Browse documentation
Documentation

Choose your tools

Choose your agent and model

Select a supported agent and model, connect your provider, and understand what can change between runs.

Updated September 7, 2026 4 min read

Rigless lets you choose the coding agent and a supported model for a hosted project. This guide shows where to make those choices and what can change after work begins.

Free Browser projects use the built-in Rigless agent. Choose a hosted plan for Linux tools and the broader agent selection. Every plan includes limited AI usage, and supported provider keys are available on Free as well as paid plans.

Understand the three choices

Choice

What it controls

When you choose it

Workspace

Where the project runs: Browser files and previews, or hosted Linux tools

When creating a project

Agent

The program that plans the work and uses tools; sometimes called a harness

When creating a project

Model

The AI model the selected agent calls

Before a task or between runs, from the supported choices

Hosted projects support Rigless, Codex, opencode, Pi, and Oh My Pi. Their available models and controls differ. Use the combinations shown in Rigless's picker for the selected agent.

Select an agent for a new hosted project

  1. Open Developer workspace. The default Create view is intended to get you building with fewer controls.

  2. Open the agent control above the task field. Its label shows the current choice, such as Rigless or Browser workspace.

  3. Choose a hosted agent. If only Browser workspace is available, check your plan in Settings.

  4. Open the model control below the task field and select an available model. Check the agent label again before starting: choosing a model can affect which agent is selected for a new project.

  5. Describe one small task and select the upward-arrow Start task button.

The agent choice is fixed for that project. To work with a different agent, start a new hosted project using your repository or a copy of your source. The GitHub guide shows how to keep work in a repository you control.

Change the model for the next run

Wait for the active run to finish. In a project, open Advanced to reach the Developer workspace; on a phone, use the code icon beside the three-dot menu. Open the model control near the message field and choose a supported model before sending the next request.

Your source files stay in the project. Changing a model does not require you to recreate the code. The available model list depends on the project's agent and your provider connections.

Starting a project with another agent creates a separate conversation. Native agent histories and tool state are not an interchangeable export format. Give the new agent the source, the task, and any decisions it needs to continue.

Connect your preferred provider

Open Settings, find API keys, and choose + Add key. Select a supported provider, enter the key, and choose Save key. Return to the project and select that provider's model in the picker.

Adding a key does not mean every agent supports every model. Check the available choices for your agent. When your own key is used, the provider's charges and limits apply; Rigless's workspace and publishing limits still apply too.

Plans and provider keys explains included usage, provider billing, and what to do when a request reaches a limit.

Set reasoning effort when available

Some models expose Reasoning effort or Thinking controls. These affect how much reasoning the model uses for a request. The options depend on the model and agent; leave the default if you are unsure.

More reasoning can take longer and use more model capacity. Try a small task, inspect the result, and adjust only when the work benefits from it. A model without this control can still edit and explain code.

Check that your choice works

Start with a request that has a result you can inspect:

Read this project's README and source files. Explain how it runs,
identify one small improvement, and tell me how we could check it.
Do not change files yet.

Confirm the response refers to your actual project. If the run fails before producing a useful result, read its error before retrying.

Problem

Next step

A preferred agent is missing

Open Developer workspace and confirm the account has hosted compute

A provider model is missing

Check the saved key and the models available for the selected agent

A key is rejected

Check the key's provider, permissions, billing, and expiration in that provider's account

Included usage is exhausted

Wait for the indicated reset or select a supported provider model using your own key

A reasoning control is absent

Use the model's defaults; this combination may not expose that control

For the other part of keeping your options open, export and run your code outside Rigless.