# `Arcana.Agent.Decomposer`
[🔗](https://github.com/georgeguimaraes/arcana/blob/main/lib/arcana/agent/decomposer.ex#L1)

Behaviour for query decomposition in the Agent pipeline.

The decomposer breaks complex questions into simpler sub-questions
that can be searched independently, improving retrieval for multi-faceted queries.

## Built-in Implementations

- `Arcana.Agent.Decomposer.LLM` - Uses your LLM to decompose queries (default)

## Implementing a Custom Decomposer

    defmodule MyApp.KeywordDecomposer do
      @behaviour Arcana.Agent.Decomposer

      @impl true
      def decompose(question, _opts) do
        # Simple keyword-based decomposition
        sub_questions =
          question
          |> String.split(~r/\s+(and|vs|versus|compared to)\s+/i)
          |> Enum.map(&String.trim/1)
          |> Enum.reject(&(&1 == ""))

        {:ok, sub_questions}
      end
    end

## Using a Custom Decomposer

    Agent.new(question, repo: repo, llm: llm)
    |> Agent.decompose(decomposer: MyApp.KeywordDecomposer)
    |> Agent.search()

## Using an Inline Function

    Agent.decompose(ctx,
      decomposer: fn question, _opts ->
        {:ok, [question]}  # No decomposition
      end
    )

# `decompose`

```elixir
@callback decompose(
  question :: String.t(),
  opts :: keyword()
) :: {:ok, [String.t()]} | {:error, term()}
```

Decomposes a complex question into simpler sub-questions.

## Parameters

- `question` - The complex question to decompose
- `opts` - Options passed to `Agent.decompose/2`, including:
  - `:llm` - The LLM function (for LLM-based decomposers)
  - `:prompt` - Custom prompt function (for LLM-based decomposers)
  - Any other options passed to `Agent.decompose/2`

## Returns

- `{:ok, sub_questions}` - List of simpler questions
- `{:error, reason}` - On failure, the original question is used as a single-item list

---

*Consult [api-reference.md](api-reference.md) for complete listing*