AskAI

A small Android helper that gets selected text or part of your screen to an AI in a couple of taps. Two ways to send, your choice:

• Open an app: hand the text or screenshot to the Claude, ChatGPT, Gemini, Perplexity, Grok or Copilot app. No keys, no accounts.
• Use an API: ask the Claude API or any OpenAI-compatible endpoint with your own key. The answer streams into a popup you can copy from.

What it does

Prompts

Save as many prompts as you like, each with a short name such as Summarize or Translate, and drag the handle to put the ones you use most at the top. When you send something, a picker shows the names in that order; with a single prompt there is nothing to pick and it is used straight away.

Selectable text

Select text in any app and pick AskAI from the selection menu (tap the ⋮ overflow if it is hidden). Choose a prompt. In open-app mode the text opens in your AI app with the prompt appended; in API mode the answer appears in a popup.

Text you can't select

Some apps (LinkedIn, images, PDFs) don't let you select text. For those:

1. Pull down Quick Settings and tap the AskAI capture tile.
2. Allow screen capture.
3. A screenshot appears with a box on it. Drag the box to move it, drag an edge or corner to resize it.
4. Tap Send and choose a prompt.

The status bar, navigation bar and display cutout are cropped out, so the image only contains app content.

Start it with a gesture. Set AskAI as the digital assistant app (AskAI settings → Choose assistant app, or Android Settings → Apps → Default apps → Digital assistant app). The assistant gesture then starts a capture: swipe up from a bottom corner with gesture navigation, or long-press Home with button navigation. If the assistant is allowed to use screenshots (a switch on that same settings page) the capture skips the consent dialog entirely; otherwise the usual consent flow runs. This replaces Gemini / Google Assistant on that gesture until you switch back.

API mode

Pick Use an API in settings, choose Claude or OpenAI-compatible, and enter a key. The model and, for OpenAI-compatible providers, the base URL can be changed, so Gemini, OpenRouter, Groq or a local server all work. The answer streams into a bottom popup. Copy puts it on the clipboard and closes the popup. Screenshots are sent as images, so the model must support vision. Keys are stored with EncryptedSharedPreferences.

Open-app mode and screenshots

AI apps ignore text that is shared together with an image, so in open-app mode the app has two ways to deliver the prompt with a screenshot:

• Write prompt into screenshots (default): the prompt is drawn as a caption under the image and the AI reads it from there.
• Off: the prompt is copied to the clipboard; long-press the chat box in the AI app and tap Paste.

Setup

1. Build and install AskAI (see below).
2. Open AskAI and pick how to send: install an AI app, or enter an API key.
3. Tap Add tile to Quick Settings. On Android 12 and older, open Quick Settings, tap the edit (pencil) button and drag the tile in.
4. Grant the notification permission when asked. It is only used for the short-lived "Capturing screen" notification Android requires during capture.

Building

Requirements: Android Studio (or a JDK 17–21 and the Android SDK), minimum Android 8.0 (API 26), target API 34.

./gradlew :app:assembleDebug
adb install -r app/build/outputs/apk/debug/app-debug.apk

Gradle 8.7 does not run on JDK 25, so if Android Studio's bundled JDK is that new, point JAVA_HOME at a JDK 17 or 21.

How it works

Piece	Role
ProcessTextActivity	Handles ACTION_PROCESS_TEXT from the selection menu, shows the prompt picker, forwards the text.
PromptPicker	The prompt chooser shared by both paths; skipped when there is at most one prompt.
AnswerActivity / LlmClient	API mode: streams from the Claude Messages API or an OpenAI-compatible /chat/completions endpoint over raw HTTP and shows the result.
CaptureTileService	The Quick Settings tile; launches CaptureActivity.
AssistService / AssistSession	Registers AskAI as a digital assistant. On the gesture, uses the system-provided screenshot when available, else launches CaptureActivity.
CaptureActivity	Invisible; asks for MediaProjection consent and measures the system bars.
CaptureService	Foreground service that grabs one frame through a VirtualDisplay, strips the bars and saves a PNG.
CropView / CropActivity	The movable, resizable box over the screenshot.
Caption	Draws the message under the cropped image.
AiApp	The supported AI apps; sends text or an image to the chosen one, or opens the Play Store if it is missing.
Settings	Send mode, chosen AI app, API provider and keys, prompts, burn-in toggle. Encrypted.

Android asks for screen-capture consent on every capture. That is a platform rule and the app can't skip it. On Android 14+ the dialog is limited to the entire screen so the single-app picker never appears.

Notes

• Adding another AI app means one line in the AiApp enum and one <package> entry in the manifest <queries> block.
• Screenshots are written to the app's private cache and overwritten on the next capture.

License

MIT