feat(config): accept GROQ_API_KEY for Groq-hosted models

README.md

@@ -235,6 +235,15 @@ export PERPLEXITY_API_KEY="your-api-key"  # for search capabilities
 export STRIX_REASONING_EFFORT="high"  # control thinking effort (default: high, quick scan: medium)
 ```
 
+Provider-native keys are also picked up automatically. For example, to use
+[Groq's](https://console.groq.com/) fast inference you can set `GROQ_API_KEY`
+directly instead of `LLM_API_KEY`:
+
+```bash
+export STRIX_LLM="groq/llama-3.3-70b-versatile"
+export GROQ_API_KEY="your-api-key"
+```
+
 > [!NOTE]
 > Strix automatically saves your configuration to `~/.strix/cli-config.json`, so you don't have to re-enter it on every run.
 

strix/config/models.py

@@ -65,6 +65,10 @@ def configure_sdk_model_defaults(settings: Settings) -> None:
     llm = settings.llm
     set_tracing_disabled(True)
     _configure_litellm_compatibility()
+    if llm.groq_api_key:
+        # LiteLLM reads GROQ_API_KEY natively; export it (without touching the
+        # generic api_key) so persisted config reaches the provider too.
+        os.environ.setdefault("GROQ_API_KEY", llm.groq_api_key)
     if llm.api_key:
         set_default_openai_key(llm.api_key, use_for_tracing=False)
         _configure_litellm_default("api_key", llm.api_key)

strix/config/settings.py

@@ -25,6 +25,10 @@ class LlmSettings(BaseSettings):
         default=None,
         validation_alias=AliasChoices("LLM_API_KEY", "OPENAI_API_KEY"),
     )
+    # Provider-native key kept separate from the generic ``api_key`` so it is only
+    # ever exported as ``GROQ_API_KEY`` and never mistaken for another provider's
+    # credential (e.g. mirrored into ``OPENAI_API_KEY``).
+    groq_api_key: str | None = Field(default=None, alias="GROQ_API_KEY")
     api_base: str | None = Field(
         default=None,
         validation_alias=AliasChoices(

tests/test_config_loader.py

@@ -20,6 +20,7 @@
     "STRIX_LLM",
     "LLM_API_KEY",
     "OPENAI_API_KEY",
+    "GROQ_API_KEY",
     "LLM_API_BASE",
     "OPENAI_API_BASE",
     "OPENAI_BASE_URL",
@@ -150,6 +151,29 @@ def test_apply_config_override_invalidates_cache(tmp_path: Path) -> None:
     assert loader.load_settings().llm.model == "second-model"
 
 
+# --------------------------------------------------------------------------- #
+# LLM provider keys
+# --------------------------------------------------------------------------- #
+
+
+def test_groq_api_key_populates_dedicated_field(monkeypatch: pytest.MonkeyPatch) -> None:
+    # GROQ_API_KEY must land on its own field, never the generic api_key, so it
+    # can't be mirrored into another provider's credential.
+    monkeypatch.setenv("GROQ_API_KEY", "gsk-test")
+    llm = loader.load_settings().llm
+    assert llm.groq_api_key == "gsk-test"
+    assert llm.api_key is None
+
+
+def test_groq_and_openai_keys_do_not_collide(monkeypatch: pytest.MonkeyPatch) -> None:
+    # Both set: each key stays on its own field instead of one overriding the other.
+    monkeypatch.setenv("OPENAI_API_KEY", "sk-openai")
+    monkeypatch.setenv("GROQ_API_KEY", "gsk-test")
+    llm = loader.load_settings().llm
+    assert llm.api_key == "sk-openai"
+    assert llm.groq_api_key == "gsk-test"
+
+
 # --------------------------------------------------------------------------- #
 # persist_current
 # --------------------------------------------------------------------------- #