fix: auto-detect OpenAI embedding provider when API key available (#72)

[zonk1024]

Problem

The embedding_provider config defaults to "local" but local embeddings aren't set up in most environments. When OPENAI_API_KEY is available, the provider should auto-select "openai". The MCP/server path worked because env vars were set at server startup, but direct Python API usage defaulted to "local" — causing failures in batch usage.

Fix

Added a Pydantic model_validator(mode='after') to CoreSettings in src/valence/core/config.py:

• If embedding_provider is still the default "local" and OPENAI_API_KEY is set and VALENCE_EMBEDDING_PROVIDER was not explicitly configured → auto-select "openai"
• Explicit user configuration via VALENCE_EMBEDDING_PROVIDER is always respected (no surprise overrides)

Tests

Added TestEmbeddingProviderAutoDetect in tests/core/test_config.py:

1. OPENAI_API_KEY set + no VALENCE_EMBEDDING_PROVIDER → provider becomes "openai"
2. VALENCE_EMBEDDING_PROVIDER=local explicit + OPENAI_API_KEY set → stays "local" (user override)
3. No key, no explicit provider → stays "local"

All 1716 tests pass (10 skipped).

Closes ourochronos/tracking#72

[Copilot]

Pull request overview

This PR updates Valence’s configuration so CoreSettings.embedding_provider auto-selects "openai" when OPENAI_API_KEY is present and the provider was not explicitly configured, preventing unexpected failures when using the Python API with default settings.

Changes:

• Added a @model_validator(mode="after") in CoreSettings to auto-select the OpenAI embedding provider when an API key is available.
• Added unit tests covering env-var-driven auto-detection and explicit env override behavior.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 3 comments.

File	Description
src/valence/core/config.py	Adds an “after” model validator to auto-select embedding_provider="openai" when OPENAI_API_KEY is set.
tests/core/test_config.py	Adds tests validating auto-detection and respecting explicit env configuration.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

These tests cover env-var-based overrides, but don’t cover the common direct-Python usage where a caller might explicitly pass embedding_provider via CoreSettings(embedding_provider=...). Given the validator logic, adding a test that an explicit embedding_provider="local" (via kwargs) is not auto-overridden when OPENAI_API_KEY is set would help prevent regressions.

Suggested change

	def test_kwarg_local_override_respected(self, monkeypatch, clean_env):
	"""When embedding_provider='local' is passed via kwargs, it should stay
	'local' even if OPENAI_API_KEY is set (direct Python override)."""
	monkeypatch.setenv("OPENAI_API_KEY", "sk-test-kwarg-local")
	monkeypatch.delenv("VALENCE_EMBEDDING_PROVIDER", raising=False)
	settings = CoreSettings(embedding_provider="local")
	assert settings.embedding_provider == "local"

[Copilot]

The explicit-configuration check relies on os.environ.get("VALENCE_EMBEDDING_PROVIDER"), but BaseSettings values can come from non-os.environ sources (e.g., passing embedding_provider="local" to CoreSettings(...), or values loaded from an env_file in non-test runs). In those cases this validator can incorrectly override an explicit local setting to openai. Prefer detecting whether the field was provided via settings inputs (e.g., using self.model_fields_set to see if embedding_provider was set) rather than inspecting os.environ directly.

Suggested change

	import os
	# Check if VALENCE_EMBEDDING_PROVIDER was explicitly set
	if not os.environ.get("VALENCE_EMBEDDING_PROVIDER"):
	# Only auto-select when the embedding_provider field was *not*
	# explicitly provided via any settings source.
	if "embedding_provider" not in self.model_fields_set:

[Copilot]

There’s a section divider comment indented under the validator after return self, which is unreachable and makes the file structure confusing. Consider removing it or unindenting it so the section headers align consistently with the rest of the module.

Suggested change

	# ==========================================================================
	# ==========================================================================