Build and run internal AI agents

Hubzoid is an open-source platform for internal AI agents. Give it your team's knowledge, rules and tools once, then use them in chat, workflows and the AI assistants you already use.

Ask it questions in chat, hand it repeatable work, or connect your own assistant to it over MCP. The rules live in files you can read, edit and keep in Git. Hubzoid is open source and runs on your laptop or your own server.

Terminal

pip install hubzoid
hubzoid init my-hub
hubzoid run my-hub

Three commands give you a working agent on your laptop. The quickstart walks through them and your first question.

**Hubzoid: one brain for your internal agents**A short film on what one shared Hub gives a team.Watch on YouTube (opens in a new tab)

Start here

1. Install

Install Hubzoid with pip, on Windows or in Docker.

2. Quickstart

Create a Hub, ask your first question and check the answer.

3. Ready-to-use agents

Start from Software Factory, Order & Invoice Desk, Customer Desk or Engineering Desk.

What one Hub gives your team

ChatWorkflowsEvalsAccess

A refund question answered in the Hubzoid web app from knowledge/house-rules.md, with the Read knowledge step shown. — Your team asks, and the agent answers from your files.Web chat

The Workflows tab of an agent in Settings, with a Monday brief that runs every Monday at 09:07. — Repeatable work runs on a schedule, from a webhook or on request.Markdown tasks

The Evals tab of an agent in Settings: one of one checks passed. — Real questions check that the agent keeps answering well.Evals

The Access tab of an agent in Settings, listing the people who can use it and what each may do. — You choose who can use each agent and each sensitive tool.Agents and access

Your assistant can use the Hub too. Claude Code and other MCP clients connect after you sign in and approve the connection, and they use the Hub with your access. See connect an assistant.

How it fits together

Everything reaches the same Hub through one bridge. The bridge checks who is asking, runs the agent or workflow with the Hub's knowledge and tools, and calls the model and systems you chose.

Where your team works

Chat
The Hubzoid web app, Slack, WhatsApp, Telegram, your own app and the OpenAI-compatible API.
Workflows
Markdown tasks and code workflows on a schedule, from a webhook, on request or from chat.
Your assistant
Claude Code and other MCP clients, through /mcp after sign-in and consent.

Hubzoid, on your machine or serverOne bridge per Hub

Identity and access
Signs people in, checks who is asking before the agent or a restricted tool runs, and records each tool decision.
Agent runtime
Builds the agent from the Hub and runs it on the model you chose.
Workflow engine
Runs scheduled, webhook and requested work durably, with a run history.

The Hub, your folder in Git

AGENTS.mdknowledge/skills/tools_local/restricted/schedule/

Instructions, knowledge, skills, tools and recurring work. Every service above reads the same files.

Settings
/settings
Administrators add people, grant access per agent and follow usage and runs. Everyone sees their own schedules and personal settings.
Your data
Accounts, conversations, grants, runs and usage, in SQLite by default or PostgreSQL.

What it calls

Your model
Claude Code or Codex signed in on the machine, or OpenAI, Anthropic, Azure OpenAI or OpenRouter with a key.
Your systems
Only through the tools and MCP servers you add to the Hub.

Assistants connected over MCP use the Hub's tools and knowledge with their own model.

Architecture follows one request through every process, and project structure lists every file a Hub can hold.

Build something

Put a task on a schedule

A Markdown file the agent runs every Monday, with its result in Settings.

Let the agent use a system

Add Python tools and MCP servers so the agent can look things up and act.

Restrict a sensitive tool

Only people you grant can make the agent call it.

Connect your assistant

Use the Hub from Claude Code or another MCP client.

Add it to Slack

Your team asks the agent where they already talk.

Run it for your team

One server, sign-in on, backups and upgrades.

Who it is for

Builders understand a business process and already get useful results from AI. They write the Hub and test it on a laptop. Teammates chat with the agent and use its reports without knowing how it is built. Administrators add people, grant access per agent and follow usage, runs and cost in Settings.

Building with Claude Code or Codex? The quickstart has a prompt that sets up a Hub for your own job.

Introduction