Yes, you can! We take all the "general" sections (currently: openai, features and edge_tts) from the top of the config, and then copy the wingman config section over, so that the wingman always takes precedence.
So if you want to override something, just mirror the general structure in your wingman config:
features > tts_provider: default is overridden by my-wingman > features > tts_provider: edge_tts.
Because of this mechanism, you can also add completely new config settings and sections to your custom Wingman. Just add something new to your wingman's config, override Wingman.validate() to validate it if necessary, and then you can use it safely.
Everything you can change (as non-dev) is stored in the config.yaml file.
Be careful: It is very indentation-sensitive!
Please DO NOT use the SPACE bar to indent lines in the config. Use the TAB key instead. Always use a single TAB to indent. If you're using an editor that automatically converts TAB to SPACE, please disable this feature for this file.
The file is very hierarchical, meaning that entries belong to other entries. The file begins with a general section. These are global (default) settings for all your wingmen.
The screenshot shows the general section of the config file. It has several sections or entries (colored), and these are all siblings to each other, meaning they have the same indentation level and must align perfectly with each other.
After the general section come the wingmen. Each wingman is defined in its own entry. They are all children of the wingmen entry and siblings to each other. Then each wingman has its own commands and other settings.
You can override any setting from the general config in a specific wingman. For example, the board-computer wingman overrides the global features > enable_robot_sound_effect: false by mirroring the exact same structure with wingmen > board-computer > features > enable_robot_sound_effect: true.
Note also the commands. The general section defines a global ResetConversationHistorycommand, meaning that every wingman has this command. Then theboard-computerwingman adds its own keystroke command to the command list. Also note that commands start with a-sign. This is because they are lists of commands. You can add as many commands as you like. Please do not remove the- sign and add it to your new commands.
The context prompt describes the character of your Wingman and controls how it behaves. It's a simple ChatGPT prompt that you can edit to your liking. You can do a lot with prompting, so play around with that.
You can use the OpenAI ChatGPT web interface or our no login required open source tool called SlickGPT to test your prompts.
Prompting is a science unto itself. If you want to get better at it, there are great resources like learnprompting.org.
Commands are the heart of the Wingman AI, as they add functions to otherwise pure conversations. Commands are defined in config.yaml and are activated by the corresponding Wingman as a result of a conversation with you. In the first version, they are mainly used to trigger keystrokes, but they can also be used for other actions such as calling external APIs or executing complex scripts.
Here are the most important properties of a command:
name: This is used to associate your request to the command. So make sure to give it a good name, e.g.RequestLandingPermission. UsecamelCasingto separate words. No spaces and don't write everything in lowercase. You do not have to say a specific phrase to trigger the command because the AI is smart enough to detect which command you meant. This is a very powerful concept and one of the reason this is way better than other (sometimes "aggressive" 🤭)voice-to-keytools.keys: A list of keys to press. They will be triggered in the order they are defined. Any key can have these properties:key: The key to press.modifier: A modifier key to press with the key. (optional)hold: The time to hold the key in milliseconds. (optional)wait: The time to wait until the next key in this command is pressed (optional)
instant_activation: A list of phrases that will trigger the command immediatale without AI round-tripping. (optional)responses: A list of responses. If the command is executed, a random response will be chosen and read out to you. (optional)
Your new command has to line up with the ones before and starts with a - name: Whatever line. Align the - with the one from the command before. Indent everything under your command with a single TAB character as separator.
Do not cut the instant_activation from the previous command (the one in the screenshot belongs to the ScanArea command.
See the full list here
Yaml is hard, we know. We're working on a more user-friendly interface to make it easier for you. For now, we recommend the following tools to help you out:
- VSCode with the
- YAML extension
Both are free and very easy to install and use. VSCode is a great editor for developers, and the YAML extension will help you with indentation by showing you errors and warnings in your config file as you type.
Notice how it has detected that something is wrong with the indentation. In this case, there is a single space before the features entry, which is not allowed. It also shows you visually how it should be indented.
It will get easier once you get used to it, we promise!
Remember: Never use SPACE, always use TAB!
Wingman supports all languages that OpenAI (or your configured AI provider) supports. Setting this up in Wingman is really easy:
Find the context setting in config/config.yaml for the wingman you want to change.
Now add a simple sentence to the context prompt:
Always answer in the language I'm using to talk to you.
or something like:
Always answer in Portuguese.
The cool thing is that you can now trigger commands in the language of your choice without changing/translating the name of the commands - the AI will do that for you.
Here's how to set the default sound output for a particular program on Windows 10 (and probably 11):
- Right-click the speaker icon in the system tray:
- There should be a speaker icon in the lower-right corner of your screen. Right-click it. Open the sound settings.
- Select "Open sound settings" from the context menu. This will open the sound settings window. Scroll down to "Advanced sound options"
- In the sound settings window, scroll down to the "Advanced sound options" section. Select the output device for each application
- Under the App volume and device settings section, you should see a list of open applications and their corresponding output devices. Use the drop-down menus to select the output device you want for each application. Note that not all applications might appear in this list, depending on whether the application supports individual sound output settings.
Thanks to our community member EagleOne for posting this guide on our Discord server.
Please contact us directly in Discord or via Patreon. Please do not open an issue on GitHub.
Are you using Cloudflare DNS?
Try nslookup https://api-test.star-head.de in your terminal. If Adress is 1.1.1.1, then you are.
If so, for now you can either change your DNS to something else (8.8.8.8 is Google) or have to disable :wingman_ai: StarHead, sorry. We'll report this to the StarHead backend team.
Please check the exact transcript text. Where there words misunderstood? Especially the API params ship, location or money?
We're aware of this issue and looking into it.
If you install EdgeTTS on your system, you can output all the voices in your terminal using edge-tts --list-voices.
Here's a dump of this list (taken 11/24/2023):
Non-techie version:. It should be like using ChatGPT: You hit enter and it starts typing immediately. Only this time with speech. But that's not happening right now because the OpenAI API is broken. Instead, we have to wait for all the audio to be generated and then start playing it.
Dev version: There is currently a known issue with Open AI's TTS API that our community member meenie discovered and reported to them. Basically, their audio streaming doesn't work in Python. This results in much higher latency when starting the TTS conversion. We're waiting for a fix.
Until then, you can always use another TTS provider like EdgeTTS or 11Labs.
Make sure you DID NOT extract Wingman to C:\Program Files or another directory that needs admin priviledges. Pick a user directory like your Desktop, Documents, Downloads etc instead.
Extract all the files from the .zip file, not just the .exe.
If it still doesn't work, please join our Discord and provide a screenshot showing the error message.
Please do not create a new GitHub issue for that.