value=None), \
patch("agent.auxiliary_client._read_codex_access_token", return_value=None):
result = auxiliary_max_tokens_param(1024)
assert result == {"max_tokens": 1024}
# ── Payment / credit exhaustion fallback ─────────────────────────────────
class TestIsPaymentError:
"""_is_payment_error detects 402 and credit-related errors."""
def test_402_status_code(self):
exc = Exception("Payment Required")
exc.status_code = 402
assert _is_payment_error(exc) is True
def test_402_with_credits_message(self):
exc = Exception("You requested up to 65535 tokens, but can only afford 8029")
exc.status_code = 402
assert _is_payment_error(exc) is True
def test_429_with_credits_message(self):
exc = Exception("insufficient credits remaining")
exc.status_code = 429
assert _is_payment_error(exc) is True
def test_429_without_credits_message_is_not_payment(self):
"""Normal rate limits should NOT be treated as payment errors."""
exc = Exception("Rate limit exceeded, try again in 2 seconds")
exc.status_code = 429
assert _is_payment_error(exc) is False
def test_generic_500_is_not_payment(self):
exc = Exception("Internal server error")
exc.status_code = 500
assert _is_payment_error(exc) is False
def test_no_status_code_with_billing_message(self):
exc = Exception("billing: payment required for this request")
assert _is_payment_error(exc) is True
def test_no_status_code_no_message(self):
exc = Exception("connection reset")
assert _is_payment_error(exc) is False
class TestGetProviderChain:
"""_get_provider_chain() resolves functions at call time (testable)."""
def test_returns_five_entries(self):
chain = _get_provider_chain()
assert len(chain) == 5
labels = [label for label, _ in chain]
assert labels == ["openrouter", "nous", "local/custom", "openai-codex", "api-key"]
def test_picks_up_patched_functions(self):
"""Patches on _try_* functions must be visible in the chain."""
sentinel = lambda: ("patched", "model")
with patch("agent.auxiliary_client._try_openrouter", sentinel):
chain = _get_provider_chain()
assert chain[0] == ("openrouter", sentinel)
class TestTryPaymentFallback:
"""_try_payment_fallback skips the failed provider and tries alternatives."""
def test_skips_failed_provider(self):
mock_client = MagicMock()
with patch("agent.auxiliary_client._try_openrouter", return_value=(None, None)), \
patch("agent.auxiliary_client._try_nous", return_value=(mock_client, "nous-model")), \
patch("agent.auxiliary_client._read_main_provider", return_value="openrouter"):
client, model, label = _try_payment_fallback("openrouter", task="compression")
assert client is mock_client
assert model == "nous-model"
assert label == "nous"
def test_returns_none_when_no_fallback(self):
with patch("agent.auxiliary_client._try_openrouter", return_value=(None, None)), \
patch("agent.auxiliary_client._try_nous", return_value=(None, None)), \
patch("agent.auxiliary_client._try_custom_endpoint", return_value=(None, None)), \
patch("agent.auxiliary_client._try_codex", return_value=(None, None)), \
patch("agent.auxiliary_client._resolve_api_key_provider", return_value=(None, None)), \
patch("agent.auxiliary_client._read_main_provider", return_value="openrouter"):
client, model, label = _try_payment_fallback("openrouter")
assert client is None
assert label == ""
def test_codex_alias_maps_to_chain_label(self):
"""'codex' should map to 'openai-codex' in the skip set."""
mock_client = MagicMock()
with patch("agent.auxiliary_client._try_openrouter", return_value=(mock_client, "or-model")), \
patch("agent.auxiliary_client._try_codex", return_value=(None, None)), \
patch("agent.auxiliary_client._read_main_provider", return_value="openai-codex"):
client, model, label = _try_payment_fallback("openai-codex", task="vision")
assert client is mock_client
assert label == "openrouter"
def test_skips_to_codex_when_or_and_nous_fail(self):
mock_codex = MagicMock()
with patch("agent.auxiliary_client._try_openrouter", return_value=(None, None)), \
patch("agent.auxiliary_client._try_nous", return_value=(None, None)), \
patch("agent.auxiliary_client._try_custom_endpoint", return_value=(None, None)), \
patch("agent.auxiliary_client._try_codex", return_value=(mock_codex, "gpt-5.2-codex")), \
patch("agent.auxiliary_client._read_main_provider", return_value="openrouter"):
client, model, label = _try_payment_fallback("openrouter")
assert client is mock_codex
assert model == "gpt-5.2-codex"
assert label == "openai-codex"
class TestCallLlmPaymentFallback:
"""call_llm() retries with a different provider on 402 / payment errors."""
def _make_402_error(self, msg="Payment Required: insufficient credits"):
exc = Exception(msg)
exc.status_code = 402
return exc
def test_402_triggers_fallback(self, monkeypatch):
"""When the primary provider returns 402, call_llm tries the next one."""
monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
primary_client = MagicMock()
primary_client.chat.completions.create.side_effect = self._make_402_error()
fallback_client = MagicMock()
fallback_response = MagicMock()
fallback_client.chat.completions.create.return_value = fallback_response
with patch("agent.auxiliary_client._get_cached_client",
return_value=(primary_client, "google/gemini-3-flash-preview")), \
patch("agent.auxiliary_client._resolve_task_provider_model",
return_value=("openrouter", "google/gemini-3-flash-preview", None, None)), \
patch("agent.auxiliary_client._try_payment_fallback",
return_value=(fallback_client, "gpt-5.2-codex", "openai-codex")) as mock_fb:
result = call_llm(
task="compression",
messages=[{"role": "user", "content": "hello"}],
)
assert result is fallback_response
mock_fb.assert_called_once_with("openrouter", "compression")
# Fallback call should use the fallback model
fb_kwargs = fallback_client.chat.completions.create.call_args.kwargs
assert fb_kwargs["model"] == "gpt-5.2-codex"
def test_non_payment_error_not_caught(self, monkeypatch):
"""Non-payment errors (500, connection, etc.) should NOT trigger fallback."""
monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
primary_client = MagicMock()
server_err = Exception("Internal Server Error")
server_err.status_code = 500
primary_client.chat.completions.create.side_effect = server_err
with patch("agent.auxiliary_client._get_cached_client",
return_value=(primary_client, "google/gemini-3-flash-preview")), \
patch("agent.auxiliary_client._resolve_task_provider_model",
return_value=("openrouter", "google/gemini-3-flash-preview", None, None)):
with pytest.raises(Exception, match="Internal Server Error"):
call_llm(
task="compression",
messages=[{"role": "user", "content": "hello"}],
)
def test_402_with_no_fallback_reraises(self, monkeypatch):
"""When 402 hits and no fallback is available, the original error propagates."""
monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
primary_client = MagicMock()
primary_client.chat.completions.create.side_effect = self._make_402_error()
with patch("agent.auxiliary_client._get_cached_client",
return_value=(primary_client, "google/gemini-3-flash-preview")), \
patch("agent.auxiliary_client._resolve_task_provider_model",
return_value=("openrouter", "google/gemini-3-flash-preview", None, None)), \
patch("agent.auxiliary_client._try_payment_fallback",
return_value=(None, None, "")):
with pytest.raises(Exception, match="insufficient credits"):
call_llm(
task="compression",
messages=[{"role": "user", "content": "hello"}],
)