About this video
Stop handing your data to OpenAI. This video shows you how to set up Open Dot, a personal AI agent that runs on local models to keep your emails and automations 100% private. Key Takeaways: - Learn to install Open Dot in under 15 mins for local-first automation. - Discover why natural language agents are superior to fragile node-based systems. - Connect Gmail and Notion privately using Composio triggers. - Configure local inference using models like Quen 3.83b to eliminate API costs. - Understand the trade-offs of local agents, including why 100% offline mode isn't always possible.
Your emails are your business, not training data for a trillion dollar corporation.
Stop letting Big Tech read your private messages under the guise of 'productivity'. I have just configured a personal AI agent called Open Dot that runs entirely on a local model. It processed my emails and acted upon them without a single byte of my content touching a cloud server like OpenAI or Grok.
Why Local Agents Matter
Traditional automation tools like n8n are notoriously fragile. One wrong value or a slight edit to a workflow can break the entire chain. Open Dot shifts the paradigm by using natural language to drive autonomous agents. If you can describe it, the AI can integrate it.
The 15 Min Setup
Getting Open Dot running on a Mac is straightforward. You clone the repository, install dependencies via pnpm, and build the application. While the default setup uses API keys, the real power lies in running it from source to enable local inference. By modifying a few environment variables, you can point the agent to a local endpoint like OMLX.
Integration via Composio
To make these agents useful, they need to talk to the world. Composio provides a private way to connect apps like Gmail, Slack, or Notion without storing your login credentials directly on the agent tool. This allows your local AI to be 'woken' by triggers, such as a new email, and perform actions based on your predefined rules.
The Reality Check
You cannot run this 100% offline if you want it to interact with cloud services. Because the triggers rely on external platforms like Gmail, your machine needs a connection to respond to those events. However, the inference, the actual 'thinking' part of the AI, remains entirely on your hardware.
If privacy and cost are your priorities, switching to local models is the only logical move. Keep your data where it belongs: with you.
Transcript▾
My AI dot, which is running off a totally local model, has just gone through my emails and acted upon them without any intervention whatsoever. You do not need OpenAI or Grok to run these personal agents. That is where I have discovered Open Dot. By the end of this video, we are going to get Open Dot up and running, but we are going to be running it completely privately and offline using local models. All it is going to take you is 15 mins of your time, completely free, but it is not for Windows or Linux users, unfortunately.
There is one feature that quietly switches off when you do go local, so stay tuned to avoid that. These dots are autonomous agents driven by natural language. They rely on triggers and automations integrated with the dot, like a new email or GitHub issue, which could then post to Notion or Slack. Traditionally, we had n8n, but they are so fragile. It is better if you are not technically minded to just describe these things with natural language.
To install, clone the repo, run pnpm install, and build it. We are going to run it from source, copy the environment example into local, and run it in dev mode. I have connected my Gmail via Composio, which allows a private way for apps to connect without storing credentials on the tool itself. I set up a 'rap bot' that summarises every email I receive into a rap.
We can take this further by making the inference completely local. I edited the code to add a local TS file that checks for a local endpoint. Using OMLX, I loaded Quen 3.83b. Now, everything the dot does uses this local model. It is a bit slower, but it is free and private. Note that you cannot run this completely offline because it relies on triggers from services like Gmail. If privacy and cost are your top concerns, Open Dot is the winner.