When It Exists

ctx.message represents the assistant message generated during the current turn.

This value is only available in post-turn hooks that run after the model finishes generating a response.

During prompt construction (onContextBuild), ctx.message is always null.

It becomes populated in post-turn hooks where the host provides the assistant message for the current turn, such as onTurnEnd.

Shape

interface TurnMessage {
  id: string
  role: 'assistant'
  content: string | null
  finishReason?: 'stop' | 'length' | 'tool_calls' | 'error'
  toolCalls?: ToolCall[]
}

Fields

• id – unique identifier of the assistant message
• role – always "assistant"
• content – textual content produced by the model
• finishReason – reason the model stopped generating
• toolCalls – reserved for tool-capable flows

content may be null in cases such as:

• generation failed
• the turn was aborted

Practical Use

Strategies typically use ctx.message in post-turn hooks to:

• skip post-processing after aborted or failed turns
• extract structured memory from generated text

Example:

export async function onTurnEnd(ctx) {
  if (!ctx.message) return

  if (ctx.message.finishReason !== 'stop') {
    return
  }

  const text = ctx.message.content
  if (!text) return

  // perform memory extraction or post-processing
}

In v0.1, strategies should primarily rely on content and finishReason.

Notes

ctx.message represents only the assistant message generated in the current turn.

It does not expose earlier messages.

To access conversation history, use:

ctx.history