# MindsHub

> An agent workspace with swappable open harnesses and your choice of model

- Category: Agents & Automation / Assistants & Chatbots
- Language: Python
- Licence: MIT
- Repository: https://github.com/mindsdb/mindshub
- Website: https://mindshub.ai
- Canonical page: https://ai-tldr.dev/tools/mindshub/

## In short

MindsHub is an agent workspace for knowledge work and software development. It sits in the Assistants & Chatbots category of the Agents & Automation tools and is open-source, written in Python and licensed MIT. To get started: try it without installing.

## Overview

MindsHub is an agent workspace for knowledge work and software development. Its pitch is model freedom: instead of being tied to one vendor's assistant, you pick which models and providers the agents run on — frontier models such as Claude, GPT and Gemini, or open ones such as DeepSeek, Qwen and Kimi — through a built-in model router that spares you wiring up a key per provider.

The workspace pairs those models with connected data and reusable output. A secure vault links systems like BigQuery, Postgres, Gmail, Drive, HubSpot, Notion and Linear, with credentials scoped per connection so agents never see raw keys. Agent output becomes artifacts — documents, dashboards, apps, analyses, code — which can be published to a live URL and shared. Cross-session memory, a reusable skill library and scheduled tasks let the same work run again without being re-explained.

The agents themselves are open and interchangeable: MindsHub ships open-source harnesses (Anton as the default, plus Hermes) that you swap from a dropdown. The GitHub repository is the platform superproject — it pulls together the desktop and web app, the agent backend and the data engine as submodules, so the whole stack can be built and run from source. It runs on your machine, in your VPC, on-prem or air-gapped, or through the hosted console; the hosted product has a free tier with a paid Pro plan.

## What it does

- Model router across frontier and open models — Claude, GPT, Gemini, DeepSeek, Qwen, Kimi — without per-provider key wiring
- Interchangeable open-source agent harnesses (Anton and Hermes), swapped from a dropdown
- A credential vault connecting BigQuery, Postgres, Gmail, Drive, HubSpot, Notion and Linear, scoped per connection
- Artifacts: turn agent output into documents, dashboards, apps and code, publishable to a live URL
- Cross-session memory, a reusable skill library, and tasks that run on a schedule
- Runs as a macOS/Windows desktop app, a web app, or from source for cloud, VPC, on-prem and air-gapped deployment

## Getting started

The fastest route is the hosted console or a desktop download; building from source is the path on Linux and for development. The commands below come from the repository README.

1. Try it without installing

   Sign in at console.mindshub.ai for the web app, or download the macOS .pkg or Windows .exe desktop build from the README's links.

2. Clone the superproject

   This repository pins the frontend, core API, agent backend and data vault as submodules, so clone recursively.

```bash
git clone --recurse-submodules https://github.com/mindsdb/minds.git
cd minds
```

3. Install dependencies and run

   make setup installs the stack. Then choose a run mode: make dev for the Electron desktop app with hot reload, make dev-web for the browser app, make build for a production build.

```bash
make setup
make dev        # desktop app with hot reload
# or: make dev-web   # web app in the browser
```

4. Reset a broken install

   make flush removes the local runtime and deletes app state in ~/.anton (provider keys) and ~/.cowork (database, hermes, projects), then the next make setup reinstalls everything. It prompts for confirmation and it does delete your conversations and saved keys.

```bash
make flush
```

## When to use it

- Give a non-engineering team an agent workspace that builds reports, dashboards and internal tools they can publish and share
- Run agents against your own databases and SaaS systems without handing raw credentials to the model
- Keep an agent platform inside your VPC, on-prem or air-gapped while still using frontier models where allowed
- Compare agent harnesses and models on the same task by switching both from a dropdown