Show HN: Anjadhe – privacy first AI assistant, no account, no server DB

Show HN: Anjadhe – privacy first AI assistant, no account, no server DB(anjadhe.com)

1 points by ram-bv 23 days ago | 1 comment

Hi HN, I am Ram. For the last few months I have been building Anjadhe. It is a personal AI assistant for macOS. The main idea is simple: A macOS app where AI does the job of a personal assistant for the user. Not a chat only app where user needs to dig through chats to understand today’s schedule or a project they planned with the ai last week, but a canvas where the user and AI work together around basic tools.

I started building this as DYI tool for myself, but at this point I see that its truly useful for me, so sharing with others to see if it can be useful to others too.

Here is a quick Demo - https://www.anjadhe.com/demo

What it can do today:

- It reads your incoming email and files the important things: bills with due dates, renewals, receipts, deliveries. A bill becomes a task with a date, by itself. Two booking emails become one trip. - You set goals by talking to it, not by filling forms. It asks you questions, makes a plan with dated tasks. Later you can say "work got crazy, push everything two weeks" and it moves the whole plan. It always shows you what will change and asks before doing it. - Routines run without you. On a schedule, or when a certain email or file arrives. Every run leaves a log you can read. - You can give it some documents you wrote. It learns your writing style and writes new content in your voice. What it learned is a page you can read and edit, not a black box.

About privacy: the AI runs on your Mac with llama.cpp. Or you can point it to your own server, or use your own OpenAI/Anthropic key, or use Anjadhe Cloud (open-weight models, free allowance, no account needed). Your data stays in SQLite on your own disk. No account. No telemetry unless you turn it on. The source code of the app and the cloud service is public, so you do not have to trust my words.

Honest limitations: it is an early alpha, macOS only. Local models need a Mac with atleast 32gb ram to work. The email features work fine on small models, but the full agent wants a bigger model or a server. Also it is Electron with vanilla JS, no framework. Looking forward to hear your opinions.

Demo (2 min): https://anjadhe.ai/demo Download: https://anjadhe.ai/download Source: https://github.com/Anjadhe/Anjadhe

I would love feedback, especially if you think my privacy claims are wrong somewhere. I will be here to answer questions.

ram-bv 23 days ago |

The stack. Electron, vanilla JS, no framework, no build step. Every app inside (Mail, Tasks, Goals, Portfolio...) is a plain JS object. Records live in SQLite on disk. Decision to use Vanilla JS started as an experiment to write code that will never fail, but eventually libraries were added. Anyone can download the code and run "npm start" and use the app or download the Built app from download page.

I started the project with the hope that i can build an assistant for myself in my 16 gb machine using an open weight model as the brain. but after building it, i realized even models claiming to work in 16gb m series macs does not really work that well when used as brain for an agent. As a chatbot it works. Due to this reason, I have introduced a hosted model option so people can try out the features. But it works well with models like Qwen 3.6 models in mac with more RAM.