View Source Anthropic.ToolRunner (anthropic_community v0.5.0)

Drives Anthropic.Messages.create/2 in a loop, executing tool calls the assistant requests until it produces a final answer (stop_reason other than "tool_use").

Example

client = Anthropic.Client.new(api_key: System.fetch_env!("ANTHROPIC_API_KEY"))

{:ok, message, _history} =
  Anthropic.ToolRunner.run(
    client,
    [model: "claude-opus-4-8", max_tokens: 1024,
     messages: [%{role: "user", content: "What's the weather in Paris?"}]],
    [MyApp.WeatherTool]
  )

Summary

Functions

run(client, opts, tools, max_iterations \\ 25)

Runs client/opts through Anthropic.Messages.create/2, executing any requested tools and feeding results back, until the assistant stops requesting tools (or max_iterations is exceeded).

Functions

run(client, opts, tools, max_iterations \\ 25)

@spec run(
  Anthropic.Client.t(),
  Anthropic.Messages.create_opts(),
  [module()],
  pos_integer()
) ::
  {:ok, Anthropic.Messages.Message.t(), [map()]} | {:error, Anthropic.Error.t()}

Runs client/opts through Anthropic.Messages.create/2, executing any requested tools and feeding results back, until the assistant stops requesting tools (or max_iterations is exceeded).

Returns {:ok, final_message, messages_history} on success, where messages_history is the full messages param list (including every assistant tool_use turn and user tool_result turn) so the caller can continue the conversation. Returns {:error, %Anthropic.Error{}} if the underlying API call fails or the loop exceeds max_iterations.