Dhenara AI is an open source Python package for calling multiple LLM providers through one typed interface. It keeps the integration surface small: provider credentials, model selection, prompt/message formatting, streaming, tool use, and structured output all flow through the same core client types.

For full documentation, visit docs.dhenara.com.

pip install dhenara-ai

Dhenara AI discovers credentials in one of two ways:

1. Pass an explicit file path to ResourceConfig.load_from_file().
2. Set DAI_SECRET_CONFIG_DIR and place dai_credentials.yaml inside that directory.

If DAI_SECRET_CONFIG_DIR is unset, the default location is /run/secrets/dai/dai_credentials.yaml.

Use the checked-in template at src/dhenara/ai/types/resource/credentials.yaml as the starting point.

mkdir -p /path/to/secrets
export DAI_SECRET_CONFIG_DIR=/path/to/secrets
cp src/dhenara/ai/types/resource/credentials.yaml "$DAI_SECRET_CONFIG_DIR/dai_credentials.yaml"

Then edit dai_credentials.yaml and keep only the providers you actually use.

from dhenara.ai import AIModelClient
from dhenara.ai.types import AIModelAPIProviderEnum, AIModelCallConfig, ResourceConfig

resource_config = ResourceConfig()
resource_config.load_from_file(credentials_file=None, init_endpoints=True)

endpoint = resource_config.get_model_endpoint(
    model_name="claude-haiku-4-5",
    api_provider=AIModelAPIProviderEnum.ANTHROPIC,
)
if endpoint is None:
    raise RuntimeError("No Anthropic endpoint configured for claude-haiku-4-5")

client = AIModelClient(
    model_endpoint=endpoint,
    config=AIModelCallConfig(
        max_output_tokens=1024,
        reasoning=False,
        streaming=False,
    ),
    is_async=False,
)

response = client.generate(
    prompt="Explain quantum computing in simple terms.",
    instructions=["Keep the answer under 120 words."],
)

if response.chat_response:
    print(response.chat_response.text())

The example programs use the same credential-loading contract. With a secret directory configured, you can run them directly:

export DAI_SECRET_CONFIG_DIR=/path/to/secrets
python examples/14_multi_turn_with_messages_api.py

If you prefer an explicit credentials file, pass that path where the example calls load_from_file().

• OpenAI direct and Azure OpenAI / Microsoft Foundry OpenAI v1 use the openai SDK.
• Google Gemini Developer API and Gemini on Vertex AI use the google-genai SDK.
• Anthropic direct, Amazon Bedrock, and Anthropic on Vertex use the anthropic SDK.
• Amazon Bedrock configuration stays under the amazon_bedrock provider block. The package passes those credentials into AnthropicBedrock; there is no separate boto client setup in your application code.
• microsoft_azure_ai is not a supported text-generation surface in dhenara-ai; use microsoft_openai with an Azure OpenAI or Microsoft Foundry OpenAI v1 endpoint instead.
• For Microsoft-hosted OpenAI-compatible deployments, the model value in requests must be the Azure deployment name, not just the underlying vendor model family.