Getting started

Start small.
See the whole process.

This guide takes you from installing Wile to reviewing your first workflow. You need an Apple Silicon Mac running macOS 15 or later and an available agent provider.

Install and connect

Check the current Mac download. Once a public release is available, open the disk image, drag Wile to Applications, and open it.

Create an account with Google or email. Verify your email when prompted, then sign in to Wile. Google accounts with a verified email do not need another email challenge.

Open Runtimes in the sidebar to see available providers and models. Configure and sign in to Codex, Claude Code, or Pi on the device that will run the work. Follow the setup guidance shown for unavailable providers.

Pi can discover custom local or cloud models from its own configuration. Provider subscriptions, API charges, and model availability depend on that provider.

Your first workflow

Choose New workflow. Describe a small task to the assistant, or configure the steps on the graph canvas. A saved workflow is the design; a run is one execution of that design.

1. Start with a manual trigger and a clear input, such as a short project brief.
2. Add an agent step with specific instructions and an available model.
3. Attach the workspace, skills, and tools it needs.
4. Connect it to a review step, then add a human approval.
5. Choose Run and inspect the stage activity and output.

A good first workflow: read a brief → draft a plan → review the plan → wait for your approval. Try local files first so you understand the handoffs before connecting more tools.

Workflow assistant

Open the Assistant beside a workflow on this Mac. Describe your process in conversation: its input, desired result, agents, tools, and where you want to review the work. The assistant can create and revise the workflow’s steps and handoffs.

Choose an available provider and model. Review the design on the canvas, refine it with follow-up requests, and start execution with Run. Creating a design does not run the workflow. The right-side assistant operates on this Mac; connected-device workflows use their configuration controls.

Review and approval

An approval step holds the run until a human decides. Inspect the available output and evidence, then use Approve & continue to release the next step, or stop the run.

Approvals bind reviewed content. If that content changes, the approval is rejected and you need a new run. Editing the saved workflow does not rewrite a run already in progress.

Agent review loops are separate: a reviewer can send work back for another attempt up to the configured limit. Human approval still belongs to you.

Tools and permissions

Open Connectors to browse available apps or add a custom MCP connector. A saved profile is configuration; it does not prove that the service is authenticated or reachable.

Enable the connector tools you need, then narrow access for each agent step in its configuration. A drafting step may need file editing without a tool that sends messages. Review both the provider’s capabilities and each connected account’s permissions.

Choose external actions deliberately. An approval protects the steps after it; it does not undo actions earlier in the workflow.

Codex apps and plugins

In Connectors → Browse, Wile discovers the app catalog available through the Codex account on each device. Connect or reconnect an app when requested, select its tools, and assign only the tools each step needs.

This compatibility covers Codex connected-app tools. Entire Codex plugin bundles, hooks, plugin skills, and arbitrary installed plugin servers are not automatically imported. Add other MCP servers explicitly and attach reusable instructions as Wile skills.

App availability and authentication belong to the device running the workflow. A catalog entry shows a discoverable app; it does not prove that a tool call will succeed.

Skills and workspaces

Skills are reusable instructions, such as a writing guide or review checklist. Create or edit them in the Skills library and attach them to the relevant steps. Wile stores their instructions in readable files and captures the resources used by each run.

Workspaces give agents a disposable copy of the files or repository they need. Connections between steps can hand off workspace content, a document, or text. External connector actions can still affect the connected service.

Schedules and triggers

Start manually while you refine a workflow. When it is ready, configure a schedule with a time zone, or use a supported polling source. The native custom-trigger setup currently supports Telegram.

Review the trigger configuration and choose Enable trigger in its inspector; after later edits, choose Apply changes. Saving a graph alone does not enable automatic execution.

Polling tracks source identifiers to avoid creating duplicate runs from the same source item. This is not a guarantee that an external action executes exactly once. Schedules require the owning device and runtime to be available; they do not wake a sleeping Mac.

Connected devices

Open Devices to enable remote access through a shared wss:// relay URL, then connect another device with a pairing code. Each workflow belongs to one device. Its providers, credentials, files, and execution live there.

Connecting a device lets you manage its work from your Mac; it does not move individual steps between machines. An offline device cannot execute a new remote action. Keep the owning device available for scheduled work.

Data and costs

Wile stores local configuration and history on the workflow’s device. Cloud model calls and connector actions may transmit prompts, files, or tool inputs to the services you select. A local model alone does not prevent external tools from sending data.

Provider billing follows the selected provider and its configuration. Check subscriptions, API usage, and limits before running a workflow repeatedly. Wile does not turn paid model usage into free usage.

Troubleshooting

• No model available: check the provider in Runtimes and complete its installation or sign-in.
• Connector unavailable: check its account, enabled tools, and connection status.
• Schedule not running: confirm the graph is enabled, the trigger was applied, and the owning device is awake.
• A failed stage: inspect its activity and error before retrying. Check whether an external action already happened.

Need help with your setup? Send feedback with the visible error and the step you were trying. Keep the returned report ID. If you cannot log in, contact us. Leave credentials and private files out of the message.