OllamaLLMClient#

OllamaLLMClient targets local/self-hosted Ollama chat inference.

Default behavior#

  • Default managed mode: manage_server=True

  • Default model: qwen2.5:1.5b-instruct

  • Default managed endpoint: http://127.0.0.1:11434

Constructor-first usage#

from design_research_agents import OllamaLLMClient
from design_research_agents.llm import LLMMessage, LLMRequest

with OllamaLLMClient() as client:
    response = client.generate(
        LLMRequest(
            messages=(LLMMessage(role="user", content="Summarize one design principle."),),
            model=client.default_model(),
        )
    )

Prefer the context-manager form so managed local runtime processes shut down deterministically. close() remains available for explicit lifecycle control.

Dependencies and environment#

  • Install and run Ollama locally (ollama serve) if using connect mode.

  • Managed mode starts ollama serve automatically using the configured ollama_executable.

  • Optional model prefetch: auto_pull_model=True.

Examples#

  • examples/clients/ollama_local_client.py

Attribution#