Jidoka.Runtime.ReqLLM (Jidoka v0.8.0-beta.1)

ReqLLM runtime support for Jidoka's LLM effect boundary.

The runtime uses a constrained JSON protocol instead of native provider tool-calling. That keeps Jidoka's Runic spine provider-neutral while still letting a real model choose between final answers and operation calls.

Summary

Types

option()

Functions

llm(opts \\ [])

Returns an LLM function suitable for Jidoka.turn/3.

Types

option()

@type option() ::
  {:model, ReqLLM.model_input()}
  | {:temperature, number()}
  | {:max_tokens, pos_integer()}
  | {:timeout, timeout()}
  | {:receive_timeout, timeout()}
  | {:provider_options, map()}
  | {:cache, term()}
  | {:api_key, String.t()}
  | {:stream, boolean()}
  | {:stream_to, pid() | {:pid, pid()}}
  | {:on_event, (Jidoka.Event.t() -> term())}

Functions

llm(opts \\ [])

@spec llm([option()]) :: Jidoka.Runtime.Capabilities.llm_capability()

Returns an LLM function suitable for Jidoka.turn/3.

llm = Jidoka.Runtime.ReqLLM.llm(model: "openai:gpt-4o-mini", temperature: 0.0)
Jidoka.turn(agent, "Use the available tool.", llm: llm, operations: ops)