# List a generation's events

> The generation's transcript, incrementally: its progress rows in seq order, after after_seq.

The generation's transcript, incrementally: its progress rows in `seq` order, after `after_seq`. Each `generations.step_ended` row carries the step it ended as a `TranscriptStep`; the other rows say when a step or a tool call started and ended, and which generation a tool call started.

Step content and tool names are the trace's content: reading them needs `traces:GetTrace` on the generation's trace beside `generations:GetGeneration`, and is refused with `403` without it. `progress_only=true` asks for the rows alone (`step` is null), which `generations:GetGeneration` alone allows. Content is re-read on every page: once a purge erases it, or under zero-retention, `step` is null.

`stream=true` answers `text/event-stream` instead and follows the generation until it settles (`completed`, `failed` or `requires_action`), or stops running without a settle row. Each row is one `event: progress` frame whose `id` is its `seq`; the stream ends with `data: [DONE]`. A `: heartbeat` comment is sent every 15 seconds. After 5 minutes the stream ends with `event: end` and `data: {"reason": "max_lifetime", "generation_id": "...", "after_seq": N}`: reconnect with that `after_seq`, or let an `EventSource` reconnect with its `Last-Event-ID` header, which this route reads as `after_seq`. Every reconnect is authorized afresh.

  Request

<ParamsDetails
  {...require("./list-generation-events.ParamsDetails.json")}
>

<RequestSchema
  {...require("./list-generation-events.RequestSchema.json")}
>

<StatusCodes
  {...require("./list-generation-events.StatusCodes.json")}
>
