Feature Request: Flexible Post-Processing Profiles with Hotkeys and Voice Commands

[BinaryRider]

Edit (November 8, 2026): The original post was unintentionally missing substantial sections due to a formatting/copy-and-paste issue. This version restores the complete proposal, including the voice-command workflow, fallback rules, and clipboard/selected-text context. Apologies for the confusion!

First of all, thank you for developing Handy! I've been using it on Windows 11 and really appreciate its lightweight, local-first approach and the experimental LLM post-processing functionality.

Having previously used SuperWhisper, I've been thinking about a few improvements that could make Handy's post-processing workflow more convenient, flexible, and powerful.

Some related suggestions have already appeared in other discussions. I'd like to bring these ideas together into a more comprehensive concept while keeping Handy's existing functionality as simple and familiar as possible.

The main idea is to preserve the current post-processing system as a "Default Profile" and introduce a separate, optional system of independently configurable "Custom Profiles".

Additionally, I'd like to suggest a few improvements to prompt context handling.

---

1. Preserve the existing post-processing functionality as the Default Profile

Handy currently provides two basic keyboard shortcuts:

• Shortcut 1: Regular speech-to-text transcription without LLM post-processing.
• Shortcut 2: Speech-to-text transcription followed by LLM post-processing.

These two fundamental shortcuts should remain unchanged.

The existing post-processing configuration would become the Default Profile.

The name "Default Profile" would be fixed and could not be renamed, making its special role immediately recognizable.

Importantly, the Default Profile would retain Handy's current behavior:

• One configured LLM provider, either local or cloud-based.
• The selected model and its configuration settings.
• Multiple saved post-processing prompts.
• One currently selected prompt.
• Shortcut 2 to start recording and subsequently execute the selected prompt.

In other words, existing users could continue working exactly as they do today.

Improvement: Switch the Default Profile's prompt through the system tray

Currently, switching between saved prompts requires opening Handy's settings and selecting a different prompt.

It would be much more convenient if users could simply right-click the Handy system tray icon and choose a prompt from a submenu.

For example:

Handy
  |
  +-- Post-processing prompt >
  |     +-- General correction  [selected]
  |     +-- Email formatting
  |     +-- Translation
  |     +-- Custom instruction
  |
  +-- Settings
  +-- ...

Selecting a prompt would only change the active prompt within the Default Profile. It would not start a recording.

The user would then press Shortcut 2 as usual to record and process speech using the newly selected prompt.

This preserves the existing workflow while eliminating the need to open the settings whenever a different prompt is required.

A related suggestion already exists in discussion #2064.

---

2. Introduce independent Custom Post-Processing Profiles

In addition to the Default Profile, Handy could offer an optional system of Custom Profiles.

Unlike the Default Profile, which retains the existing shared LLM configuration and multiple selectable prompts, each Custom Profile would have its own complete post-processing configuration.

Each Custom Profile would contain:

• A freely configurable profile name.
• Its own LLM provider (local or cloud-based).
• Its own model and relevant configuration settings.
• Its own post-processing prompt.
• A required keyboard shortcut.
• An optional spoken command word.

For example, users could create separate profiles for:

• Translating dictated text into Spanish.
• Formatting dictated text as a professional email.
• Summarizing complex information.
• Correcting spelling and grammar.
• Processing clipboard contents using spoken instructions.

Each profile could use the LLM best suited to its task.

A lightweight local model might be sufficient for basic corrections, while a more capable local or cloud-based model could be used for complex editing tasks.

Custom Profiles would operate independently of the Default Profile.

Executing a Custom Profile must not change the Default Profile's LLM configuration or currently selected prompt.

Similarly, changing the Default Profile's prompt through the system tray must not affect any Custom Profile.

This separation would keep the system predictable and avoid unexpected changes between workflows.

---

3. Selecting Custom Profiles using keyboard shortcuts and optional voice commands

This is probably the most interesting part of the proposal.

Every Custom Profile would require a keyboard shortcut.

That shortcut must be different from Shortcut 1 (regular transcription) and Shortcut 2 (Default Profile).

However, multiple Custom Profiles could share the same shortcut if they use different spoken command words.

Example configuration

Profile	Shortcut	Spoken command
Default Profile	Ctrl+Alt+D	Not applicable
Spanish translation	Ctrl+Alt+H	SPANISH
Email formatting	Ctrl+Alt+H	EMAIL
General text correction	Ctrl+Alt+H	None
French translation	Ctrl+Alt+F	FRENCH
Summarization	Ctrl+Alt+S	None

Shortcut 1 for regular speech-to-text is configured separately and is not shown in this table.

These shortcuts and command words are only illustrative examples.

How spoken command recognition would work

From the user's perspective, the workflow would be straightforward:

The user starts recording with the keyboard shortcut assigned to one or more Custom Profiles, speaks the desired command word followed by the text to be processed, and finishes the recording as usual.

Handy would first transcribe the complete spoken input. Only after transcription would it determine which Custom Profile should process the text.

For example, suppose the user presses Ctrl+Alt+H and says:

"SPANISH I am looking forward to our meeting."

Handy would perform the following steps:

1. Record the speech and generate the transcription.
2. Examine the beginning of the transcribed text for a recognized command word associated with Ctrl+Alt+H.
3. Find the matching Custom Profile.
4. Remove the command word from the transcription.
5. Send the remaining text to the LLM configured in that profile, using its associated prompt.
6. Return the post-processed result.

In this example, Handy would identify the command SPANISH and select the Spanish translation profile.

The text passed to the LLM would be:

I am looking forward to our meeting.

The resulting translation might be:

Espero con ilusión nuestro encuentro.

The important detail is that the command recognition would happen inside Handy, before any LLM request is made.

There would be no need to ask an LLM to interpret the command or select a processing mode.

This approach could reduce unnecessary processing, simplify prompts, and improve responsiveness, especially when working with local LLMs.

Command matching should ideally be case-insensitive, recognize complete command words rather than arbitrary prefixes, and tolerate normal transcription punctuation.

Profiles without spoken commands

A Custom Profile could also be assigned a shortcut without any spoken command.

For example, pressing Ctrl+Alt+S would directly execute the Summarization profile after recording.

No spoken prefix would be required.

Fallback behavior

If multiple profiles share a shortcut, I would suggest the following selection logic:

Case A: A matching spoken command exists

Execute the corresponding Custom Profile and remove the recognized command from the transcription.

Case B: No matching spoken command exists, but a profile without a command is assigned to the same shortcut

Execute that commandless profile using the complete, unmodified transcription.

Case C: No matching command exists and no commandless profile is assigned to that shortcut

Output the original transcription without LLM post-processing, just as regular speech-to-text would do.

This avoids silently discarding dictated text and does not require any mandatory global fallback profile.

The fallback profile is therefore simply the commandless Custom Profile assigned to the same shortcut.

Configuration rules

To prevent ambiguity, Handy could enforce a few simple rules:

• Shortcut 1 and Shortcut 2 remain reserved for their existing functions.
• Custom Profiles must use other shortcuts.
• Multiple Custom Profiles may share a shortcut.
• For a given shortcut, each spoken command must be unique.
• Only one commandless Custom Profile may exist per shortcut.
• Calling any Custom Profile must not modify the Default Profile.

These rules would allow flexible configurations while keeping the selection behavior deterministic.

---

4. Extend prompt variables with clipboard and selected-text context

Handy already supports the ${output} placeholder, which inserts the recognized speech into a post-processing prompt.

It would be extremely useful to introduce at least one additional placeholder, preferably two:

• ${clipboard} – The current clipboard contents.
• ${selection} – Text currently selected in the active application, where supported.

These placeholders should work in both the Default Profile's prompts and Custom Profiles.

A. Clipboard context: ${clipboard}

Imagine that a user has copied a paragraph to the clipboard and wants to translate, summarize, or rewrite it.

Instead of creating separate fixed prompts for every possible operation, the user could configure a general-purpose prompt:

Instruction:
${output}

Text to process:
${clipboard}

Follow the instruction and return only the result.

Then the user could simply say:

"Translate this text into Spanish."

Or:

"Summarize this in three sentences."

Or:

"Rewrite this in a more professional tone."

The recognized speech would become the instruction, while the clipboard contents would provide the text to process.

This would effectively turn Handy into a voice-controlled text-processing assistant without requiring users to dictate the entire text again.

B. Selected-text context: ${selection}

An even more convenient option would be to make currently selected text available to the prompt.

For example:

1. Select a paragraph in a text editor or another application.
2. Start Handy using the appropriate post-processing shortcut.
3. Say: "Rewrite this paragraph in a more formal style."
4. Handy passes the spoken instruction and selected text to the configured LLM.

The user would not need to copy the selected text manually.

Ideally, this could work without modifying the clipboard.

I understand that reliably accessing selected text may be platform-dependent and could require different implementations on Windows, macOS, and Linux.

Therefore, I would consider ${clipboard} the simpler initial enhancement, with ${selection} as a possible subsequent improvement.

Privacy considerations

Clipboard contents and selected text may contain sensitive information.

I would therefore suggest that these values should only be read and included when the executed prompt explicitly references the corresponding placeholder.

They should not automatically become part of every LLM request.

This would be particularly important when using cloud-based LLM providers.

---

5. How the three workflows would coexist

The intention is to keep the overall design understandable and avoid interfering with established functionality.

Workflow	Trigger	Processing
Regular speech-to-text	Shortcut 1	Transcription only
Default Profile	Shortcut 2	Current Default Profile configuration and selected prompt
Custom Profile	Its assigned shortcut, optionally combined with a spoken command	Independent LLM configuration and prompt

The system tray would only be responsible for switching between the saved prompts of the Default Profile.

Custom Profiles would not appear in this prompt-selection submenu and would not replace or modify the Default Profile.

This is an intentional separation.

Users who only need Handy's existing functionality would not have to configure any Custom Profiles.

Users with more advanced workflows could introduce additional shortcuts and spoken commands as needed.

---

6. Related GitHub discussions

I found several existing discussions addressing parts of this proposal:

• Multiple post-processing configs #1380 – Multiple post-processing configurations
• Tray menu option to switch the post-processing prompt #2064 – Switching post-processing prompts through the tray menu
• Feature request: Custom keyboard shortcuts for post-processing prompts #1164 – Custom keyboard shortcuts for post-processing prompts
• How Handy would it be if Handy used the clipboard as context? #1249 – Using clipboard contents as LLM context
• [FEATURE] AI writing assistant for selected-text correction and memory-aware generation #1818 – Working with selected text

The intention is not to disregard or duplicate these contributions.

Instead, I'd like to explore how these ideas might fit together while maintaining a clear distinction between Handy's existing post-processing workflow and the proposed optional Custom Profiles.

---

7. Feedback and alternative suggestions are very welcome!

This is a conceptual proposal based on my experience using Handy and SuperWhisper, not a rigid technical specification.

I don't know enough about Handy's internal architecture to judge the implementation effort, and I certainly wouldn't expect all these features to be implemented at once.

The different parts could potentially be developed independently.

I'm especially interested in hearing what the developer, contributors, and other Handy users think about:

• Keeping the existing post-processing configuration as a separate Default Profile.
• Introducing fully independent Custom Profiles with their own LLM configurations.
• Selecting profiles through keyboard shortcuts combined with spoken commands.
• The proposed fallback behavior when no command matches.
• Switching the Default Profile's prompt through the system tray.
• Adding clipboard and selected-text variables.

There may be simpler, more elegant, or more intuitive solutions.

I'd be very happy if other users contributed their own experiences, suggestions, or alternative approaches.

My main goal is to make Handy even more convenient for everyday use while preserving the simplicity that makes it such a great application.

Thank you for reading, and thanks again to everyone contributing to Handy!

[BinaryRider]

Quick update: Due to an unexpected formatting issue when I originally submitted this discussion, several important sections were accidentally omitted.

I've now restored the complete proposal, including the detailed voice-command workflow, fallback behavior, and clipboard/selected-text integration.

Apologies for the confusion, especially to anyone who may have read the incomplete version. The discussion is now complete, and I'd be very happy to hear your thoughts and suggestions!

[timh-86]

YES!!! Custom profiles that combine "1 model + postprocessing on/off + 1 specific post-processing prompt".

Glad you suggested this 'cause I was about to write the same thing.

Also yes to context awareness (clipboard & text selection access).
Speakoflow already has this, although I haven't tried it out yet. I do know it's a fork of Handy.