Manus Ships Branch to Let AI Agents Fork Sessions Without Losing Context

Manus just shipped a feature that addresses one of the most frustrating friction points in long-running AI agent sessions: the moment you realize you need to go in two directions at once. The new Branch feature lets you fork any conversation at any point, spinning up a parallel session that inherits your full context , files, instructions, and history , while leaving the original completely untouched.

The problem with linear AI sessions

Most AI tools force work into a single forward-moving thread. That works fine for short tasks, but it breaks down fast when a session grows complex. A single research report can become API documentation, a user manual, a flowchart, and a slide deck , an initial idea radiates outward into many parallel tasks of different types. If you try to produce all of those in the same session, the conversation grows long and the model's focus drifts. The only alternative until now was starting over, which means losing all the accumulated understanding the agent built up.

Branch treats that understanding as a reusable asset. Click the branch icon beneath any message and Manus opens a parallel session that inherits everything up to that point , your instructions, your files, and the full history. The original stays exactly as it was. The key insight: you can spend the same context many times without depleting it.

How it actually works

When you branch, the new session inherits your files, your instructions, and the full conversation history up to that point. You don't need to re-upload or re-explain anything. Your original session stays untouched. A "Branched from" breadcrumb ties every new session back to its origin, so navigation stays manageable as your workspace grows. You can also branch from a branch, enabling deep decision trees without losing your place.

The steps to create one are minimal:

1. Hover over any message in your conversation
2. Click the branch icon that appears beneath it
3. Wait for the "Session branched" confirmation
4. Continue in the new branch with all prior context already loaded

Where this actually saves you time

The feature shines in a specific class of workflows: those where a single body of research or context needs to produce multiple distinct deliverables. Manus describes four concrete patterns:

• Meeting notes to multiple outputs: Branch from the point where meeting notes were finalized. Each branch focuses on one output , a to-do list, a manager update, a slide outline , while the notes stay intact in every direction.
• Phased research deliverables: Branch from the current point of a market research project and each branch inherits everything gathered so far. You produce phased outputs while the main thread keeps moving forward.
• Content repurposing: Branch a finished blog post to create a video script in a new session, keeping the script grounded in the original piece's narrative and tone without tangling the two.
• Living project log: One ongoing session acts like a living project log, accumulating notes and context over time. Whenever you need a specific deliverable, you branch from the current point to produce it , a plan, a summary, a presentation , and the main thread continues recording and growing.

The broader context: Manus as an agent platform

Manus is an autonomous, general-purpose AI agent. You give it an objective in plain language, and it independently plans the steps, carries them out in a cloud environment, and returns a finished result. It uses a multi-agent architecture , Planner, Executor, and Verifier , that chains together browser actions, code execution, spreadsheet generation, slide decks, and PDF research over multi-hour sessions from a single prompt. Branch is a natural extension of that architecture: rather than just running tasks in parallel, you can now fork the context itself in parallel.

This matters because context drift is a real problem at scale. Complex, multi-stage workflows with lots of decision forks can fall apart. The more branching logic a task requires, the higher the chance the agent loses track of where it is, repeats steps, or delivers incomplete output. Branch sidesteps this by keeping each direction in its own clean session rather than forcing the agent to juggle multiple goals in one thread.

Availability and limits

Session Branching is available to all users in standard Manus chat sessions. It is not currently supported in Web Builder sessions , branching applies to regular conversations and tasks, and support for additional session types may follow.

On pricing: there is a free plan with 300 daily refresh credits and access to Manus 1.6 Lite, enough to test simple tasks. Paid Pro plans start at $20 per month, and a Team plan starts at $20 per seat per month. Branch itself does not appear to carry an additional cost , it's a session management feature, not a compute-intensive one. That said, a single complex agent task can burn between 500 and 900 credits. On the $20 Standard plan with 4,000 monthly credits, you're looking at roughly five to eight serious tasks per month before you run out , so branching into multiple parallel directions will consume credits proportionally across each branch.

The real shift here is conceptual. AI sessions have always been treated as disposable , you start one, finish it, move on. Branch reframes a session as a reusable foundation: something you build up carefully and then fork as many times as you need. Whether that becomes a core workflow pattern depends on how reliably Manus can maintain context fidelity across branches at scale, but as a starting point, it's a meaningful step toward AI that mirrors how complex work actually unfolds.