Voice actions for every app

Speak once.
Invoke handles the rest.

Turn speech into messages, notes, snippets, searches, and app actions. Use local AI through Ollama, connect tools through Composio, and keep control of your data with privacy-first setup.

View GitHub See how it works

Listening

Natural speech to structured actions

"Clean this sentence and make it sound professional."

TEXT_CLEANUPready

"Create a GitHub issue for the login bug."

GITHUB_CREATE_ISSUEqueued

"Search the web for Android privacy changes."

COMPOSIO_SEARCH_WEBdone

Voice input > Whisper > Qwen 3 0.6B > Composio action

The Problem

Typing, switching apps, and copying text slows everything down.

Most productivity tools still make you open five windows, type a prompt, copy the result, and paste it somewhere else.

Dictation apps stop at text. Invoke goes further: it understands what you meant, picks the right route, and helps finish the action.

What Invoke Does

Use your voice like a command bar.

Invoke understands natural speech and turns it into structured actions. Dictate anywhere, clean up rough thoughts, save reusable snippets, and trigger workflows across connected apps.

Voice actions across apps

Speak naturally and turn commands into messages, notes, snippets, searches, and tool actions.

Local intent routing

Qwen 3 0.6B runs through Ollama for fast local classification without needing a cloud LLM for core routing.

Composio tool execution

Connect Gmail, GitHub, Slack, Calendar, Notion, Todoist, Docs, and web search through one action layer.

Writing cleanup and snippets

Dictate rough thoughts, clean tone, apply style presets, and save reusable phrases or shortcuts.

Real Life

Built for the moments where typing falls apart.

Invoke is not only for quiet desks. It is for commutes, shared spaces, airport gates, and rushed work moments where voice is the fastest input you still control.

Noise-aware capture + confirmation

Crowded commute

Review a PR while standing in a packed train.

You do not have a seat, your laptop is closed, and the train is loud. Open the Android mic bubble, use an earbud mic, and say what you want reviewed.

"Review the auth PR, summarize the risky files, and draft one comment about token refresh."

Invoke transcribes, classifies the GitHub action, drafts the review note, and asks before posting.

Whisper-friendly dictation

Shared office

Whisper a polished reply without disturbing anyone.

You are in an open office or library and need to answer quickly. Speak quietly into a close mic and let Invoke clean the wording.

"Reply to Sarah that I can join tomorrow, make it warm and professional."

Invoke drafts the message with your work style preset and keeps private mode local-first.

Multi-step routing

Between gates

Turn a passing thought into an organized task before it disappears.

You are walking through an airport with one hand free. Instead of opening Notion, Todoist, and Calendar, capture the whole workflow by voice.

"Save this as a product idea, create a follow-up task for Friday, and search examples of Android voice bubbles."

Invoke routes each part to the right connected tool and shows a clear action summary.

How It Works

Speak naturally. Invoke routes your words into actions.

The core architecture is simple: voice input, speech-to-text, local intent classification, tool execution, then a result or completed action.

Record

Invoke captures your voice from desktop or Android.

Transcribe

Whisper converts speech into text locally.

Classify

Qwen 3 0.6B maps intent to a structured action.

Execute

Composio or local tools complete the task.

Local AI

Powered by Qwen 3 0.6B through Ollama.

Invoke does not use Qwen as a big chatbot. It uses the model for one focused job: understanding what the user wants and converting speech into structured actions.

Small enough to run locally

Fast for intent classification

No cloud LLM required for routing

Upgradeable for larger models

Intent payloadlocal

User:
"Search the web for OpenAI official website"

Qwen 3 0.6B:
{
  tool: "COMPOSIO_SEARCH_WEB",
  parameters: {
    query: "OpenAI official website"
  }
}

Invoke:
execute(action)

Integrations

One voice layer for the tools you already use.

Composio lets Invoke connect spoken intent to real app actions, from developer workflows to everyday messages.

Gmail

"Draft an email to Alex about the project update."

Calendar

"What meetings do I have tomorrow?"

GitHub

"Create an issue for the login button bug."

Slack

"Tell the team I'm joining late."

Notion

"Create a note from this idea."

Web Search

"Find the best local AI models for Android."

Desktop + Android

One assistant across computer and phone.

Windows desktop app

Tauri desktop app with global shortcuts, local Whisper transcription, Ollama settings, Qwen 3 0.6B model support, and Composio actions.

Android voice bubble

Kotlin Android app concept with floating mic access, permissions onboarding, privacy mode, dictionary, styles, snippets, and advanced setup when needed.

Privacy

Private by design.

Invoke supports local AI through Ollama, so core intent classification can run on your own machine. Privacy mode keeps data stored on your device. Cloud sync and app integrations are optional and controlled by the user.

Examples

Real commands, real outcomes.

"Search the web for the latest Kotlin Android updates."

"Create a GitHub issue titled login crash."

"Draft an email to Maya about the invoice."