Jido.AI.Turn (Jido AI v2.1.0)

View Source

Canonical representation of a single LLM turn.

A turn captures the normalized response shape consumed by strategies and directives:

  • Response classification (:tool_calls or :final_answer)
  • Extracted text and optional thinking content
  • Normalized tool calls
  • Usage/model metadata
  • Optional executed tool results

Summary

Types

execute_opts()
execute_result()
response_type()
run_opts()
t()
tool_result()
tool_result_content()
tools_map()

Functions

assistant_message(turn)

Projects the turn into an assistant message compatible with ReqLLM context.

build_tools_map(module)

Builds a tools map from action modules.

execute(tool_name, params, context, opts \\ [])

Executes a tool by name using a tools map and returns raw action output.

execute_module(module, params, context, opts \\ [])

Executes an action module directly without registry lookup.

extract_from_content(content)

Extracts text from a content value (not wrapped in response structure).

extract_text(content)

Extracts text content from an LLM response or content value.

format_tool_result_content(other)

Formats a raw tool execution result into tool message content.

from_response(response, opts \\ [])

Builds a turn from a ReqLLM response.

from_result_map(turn)

Builds a turn from a map that is already in classified result shape.

needs_tools?(turn)

Returns true when the turn requests tool execution.

normalize_params(params, schema)

Normalizes parameters from LLM format to schema-compliant format.

run_tool_calls(tool_calls, context, opts \\ [])

Executes normalized tool calls and returns normalized tool results.

run_tools(turn, context, opts \\ [])

Executes all requested tools for the turn and returns the updated turn.

to_result_map(turn)

Converts a turn to a plain result map for public action/plugin outputs.

tool_messages(tool_results)

Projects tool results into role: :tool messages.

with_tool_results(turn, tool_results)

Returns a copy of the turn with normalized tool results attached.

Types

execute_opts()

@type execute_opts() :: [
  timeout: pos_integer() | nil,
  tools: tools_map() | [module()] | module() | nil
]

execute_result()

@type execute_result() :: {:ok, term(), [term()]} | {:error, term(), [term()]}

response_type()

@type response_type() :: :tool_calls | :final_answer

run_opts()

@type run_opts() :: [
  timeout: pos_integer() | nil,
  tools: map() | [module()] | module() | nil
]

t()

@type t() :: %Jido.AI.Turn{
  message_metadata: map(),
  model: String.t() | nil,
  reasoning_details: list() | nil,
  text: String.t(),
  thinking_content: String.t() | nil,
  tool_calls: [term()],
  tool_results: [tool_result()],
  type: response_type(),
  usage: map() | nil
}

tool_result()

@type tool_result() :: %{
  id: String.t(),
  name: String.t(),
  content: tool_result_content(),
  raw_result: execute_result()
}

tool_result_content()

@type tool_result_content() :: String.t() | [ReqLLM.Message.ContentPart.t()]

tools_map()

@type tools_map() :: %{required(String.t()) => module()}

Functions

assistant_message(turn)

@spec assistant_message(t()) :: ReqLLM.Message.t()

Projects the turn into an assistant message compatible with ReqLLM context.

build_tools_map(module)

@spec build_tools_map(module() | [module()]) :: tools_map()

Builds a tools map from action modules.

execute(tool_name, params, context, opts \\ [])

@spec execute(String.t(), map(), map(), execute_opts()) :: execute_result()

Executes a tool by name using a tools map and returns raw action output.

execute_module(module, params, context, opts \\ [])

@spec execute_module(module(), map(), map(), execute_opts()) :: execute_result()

Executes an action module directly without registry lookup.

extract_from_content(content)

@spec extract_from_content(term()) :: String.t()

Extracts text from a content value (not wrapped in response structure).

extract_text(content)

@spec extract_text(term()) :: String.t()

Extracts text content from an LLM response or content value.

This supports the canonical response/content normalization shapes used across actions and strategy flows.

format_tool_result_content(other)

@spec format_tool_result_content(execute_result() | {:ok, term()} | {:error, term()}) ::
  tool_result_content()

Formats a raw tool execution result into tool message content.

from_response(response, opts \\ [])

@spec from_response(
  map() | ReqLLM.Response.t() | t(),
  keyword()
) :: t()

Builds a turn from a ReqLLM response.

Options:

  • :model - Override model from the response payload

from_result_map(turn)

@spec from_result_map(map() | t()) :: t()

Builds a turn from a map that is already in classified result shape.

needs_tools?(turn)

@spec needs_tools?(t()) :: boolean()

Returns true when the turn requests tool execution.

normalize_params(params, schema)

@spec normalize_params(map(), keyword() | struct()) :: map()

Normalizes parameters from LLM format to schema-compliant format.

run_tool_calls(tool_calls, context, opts \\ [])

@spec run_tool_calls([term()], map(), run_opts()) :: {:ok, [tool_result()]}

Executes normalized tool calls and returns normalized tool results.

run_tools(turn, context, opts \\ [])

@spec run_tools(t(), map(), run_opts()) :: {:ok, t()} | {:error, term()}

Executes all requested tools for the turn and returns the updated turn.

to_result_map(turn)

@spec to_result_map(t()) :: map()

Converts a turn to a plain result map for public action/plugin outputs.

tool_messages(tool_results)

@spec tool_messages(t() | [map()]) :: [ReqLLM.Message.t()]

Projects tool results into role: :tool messages.

with_tool_results(turn, tool_results)

@spec with_tool_results(t(), [map()]) :: t()

Returns a copy of the turn with normalized tool results attached.