Overlay: Your AI-Powered Browsing Assistant
Hey Indie Hackers! I'm excited to share Overlay, a Chrome extension I've been working on that transforms your browsing experience using LLMs.
Website:
https://overlay.one
What is Overlay?
Overlay is an intelligent browsing companion that integrates with your Chrome browser to provide context-aware assistance, content analysis, and task automation—all while respecting your privacy.
Key Features:
- Multi-Model Support: Works with OpenAI, Claude, Gemini, and even local Ollama models
- Text processing: Summarization
- Structured Data Extraction: Pulls relevant information from any webpage
- Sidepanel Integration: Seamlessly fits into your browsing workflow
- Privacy-Focused: Options for local models if you prefer to keep your data private
- Image Support: Analyze and understand visual content on webpages
- PDF Support: Analyze and understand pdf content on webpages
- Prompt Manager: Intelligently manages prompts across different LLM providers for optimal results
- Tab Contexts: Maintains contextual awareness across your different browser tabs
- Note-Taking: Quickly capture ideas and information while browsing without switching contexts
- Google Tasks Integration: Create and manage tasks directly from your browser
Why I Built This
I wanted to create a tool that actually understands user intent while browsing and makes the web more productive without being intrusive.
The best part? Overlay is completely FREE right now as I gather feedback and continue improving it.
Would love for you all to check it out and let me know what you think! Any feedback on features, use cases, or improvements would be incredibly valuable.
Solid launch. Three things from shipping something in the same lane:
Multi-model routing: the per-question destination switch is the right axis - once users actually start thinking "this question is a Claude job, this one Gemini for grounded, this one local Ollama for cost-sensitive", single-model extensions start to feel limiting. The Ollama path especially buys you the privacy segment that wouldn't touch a hosted model.
Sidepanel vs popup: sidepanel keeps the source page visible while the model answers, which matters when the trigger is "explain this chart" or "rewrite this paragraph" - the user goes back and forth. Popup feels snappier the first click but the icon gets lost. One thing I'd watch: per-tab context retention - does the sidepanel state reset on tab switch?
Image + PDF together: that's the combo I see under-served. Most extensions pick one (text-focused vs screenshot-focused). Having both in one trigger means a research paper PDF -> page summary -> annotate a chart in the PDF -> ask the model about the chart, all without alt-tabbing.
Curious - how are you handling the "send to multiple models and compare" workflow? Single prompt, one model, or fan-out?
Interesting! What model do you use for this? Some models are really much smarter than others. It will be great if the best models can be used.
You can use all the models from Gemini and Anthropic as of now. Sonnet 3.7 & Gemini 2.5 Pro can be used for free. You just need to sign in and start using them.