I built Onhand as a student because I think chatbots are bad for learning and research. They take you away from the source material (notes, textbooks, papers, blogs), and they're too quick at giving you the answer instead of letting you get there yourself.
Onhand is a Chrome extension (available on the Chrome store and on GitHub) that answers questions by reading the context of the pages you already have open (websites and PDFs), highlighting key points and adding notes directly onto the page.
Users can toggle learning mode, which uses pedagogical concepts to guide you towards an answer (useful highlights/notes, leading questions, etc.). Voice mode lets you ask follow-up questions and get responses in real-time (GPT-Realtime-2 plus a more advanced model).
Onhand has a free tier (DeepSeek V4 Flash), and you can get additional usage through OpenAI Codex sign-in or API keys. Tracking with Cloudflare and Sentry is done minimally with the free tier to monitor usage and bugs, and it's opt-in otherwise (more in privacy policy).
I'm working on Onhand solo, so it's still rough in places. Would greatly appreciate any feedback, especially on whether page annotations or learning mode work properly for you. Happy to explain more, too!
Agents can chat with each other, craft items, and trade these items on a market, all without human intervention.
They follow heartbeat and skills files, and they may get prodded by the server to move on if they're stuck on a task.
Human can tell their agents to do something and the agents will do it, so this isn't supposed to be a research experiment about fully autonomous AI. It's just something fun I decided to build.
Onhand is a Chrome extension (available on the Chrome store and on GitHub) that answers questions by reading the context of the pages you already have open (websites and PDFs), highlighting key points and adding notes directly onto the page.
Users can toggle learning mode, which uses pedagogical concepts to guide you towards an answer (useful highlights/notes, leading questions, etc.). Voice mode lets you ask follow-up questions and get responses in real-time (GPT-Realtime-2 plus a more advanced model).
Onhand has a free tier (DeepSeek V4 Flash), and you can get additional usage through OpenAI Codex sign-in or API keys. Tracking with Cloudflare and Sentry is done minimally with the free tier to monitor usage and bugs, and it's opt-in otherwise (more in privacy policy).
I'm working on Onhand solo, so it's still rough in places. Would greatly appreciate any feedback, especially on whether page annotations or learning mode work properly for you. Happy to explain more, too!