Manus (@ManusAI) launched Branch on July 9th, adding a way to split one conversation into parallel sessions that inherit the same context while leaving the original thread unchanged.
The feature is built for a common agent workflow: one long research or planning session becomes the source material for several outputs. In Manus's example, a single research session can turn into a report, a slide deck and an investor brief without forcing the user to restart the context or pile all the follow-up work into the same thread. Manus described the release in a two-post thread on X and a company blog post published Thursday.
Branch is a small interface change with larger product consequences for Manus, the general-purpose AI agent associated with co-founders Xiao Hong, Ji Yichao and Zhang Tao. Ji, who has been described as the technical core of Manus, built his reputation before Manus through product-heavy consumer software work, including Peak Labs and the mobile browser Mammoth, according to Dealroom. That history matters because Branch is less about a new model claim and more about the operating surface around agentic work: how a user keeps an AI system useful after it has accumulated files, instructions and task history.
In the Branch announcement, Manus says users can hover over a previous message, click a branch icon, and create a new session that carries forward the conversation history, uploaded files and instructions up to that point. The original session remains unchanged. The new session gets its own workspace, and a "Branched from" breadcrumb ties it back to the source conversation.
That makes the product closer to a project tree than a single chat log. A user could keep one ongoing session as the main record for a market research project, then branch midway through to draft a client brief, a slide outline or an internal analysis. Manus also says users can create more than one branch from the same message and can branch from an existing branch.
The limitation is also clear: Manus says Session Branching is available to all users in standard Manus chat sessions, but it is not currently supported in Web Builder sessions. That caveat matters because Web Builder is one of the more concrete ways Manus turns conversation into output. Branch, at launch, applies to regular conversations and tasks rather than every Manus surface.
Manus's release lands against a messy corporate backdrop. The company's own site still carries the banner "Manus is now part of Meta" and a 2026 Meta copyright notice on its homepage and blog. But The Associated Press reported that China's National Development and Reform Commission blocked Meta's acquisition of Manus on April 27th and required the parties to withdraw from the deal. AP reported that Meta had announced in December that it was acquiring Manus, and that the deal was expected to expand Meta's AI offerings across its platforms.
The tension between Manus's current product cadence and the reported unwinding of the Meta deal is the useful context for Branch. Tom's Hardware, citing Bloomberg, reported on June 11th that Meta had begun separating operations from Manus after Beijing's order and that Manus employees had been locked out of Meta's internal data systems. Manus has continued to ship under its own brand while the public-facing site presents the product as part of Meta.
For users, Branch addresses the part of AI agent work that demos usually skip: what happens after the first useful answer. Long-running agent sessions can become crowded with alternate instructions, partially finished deliverables and context that matters for one output but harms another. Manus is turning that pain into a product primitive. Instead of asking users to duplicate a conversation manually or explain the same background again, Branch treats accumulated context as reusable material. The feature also points to where competition in AI agents is moving. Manus's about page says it builds "general AI agents" as an "Action Engine" for life, with product surfaces for research, slides, websites, images, videos, charts, browser operation, Slack and API access. Branch does not prove Manus can execute every task better than rival assistants. It does show Manus investing in the workflow layer around agent output, where retention may depend on whether users can manage work across many directions without losing the thread.
That is the practical bet behind the launch. If AI agents are going to handle more than one-off prompts, users need ways to preserve project memory, separate experiments and turn one body of work into several deliverables. Branch gives Manus a cleaner answer to that problem than another longer chat window.