LLM Providers

Every LLM in Dynamiq is a node class in dynamiq.nodes.llms, and all of them subclass the same BaseLLM. Swapping OpenAI for Anthropic, Gemini, or a self-hosted Ollama model means changing the class, the model name, and the connection — the prompt, parameters, streaming, tools, and output shape stay identical.

Quick example

Each provider node pairs with a connection class of the same name in dynamiq.connections. Connections read their credentials from environment variables by default, so the minimal setup is one import and one env var:

# export OPENAI_API_KEY=...
from dynamiq.connections import OpenAI as OpenAIConnection
from dynamiq.nodes.llms import OpenAI
from dynamiq.prompts import Message, Prompt

llm = OpenAI(
    connection=OpenAIConnection(),
    model="gpt-4o-mini",
    prompt=Prompt(messages=[Message(role="user", content="{{question}}")]),
)

result = llm.run(input_data={"question": "What is an embedding?"})
print(result.output["content"])

# export ANTHROPIC_API_KEY=...
from dynamiq.connections import Anthropic as AnthropicConnection
from dynamiq.nodes.llms import Anthropic
from dynamiq.prompts import Message, Prompt

llm = Anthropic(
    connection=AnthropicConnection(),
    model="claude-3-5-sonnet-20240620",
    prompt=Prompt(messages=[Message(role="user", content="{{question}}")]),
)

result = llm.run(input_data={"question": "What is an embedding?"})
print(result.output["content"])

# No API key needed — Ollama runs locally.
from dynamiq.connections import Ollama as OllamaConnection
from dynamiq.nodes.llms import Ollama
from dynamiq.prompts import Message, Prompt

llm = Ollama(
    connection=OllamaConnection(url="http://localhost:11434"),
    model="llama3",
    prompt=Prompt(messages=[Message(role="user", content="{{question}}")]),
)

result = llm.run(input_data={"question": "What is an embedding?"})
print(result.output["content"])

The output dictionary always contains content (the generated text) and, when the model calls tools, a tool_calls list with parsed JSON arguments.

Provider table

All classes are imported from dynamiq.nodes.llms; the matching connection classes live in dynamiq.connections. Dynamiq dispatches requests through litellm and prepends the model prefix automatically — OpenAI(model="gpt-4o") and OpenAI(model="openai/gpt-4o") are equivalent.

CustomLLM targets any OpenAI-compatible endpoint: give it an HttpApiKey connection with the base url and api_key, and set provider_prefix (for example "openai") so the request is routed with the right protocol. This is also how you point a Dynamiq LLM node at the platform's AI Gateway.

Common parameters

Because every node inherits BaseLLM, these parameters work across all providers:

Unknown extra keyword arguments are passed through to the underlying completion call, so provider-specific parameters (for example OpenAI's reasoning_effort, which the OpenAI node also exposes directly) remain available.

Vision and file input

A node reports its multimodal capabilities through two properties:

from dynamiq.connections import OpenAI as OpenAIConnection
from dynamiq.nodes.llms import OpenAI

llm = OpenAI(connection=OpenAIConnection(), model="gpt-4o-mini")
print(llm.is_vision_supported)     # True
print(llm.is_pdf_input_supported)  # model-dependent

Capability and token-limit lookups come from litellm's model database, with a bundled registry (dynamiq/nodes/llms/model_registry.json) as a fallback for models litellm does not know. For self-hosted or brand-new models missing from both, override the metadata yourself:

from dynamiq.connections import OpenAI as OpenAIConnection
from dynamiq.nodes.llms import OpenAI
from dynamiq.nodes.llms.base import ModelInfo

llm = OpenAI(
    connection=OpenAIConnection(url="https://my-gateway.internal/v1"),
    model="my-finetuned-model",
    model_info=ModelInfo(max_input_tokens=128000, supports_vision=True),
)

To actually send images or files, build a VisionMessage prompt — see Prompts & Messages.

Fallbacks

Attach a FallbackConfig to run a secondary LLM when the primary fails. Triggers can be any error, rate_limit, or connection:

from dynamiq.connections import Anthropic as AnthropicConnection
from dynamiq.connections import OpenAI as OpenAIConnection
from dynamiq.nodes.llms import Anthropic, OpenAI
from dynamiq.nodes.llms.base import FallbackConfig, FallbackTrigger
from dynamiq.prompts import Message, Prompt

backup = Anthropic(connection=AnthropicConnection(), model="claude-3-5-sonnet-20240620")

llm = OpenAI(
    connection=OpenAIConnection(),
    model="gpt-4o",
    prompt=Prompt(messages=[Message(role="user", content="{{question}}")]),
    fallback=FallbackConfig(
        llm=backup,
        enabled=True,
        triggers=[FallbackTrigger.RATE_LIMIT, FallbackTrigger.CONNECTION],
    ),
)

result = llm.run(input_data={"question": "Summarize RAG in one sentence."})
print(result.output["content"])

The fallback receives the same prepared input as the primary, and the primary node's output transformer is applied to the fallback's result, so downstream nodes see no difference.

Next steps