Allow to use credentials from environment variables (#2)

Author: vrslev

Justfile

@@ -2,7 +2,7 @@ default: install lint test
 
 install:
     uv lock --upgrade
-    uv sync --frozen
+    uv sync --frozen --all-groups
 
 lint:
     uv run --group lint ruff check

README.md

@@ -102,7 +102,16 @@ config = any_llm_client.MockLLMConfig(
 client = any_llm_client.get_client(config, ...)
 ```
 
-#### Using dynamic LLM config from environment with [pydantic-settings](https://docs.pydantic.dev/latest/concepts/pydantic_settings/)
+#### Configuration with environment variables
+
+##### Credentials
+
+Instead of passing credentials directly, you can set corresponding environment variables:
+
+- OpenAI: `ANY_LLM_CLIENT_OPENAI_AUTH_TOKEN`,
+- YandexGPT: `ANY_LLM_CLIENT_YANDEXGPT_AUTH_HEADER`, `ANY_LLM_CLIENT_YANDEXGPT_FOLDER_ID`.
+
+##### LLM model config (with [pydantic-settings](https://docs.pydantic.dev/latest/concepts/pydantic_settings/))
 
 ```python
 import os
@@ -125,6 +134,8 @@ settings = Settings()
 client = any_llm_client.get_client(settings.llm_model, ...)
 ```
 
+Combining with environment variables from previous section, you can keep LLM model configuration and secrets separate.
+
 #### Using clients directly
 
 The recommended way to get LLM client is to call `any_llm_client.get_client()`. This way you can easily swap LLM models. If you prefer, you can use `any_llm_client.OpenAIClient` or `any_llm_client.YandexGPTClient` directly:

any_llm_client/clients/openai.py

@@ -1,5 +1,6 @@
 import contextlib
 import dataclasses
+import os
 import types
 import typing
 from http import HTTPStatus
@@ -15,12 +16,15 @@
 from any_llm_client.retry import RequestRetryConfig
 
 
+OPENAI_AUTH_TOKEN_ENV_NAME: typing.Final = "ANY_LLM_CLIENT_OPENAI_AUTH_TOKEN"
+
+
 class OpenAIConfig(LLMConfig):
     if typing.TYPE_CHECKING:
         url: str  # pragma: no cover
     else:
         url: pydantic.HttpUrl
-    auth_token: str | None = None
+    auth_token: str | None = pydantic.Field(default_factory=lambda: os.environ.get(OPENAI_AUTH_TOKEN_ENV_NAME))
     model_name: str
     force_user_assistant_message_alternation: bool = False
     "Gemma 2 doesn't support {role: system, text: ...} message, and requires alternated messages"

any_llm_client/clients/yandexgpt.py

@@ -1,5 +1,6 @@
 import contextlib
 import dataclasses
+import os
 import types
 import typing
 from http import HTTPStatus
@@ -14,13 +15,21 @@
 from any_llm_client.retry import RequestRetryConfig
 
 
+YANDEXGPT_AUTH_HEADER_ENV_NAME: typing.Final = "ANY_LLM_CLIENT_YANDEXGPT_AUTH_HEADER"
+YANDEXGPT_FOLDER_ID_ENV_NAME: typing.Final = "ANY_LLM_CLIENT_YANDEXGPT_FOLDER_ID"
+
+
 class YandexGPTConfig(LLMConfig):
     if typing.TYPE_CHECKING:
         url: str = "https://llm.api.cloud.yandex.net/foundationModels/v1/completion"  # pragma: no cover
     else:
         url: pydantic.HttpUrl = "https://llm.api.cloud.yandex.net/foundationModels/v1/completion"
-    auth_header: str | None = None
-    folder_id: str | None = None
+    auth_header: str = pydantic.Field(  # type: ignore[assignment]
+        default_factory=lambda: os.environ.get(YANDEXGPT_AUTH_HEADER_ENV_NAME), validate_default=True
+    )
+    folder_id: str = pydantic.Field(  # type: ignore[assignment]
+        default_factory=lambda: os.environ.get(YANDEXGPT_FOLDER_ID_ENV_NAME), validate_default=True
+    )
     model_name: str
     model_version: str = "latest"
     max_tokens: int = 7400
@@ -79,10 +88,12 @@ def __init__(
         self.httpx_client = get_http_client_from_kwargs(httpx_kwargs)
 
     def _build_request(self, payload: dict[str, typing.Any]) -> httpx.Request:
-        headers: typing.Final = {"x-data-logging-enabled": "false"}
-        if self.config.auth_header:
-            headers["Authorization"] = self.config.auth_header
-        return self.httpx_client.build_request(method="POST", url=str(self.config.url), json=payload, headers=headers)
+        return self.httpx_client.build_request(
+            method="POST",
+            url=str(self.config.url),
+            json=payload,
+            headers={"Authorization": self.config.auth_header, "x-data-logging-enabled": "false"},
+        )
 
     def _prepare_payload(
         self, *, messages: str | list[Message], temperature: float = 0.2, stream: bool

tests/test_http.py

@@ -11,17 +11,17 @@ def test_http_timeout_is_added(self) -> None:
         original_kwargs: typing.Final = {"mounts": {"http://": None}}
         passed_kwargs: typing.Final = copy.deepcopy(original_kwargs)
 
-        result: typing.Final = get_http_client_from_kwargs(passed_kwargs)
+        client: typing.Final = get_http_client_from_kwargs(passed_kwargs)
 
-        assert result.timeout == DEFAULT_HTTP_TIMEOUT
+        assert client.timeout == DEFAULT_HTTP_TIMEOUT
         assert original_kwargs == passed_kwargs
 
     def test_http_timeout_is_not_modified_if_set(self) -> None:
         timeout: typing.Final = httpx.Timeout(7, connect=5, read=3)
         original_kwargs: typing.Final = {"mounts": {"http://": None}, "timeout": timeout}
         passed_kwargs: typing.Final = copy.deepcopy(original_kwargs)
 
-        result: typing.Final = get_http_client_from_kwargs(passed_kwargs)
+        client: typing.Final = get_http_client_from_kwargs(passed_kwargs)
 
-        assert result.timeout == timeout
+        assert client.timeout == timeout
         assert original_kwargs == passed_kwargs