Model providers: OpenAI/xAI/Ollama + run several at once (registry)

Extends the #215 abstraction:
- OpenAICompatibleLLMProvider / OpenAICompatibleEmbeddingProvider — one impl (via
  the official openai SDK) covers OpenAI, xAI (api.x.ai/v1), Ollama
  (…:11434/v1), OpenRouter, etc.; they differ only by base_url, key, and model.
- Registry factory: build_llm_providers() / build_embedding_providers() return
  every provider whose credentials are configured, so you can run several
  concurrently. get_llm_provider(name)/get_embedding_provider(name) select by
  name, falling back to default_*_provider, then Null.
- Per-provider env config (ANTHROPIC_*, OPENAI_*, XAI_*, OLLAMA_*) +
  DEFAULT_LLM_PROVIDER / DEFAULT_EMBEDDING_PROVIDER; documented in .env.example.
  Defaults keep AI off (empty registry).

Embeddings now have real backends (OpenAI/Ollama), still separate from the LLM
since Anthropic offers no embeddings endpoint. Tests cover multi-provider
selection, default resolution, disabled-without-credentials, and null fail-loud.
Full suite 87 passed.

Relates to #215.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Signed-off-by: Justin Paul <justin@jpaul.me>

backend/tests/test_model_providers.py

@@ -1,43 +1,84 @@
-"""Model-provider selection + the null-provider fail-loud behavior.
-
-No network: we only assert which provider the factory returns and that the null
-providers raise a clear error. (Live LLM/embedding calls aren't unit-tested.)
+"""Model-provider registry: configure several vendors at once, select by name,
+default selection, and the null fail-loud behavior. No network — we only assert
+which provider the factory returns and that null providers raise.
 """
 
 import pytest
 
-from app.api.deps import get_embedding_provider, get_llm_provider
+from app.api.deps import (
+    build_embedding_providers,
+    build_llm_providers,
+    get_embedding_provider,
+    get_llm_provider,
+)
 from app.core.config import get_settings
 from app.integrations.models.anthropic_provider import AnthropicLLMProvider
 from app.integrations.models.base import ModelProviderNotConfigured
 from app.integrations.models.null import NullEmbeddingProvider, NullLLMProvider
+from app.integrations.models.openai_compat import (
+    OpenAICompatibleEmbeddingProvider,
+    OpenAICompatibleLLMProvider,
+)
 
 
-async def test_default_llm_is_null_and_fails_loud(monkeypatch):
-    monkeypatch.setattr(get_settings(), "model_provider", "null")
+def _reset(monkeypatch):
+    s = get_settings()
+    for attr, val in {
+        "default_llm_provider": "null",
+        "default_embedding_provider": "null",
+        "anthropic_api_key": None,
+        "openai_api_key": None,
+        "xai_api_key": None,
+        "ollama_enabled": False,
+    }.items():
+        monkeypatch.setattr(s, attr, val)
+    return s
+
+
+async def test_default_is_null_and_fails_loud(monkeypatch):
+    _reset(monkeypatch)
     provider = get_llm_provider()
     assert isinstance(provider, NullLLMProvider)
     with pytest.raises(ModelProviderNotConfigured):
         await provider.complete(prompt="hello")
+    assert isinstance(get_embedding_provider(), NullEmbeddingProvider)
 
 
-async def test_anthropic_selected_when_configured(monkeypatch):
-    s = get_settings()
-    monkeypatch.setattr(s, "model_provider", "anthropic")
-    monkeypatch.setattr(s, "anthropic_api_key", "sk-ant-test-key")
-    monkeypatch.setattr(s, "llm_model", "claude-opus-4-8")
-    assert isinstance(get_llm_provider(), AnthropicLLMProvider)  # no network call
+async def test_multiple_llm_providers_at_once(monkeypatch):
+    s = _reset(monkeypatch)
+    monkeypatch.setattr(s, "anthropic_api_key", "sk-ant-x")
+    monkeypatch.setattr(s, "openai_api_key", "sk-openai-x")
+    monkeypatch.setattr(s, "xai_api_key", "xai-x")
+    monkeypatch.setattr(s, "ollama_enabled", True)
+    monkeypatch.setattr(s, "default_llm_provider", "anthropic")
+
+    registry = build_llm_providers()
+    assert set(registry) == {"anthropic", "openai", "xai", "ollama"}
+    # Select any by name.
+    assert isinstance(get_llm_provider("anthropic"), AnthropicLLMProvider)
+    assert isinstance(get_llm_provider("openai"), OpenAICompatibleLLMProvider)
+    assert isinstance(get_llm_provider("xai"), OpenAICompatibleLLMProvider)
+    assert isinstance(get_llm_provider("ollama"), OpenAICompatibleLLMProvider)
+    # Default resolves to the configured default.
+    assert isinstance(get_llm_provider(), AnthropicLLMProvider)
+    # Unknown name → null.
+    assert isinstance(get_llm_provider("nope"), NullLLMProvider)
 
 
-async def test_anthropic_without_key_falls_back_to_null(monkeypatch):
-    s = get_settings()
-    monkeypatch.setattr(s, "model_provider", "anthropic")
-    monkeypatch.setattr(s, "anthropic_api_key", None)
+async def test_provider_disabled_without_credentials(monkeypatch):
+    s = _reset(monkeypatch)
+    monkeypatch.setattr(s, "default_llm_provider", "openai")  # default names openai…
+    # …but no openai key → registry empty → null fallback.
+    assert build_llm_providers() == {}
     assert isinstance(get_llm_provider(), NullLLMProvider)
 
 
-async def test_embedding_default_is_null_and_fails_loud():
-    provider = get_embedding_provider()
-    assert isinstance(provider, NullEmbeddingProvider)
-    with pytest.raises(ModelProviderNotConfigured):
-        await provider.embed(["text"])
+async def test_embedding_providers(monkeypatch):
+    s = _reset(monkeypatch)
+    monkeypatch.setattr(s, "openai_api_key", "sk-openai-x")
+    monkeypatch.setattr(s, "ollama_enabled", True)
+    monkeypatch.setattr(s, "default_embedding_provider", "openai")
+    registry = build_embedding_providers()
+    assert set(registry) == {"openai", "ollama"}
+    assert isinstance(get_embedding_provider(), OpenAICompatibleEmbeddingProvider)
+    assert isinstance(get_embedding_provider("ollama"), OpenAICompatibleEmbeddingProvider)