Arcana.Loop.Context (Arcana v2.0.1)

Copy Markdown View Source

State carried through an Arcana.Loop run.

This is the loop's working memory. It accumulates chunks across iterations, records the tool history, and tracks termination state.

Fields

  • :question - The original user question.
  • :repo - Ecto repo to use for retrieval tools.
  • :collections - Collections to scope retrieval tools to (list of names).
  • :messages - The ReqLLM.Context carrying the running conversation.
  • :chunks - Accumulated chunks from search tool calls (capped via :chunk_cap).
  • :tool_history - List of %{tool, args, iteration, summary} entries in call order.
  • :iterations - Number of LLM controller turns executed.
  • :answer - Final answer text once the loop terminates.
  • :terminated_by - One of :answered, :gave_up, :max_iterations, :error, or nil.
  • :error - Error reason when terminated_by == :error.
  • :grounding - Populated by Arcana.Loop.ground/2 with an %Arcana.Grounding.Result{} scoring the answer against ctx.chunks. nil until grounding runs.

Summary

Types

t()
tool_history_entry()

Types

t()

@type t() :: %Arcana.Loop.Context{
  answer: String.t() | nil,
  chunks: [map()],
  collections: [String.t() | nil],
  error: term() | nil,
  grounding: Arcana.Grounding.Result.t() | nil,
  iterations: non_neg_integer(),
  messages: term() | nil,
  question: String.t(),
  repo: module() | nil,
  terminated_by: atom() | nil,
  tool_history: [tool_history_entry()]
}

tool_history_entry()

@type tool_history_entry() :: %{
  tool: atom(),
  args: map(),
  iteration: non_neg_integer(),
  summary: String.t()
}