Get started
AI providers and keys
Choose an AI provider, add your API key, and understand how the key is checked and stored.
Supported providers
ActionWright uses an AI model from a provider you choose, with your own account and API key. At the time of writing, the Provider list in Provider Setup has five entries:
| Provider | Speech built in | Where to get a key |
|---|---|---|
| OpenAI | Yes | OpenAI API keys |
| xAI / Grok | Yes | xAI API keys |
| Google / Gemini | Yes | Google AI Studio API keys |
| Anthropic | No. See Task provider and voice provider. | Anthropic API keys |
| OpenRouter | Yes | OpenRouter API keys |
OpenRouter is a gateway: one key reaches models from many companies, and you can pick the model from a list inside ActionWright. See Models.
In a new install, OpenAI is selected. ActionWright needs a saved key before it can run any task.
Task provider and voice provider
The Task Provider runs your tasks: the orchestrator and every agent use its model.
Speech uses a voice provider. It turns your speech into text for Push-to-Talk and Voice mode, and it speaks replies. With OpenAI, xAI / Grok, Google / Gemini and OpenRouter, the task provider and its key also handle speech, so there is nothing more to set up.
Anthropic has no built-in voice model. When you choose Anthropic as the task provider, a Voice Provider section appears with this note: "This provider has no built-in voice model." Choose one of the four providers with speech and enter its key there. If you leave that API Key blank, ActionWright saves your setup with speech off: tasks work, but Push-to-Talk dictation, Voice mode and spoken replies do not.
Add or change your key
- Open Action Wright — Provider Setup. Right-click the tray icon and choose Settings, or in the task window click Menu and then Settings. On the first launch it opens after the tutorial.
- Under Task Provider, choose your provider in Provider.
- Paste your key into API Key. The field hides what you type.
- Leave Task Model as it is, or enter another model name. See Models.
- If you chose Anthropic, fill in the Voice Provider section, or leave its key blank to keep speech off.
- Click Save. The button shows Validating… while ActionWright tests your key.
If the test passes, the dialog closes and your setup is saved. If it fails, an error appears in the dialog and nothing is saved. Fix the problem and click Save again. See What the check tests.
ActionWright keeps one setup at a time. Saving a new provider replaces the previous provider and its key. Keys you type for other providers while the dialog is open are kept only until you close it.
What the check tests
When you click Save, ActionWright sends one short test request to your provider with the key and task model you entered. The test shows whether:
- the provider accepts the key;
- your account can pay for requests, for example that it has credit or quota left;
- the model name exists and your key may use it;
- the model accepts tool calls, which every agent needs.
With a separate voice provider, ActionWright also tests that provider's key with a short text request.
The test does not try the speech models or image input. A wrong STT Model or TTS Model name shows up only the first time you use speech. A model that cannot read images fails only when a task sends screenshots or images.
The test is a real request, so it may appear as a very small amount of usage at your provider.
Errors you may see
The message names your provider and often adds what the provider said. If it starts with "Voice provider:", the problem is with the voice provider's key, not the task provider's.
| Message starts with | What it means | What to do |
|---|---|---|
| "API key is required." | The API Key field is empty. | Paste your key. |
| "The API key was rejected by …", "This API key has expired …", or that the key was reported as leaked | The key is wrong, expired or revoked. | Copy the key again, or create a new one on your provider's key page. |
| "Not enough credit at …", "A spending limit … was reached", or "Your … plan has no quota …" | Your provider account cannot pay for the request. | Add credit, raise the limit, enable billing, or choose a cheaper model. |
| That access was denied, that the key lacks a permission, or that your organization must be verified | The key or account is not allowed to use this model. | Check the key's permissions in your provider's dashboard, or choose another model. |
| That the provider has blocked this account or project | Your provider has blocked the account. | Check your provider's dashboard. |
| That the provider is not available from your current location | Your provider does not serve your region. | If you use a VPN, try turning it off. |
| "Model '…' was not found or is not available at …" | The model name is wrong, or your key cannot use it. | Check Task Model. Clear the field to use the provider's default. |
| "Model '…' does not support tool calls, which every agent needs." | This model cannot run agents. | Choose a model with tool support. |
| That the provider is limiting how many requests can be sent | You hit a rate or usage limit. | Wait a minute and try again. If it keeps happening, check your plan's limits. |
| That the provider is overloaded or having server trouble | A temporary problem at the provider. | Try again in a few minutes. |
| "Could not reach …", or that a proxy refused the connection | No internet, or a network or proxy problem. | Check your connection and proxy settings. |
| "Model '…' did not answer in time." | The provider took too long. | Try again, or choose a faster model. |
| "The test call took more than 30 seconds." | The provider did not answer in time. It may be busy. | Try again. |
| "Unexpected", or "Provider returned an error (…). Check your key and model name." | Any other error. When the message starts with "Unexpected", it ends with the provider's own error text. | Read the provider's message. Check the key and model name. |
| "Could not reach the Action Wright server" | ActionWright's own background service did not answer. | Wait a moment and click Save again. If it keeps happening, choose Quit in the tray menu and open ActionWright again. |
Where your key is stored
ActionWright stores your API keys in Windows Credential Manager, not in its settings. Windows encrypts them for your Windows account. Your provider choice and model names are saved in ActionWright's settings; the keys are not.
You can see that ActionWright's entries exist in Control Panel, then Credential Manager, then Windows Credentials.
This keeps the key away from other people who sign in to the PC with their own Windows account. It does not protect the key from harmful software running under your own account. Keep your PC free of software you do not trust.
Delete Saved Keys
Delete Saved Keys in Provider Setup removes the task key and the voice key from Windows Credential Manager at once. It does not ask for confirmation, and Cancel does not undo it. The dialog shows "API keys deleted from secure storage."
Your provider choice and model names stay. Until you save a key again, tasks stop with "No API key configured for …".
Deleting a key in ActionWright does not cancel it at your provider. To revoke a key, delete it on your provider's key page.
Models
Task Model is the model that runs your tasks. Each provider fills in a default. If you clear the field and click Save, ActionWright uses that provider's default.
| Provider | Default task model (example, at the time of writing) |
|---|---|
| OpenAI | gpt-6.1-terra |
| xAI / Grok | grok-4.7 |
| Google / Gemini | gemini-3.8-flash |
| Anthropic | claude-sonnet-5-5 |
| OpenRouter | openai/gpt-6.1-terra |
Providers rename and retire models often, so these defaults change between versions. If you enter your own model, take the name from your provider's documentation. It must support tool calls. For the browser agent and Computer Use, it must also accept images, because those agents send screenshots.
With OpenRouter, a Browse… button appears next to each model field. Enter your OpenRouter key first, then click Browse… to load OpenRouter's model list. Search it and double-click a model to fill in the field. You can also type a model name yourself.
Speech models
The Voice field shows Standard. To change how speech works, click Voice Advanced Settings. It has STT Model (speech to text), TTS Model (text to speech) and TTS Voice.
If you leave a speech field at its default, ActionWright follows the default of the version you run, including after updates. A value you change is kept exactly as you typed it, even if the provider later retires that model.
Costs
Your provider bills you for every request ActionWright makes with your key, at your provider's prices. ActionWright does not include AI usage and cannot see your provider bill.
- A task makes many requests, usually at least one for each step.
- Long tasks, large documents and screenshots make requests bigger.
- Push-to-Talk, Voice mode and spoken replies make speech requests.
- The check when you click Save makes one small request. See What the check tests.
After a task, the task window can show the tokens it used, for example "Tokens: 12,345 total (12,000 in / 345 out)". Your provider's dashboard is the place to see what you were charged.
If your credit or a spending limit runs out during a task, ActionWright pauses the task and shows what failed. Fix it, for example by adding credit or changing the model or key in Settings, then click Continue. Stop ends the task and keeps what was done.
