Local-first by design
Conversations, assistants and settings live in SQLite on the machine you run it on. Provider keys are encrypted at rest.
self-hosted · open source · MIT
A personal AI agent workspace that asks before it acts.
Arvind Dots is a small, inspectable workspace for chatting with assistants, connecting apps, and running computer tasks, with every risky step routed through an approval prompt. Run it on your own box.
Approval-gated actions
Workspace and computer operations go through a deny-by-default gateway. Higher-risk actions pause for your approval and leave an audit trail.
Bring your own model
Point it at any Responses-compatible endpoint, list the model IDs you want, and switch models per assistant from the chat header.
how it works
Three steps to a working assistant
- 01
Sign in
Use the owner token the server generated on first start. No accounts, no sign-up.
- 02
Add a model
Open Settings, paste your API base URL and key, and save the model IDs you want to use.
- 03
Create an assistant
Give it a name, a role and a model. Then chat, attach images, and let it ask before it acts.
What it is, and what it is not
It is
A single-owner workspace with streaming chat, image attachments, a connector marketplace, an optional Docker computer runtime, and an audit log of every gated action.
It is not
A multi-user product, a hardened sandbox for hostile websites, or a managed cloud service. It is a personal tool you run and inspect yourself.