Skip to content

Creating a task

You can create a task and work agentically from the browser in Cloud.

Cosine Cloud create-task form showing the main configuration fields before starting a task

Here you can create a new task or open a previous task.

You can get started by adding a task to the input field straight away. You can also adjust the task settings before you launch it.

Create-task mode selector showing the available agent modes in Cosine Cloud

When creating a task you can choose between:

  • Manual for a straightforward single-agent workflow
  • Plan when you want Cosine to research the task, inspect the codebase, and propose an approach without making changes yet
  • Swarm when you want multiple agents to split up the work
  • Auto when you want Cosine to switch between approaches automatically as the task develops

These modes are essential because they shape how much autonomy the agent has and how it breaks down the work.

See Models and Pricing for a dedicated overview of the available modes.

Choose the GitHub repository Cosine should work in.

Repository picker in the create-task flow for selecting the GitHub repository the task will run against

Every task runs against a specific imported repository. Pick the repository that contains the code, docs, or configuration you want the agent to work on.

If you have several repositories connected to the same team, make sure you select the one that actually owns the change. This keeps the agent grounded in the right codebase from the start.

Choose the model that fits the complexity and cost profile of the task.

Model picker in the create-task flow for choosing the model to run the task with

For smaller or well-scoped tasks, a lighter model may be enough. For more ambiguous, research-heavy, or multi-step work, a stronger model can be a better fit.

See Models and Pricing for the current model catalog and pricing multipliers.

Choose the environment where the task should run.

Environment selector in the create-task flow for choosing which runtime environment to use

The environment determines the workspace the agent will use while carrying out the task. It controls what tools, languages, and dependencies are available.

Base environments like Ubuntu, Browser, Desktop, VS Code, and Slim work immediately without setup. Custom environments created in your project settings provide project-specific tooling and dependencies.

See Environments for more detail on available environments and how to create custom ones.

Choose the base branch and the working branch for the task.

GitHub branching settings in the create-task flow showing base branch selection and branch naming for the agent

This is where you define:

  • the base branch the task should start from
  • the branch name the agent should use for its work

This is useful when you want the task to target a release branch, a staging branch, or a clearly named feature branch instead of relying on a default.