Open source · framework neutral
Your app.
Ready to answer.
Give any web application a contextual assistant grounded in its rendered UI, a selected element and trusted documentation.
$npm i @123toto/ai-app-assistant-server @123toto/ai-app-assistant-clientnpm i …server …client
Requests
Approval is waiting
One required check is still incomplete.
How answers stay relevant
Useful context in.
Grounded help out.
The assistant combines what the user sees, what they point to and the trusted knowledge supplied by the application.
Library structure
Four parts.
One assistant.
Each part has a clear role and can evolve independently. Use the included options today, then add a focused adapter where your application needs one.
Browser clientCaptures the rendered page, selection and conversation state.
Assistant serverConnects application context, documents and streaming responses.
Model contractKeeps the answer format independent from the chosen model.
Document contractTurns trusted application knowledge into reusable sources.
A faithful, zero-cost simulation
Try the real
interaction flow.
The demo reproduces the library’s launcher, element picker, selection event, streaming phases, structured answer and confidence display—without calling an LLM.
Launch the guided demo →- 01Open the assistantThe real sparkling launcher expands into the panel.
- 02Select a UI elementThe panel hides while the picker focuses the page.
- 03Ask in contextThe selection, page and documents support the answer.
- 04Check reliabilityStructured output shows confidence and limitations.
Privacy controlled by your app
Your context stays under your control.
✓ Your application chooses which page areas are shared or excluded.
✓ Form values and marked sensitive areas are excluded by default.
✓ Identity, permissions and privacy rules remain application controlled.
✓ Answers are structured, validated and linked to retained evidence.
Two packages. One boundary.