ContexTabs

Inspiration

We wanted to make AI feel like it actually understands the user’s current context instead of starting from a blank prompt every time. As people bounce between tabs, chats, docs, and research pages, it’s easy to lose the key facts that make a good prompt. ContexTabs was inspired by the idea of a lightweight “ambient assistant” that quietly gathers the most relevant context from the current page and nearby tabs and helps users turn a rough thought into a clearer, more grounded prompt.

What it does

ContexTabs is a Chrome extension that detects AI prompt fields and adds a floating helper beside them. When a user clicks the widget, it gathers context from the active page, selected text, recent tabs, and saved preferences, then sends that information to a local API. The API uses Gemini to rewrite the user’s prompt into a more specific, higher-signal version and inserts it back into the field.

The extension also includes a related-tabs suggestion panel that surfaces grouped research tabs and lets users keep or dismiss those clusters, plus support for saved research sessions and approved memory preferences. It does not answer the task for the user, instead, it helps them ask better questions with the right context attached.

How we built it

We built the project as a monorepo with a browser extension and a local Fastify API. The extension uses Plasmo, React, and TypeScript to detect editable fields, mount a floating widget, and collect browser context from the page and active tabs. The background script handles tab metadata, grouping logic, and local storage.

The API is a Fastify server that receives the user’s prompt plus page/tab context, then calls Gemini via the Google GenAI SDK. The prompt engineering layer filters noise, prioritizes relevant evidence, and returns a refined prompt optimized for downstream AI use. We also used structured schemas and local storage to keep the flow manageable and reliable in a hackathon environment.

Challenges we ran into

One of the biggest challenges was figuring out how to reliably detect real AI prompt fields across different websites without interfering with normal form inputs. Each site has different DOM structure, selectors, and behaviors, so the extension had to be careful to avoid false positives.

Another challenge was balancing usefulness and latency: collecting page and tab context adds time, but too much context makes the prompt noisy or slow. We had to tune what data gets included and how aggressively the tool triggers. We also ran into the normal hackathon pain points of local dev setup, stale port conflicts, and environment issues while debugging the API and extension together.

Accomplishments that we’re proud of

We’re proud that we built a working end-to-end flow where a user can focus a prompt field, click a lightweight helper, and get a refined prompt based on live context from the browser. The project demonstrates a real concept: turning ambient browsing context into better AI prompts instead of forcing users to manually paste links and summaries.

We’re also proud of the system design around it: local tab/context collection, saved memory, clustered tab suggestions, and a modular API that can be expanded beyond the hackathon demo. It feels like a real product concept, not just a mocked prototype.

What we learned

The biggest lesson was that prompt quality is not just about model intelligence; it’s heavily about context quality. A good prompt is often the result of gathering the right evidence and pruning the irrelevant parts. We also learned that browser extension UX needs to be subtle and respectful: it should help without interrupting the user or feeling invasive.

Another important lesson was around shipping an MVP quickly: we had to focus on the highest-value workflow, keep the architecture modular, and leave room for future improvements like stronger authentication, deeper memory models, and richer task automation.

What’s next for ContexTabs

The next step is to make ContexTabs smarter and more personalized. We’d like to add user authentication, persistent memory beyond local storage, and better source weighting so the extension can prioritize only the most relevant tabs and page content. We also want to support more websites and more agent-like workflows such as task execution, research summarization, and multi-step prompt refinement.

Longer term, we imagine ContexTabs as an ambient AI workspace that helps users move from “rough idea” to “clear, informed prompt” without manually copying context from dozens of tabs.

Built With

• chrome
• fastify
• gemini
• node.js
• plasmo
• react
• typescript