Skip to content
The Docs Report
Tuesday, November 17, 2026NewsletterRSS

GitBook Agent Now Splits Big Docs Jobs Across Parallel Agents

GitBook's Sept. 30 update lets its Agent run background agents in parallel, serves Markdown to ChatGPT Search's crawler and paginates llms-full.txt.

2 min read

GitBookAIllms.txt

Isometric 3D illustration of glossy cobalt blue toy robots on a tiled floor

GitBook Agent can now split a large task across several background agents that run in parallel, GitBook said in its Sept. 30 changelog. The Agent merges their findings into one conversation. GitBook says broad questions and multi-page edits should finish faster.

Parallel agents in one conversation

Each background agent gets its own conversation, which users can open with one click. Tool calls, thoughts and background agents now sit on one grid in the chat view. The Agent also reads pages in parallel. GitBook gave no speed figures, limits on how many agents run at once or pricing details.

Published GitBook sites now serve Markdown to ChatGPT Search’s crawler, so the search engine can index docs in what GitBook calls “a more LLM-friendly language.” GitBook didn’t say whether the Markdown is a separate rendering or the same page .md twin it already offers, or whether the change is on by default.

A paginated llms-full.txt that points onward

Each part of a paginated llms-full.txt now says at the top which part it is and links to the previous and next parts. GitBook’s stated reason is that AI tools that would otherwise extract a single page know to fetch the rest.

Channels, audit logs and other changes

GitBook’s channel setup for Slack, GitHub and Linear now lets admins pick channels, repositories or teams from a list instead of typing names or IDs. Enterprise audit logs now record changes to agent instructions, custom fonts, OpenAPI specifications and translations, and mark actions GitBook Support takes on a customer’s behalf. AI Assistant answers can be exported as CSV, and an API setting for an organization’s support cost feeds the AI Assistant report’s savings estimate. GitBook also fixed a bug where its Agent didn’t reply to Linear messages when the connection was limited to specific teams.

Key takeaways

  • GitBook Agent can fan a large task out to parallel background agents and merge the results.
  • Published sites serve Markdown to ChatGPT Search’s crawler.
  • Paginated llms-full.txt parts link to their neighbors.
  • Enterprise audit logs now cover agent instructions and Support actions.

The take

We think the paginated llms-full.txt change is the most useful item here, because it admits a failure: AI tools that fetch one page of a multipart file stop there and never see the rest. Linking the parts fixes that, but only for clients that follow the links. On parallel agents, the changelog promises speed and gives nothing we can check. We’d want GitBook to publish how many agents can run at once and what parallel runs cost. The ChatGPT Search change fits a pattern across docs platforms of tuning output for named AI crawlers. That kind of claim is easy to make and hard to test, so we’d want logs showing those crawlers actually use it.