Class: OllamaAgent::Skills::LlmClient

Inherits:
Object
  • Object
show all
Defined in:
lib/ollama_agent/skills/llm_client.rb

Overview

Thin facade over Providers::Registry tuned for deterministic, single-shot JSON generation. Defaults to the local Ollama provider so skills stay local-first and auditable. Inject any object responding to #generate in tests.

Constant Summary collapse

DEFAULT_TEMPERATURE = 
0.0
DEFAULT_PROVIDER = 
"ollama"

Instance Method Summary collapse

Constructor Details

#initialize(provider: nil, model: nil, temperature: DEFAULT_TEMPERATURE) ⇒ LlmClient

Returns a new instance of LlmClient.



15
16
17
18
19
 
# File 'lib/ollama_agent/skills/llm_client.rb', line 15

def initialize(provider: nil, model: nil, temperature: DEFAULT_TEMPERATURE)
  @provider    = provider || Providers::Registry.resolve(DEFAULT_PROVIDER)
  @model       = model || ENV.fetch("OLLAMA_AGENT_SKILL_MODEL", default_model)
  @temperature = temperature
end

Instance Method Details

#generate(prompt) ⇒ String

Returns raw assistant content.

Parameters:

  • prompt (String)

Returns:

  • (String) —

    raw assistant content

Raises:



23
24
25
26
27
28
29
 
# File 'lib/ollama_agent/skills/llm_client.rb', line 23

def generate(prompt)
  response = @provider.chat(messages: [user_message(prompt)], model: @model, temperature: @temperature)
  content  = response.content
  raise OllamaAgent::Error, "empty response from provider" if content.to_s.strip.empty?

  content
end