hubzoid 2.0.1

One brain for your internal agents.

Your knowledge. Your way of working. Useful across your assistants and your team.

Documentation · Quickstart · Website · Contribute

Hubzoid turns a folder of your knowledge, rules and tools into an agent you can use yourself and share with your team. Ask it questions in chat, run repeatable workflows, or connect your own assistant through MCP.

Start with one useful job: a stock check, a policy answer, or a weekly report. Keep the context and rules in files you can read, edit and version. Add teammates with access to the agents and capabilities they need.

Get your first useful result

Use Claude Code or Codex with the setup prompt below, or run these commands yourself.

Use Python 3.11 or 3.12 in a virtual environment:

python3.12 -m venv .venv
source .venv/bin/activate
pip install hubzoid
hubzoid init my-hub
hubzoid run my-hub

The web app opens at http://127.0.0.1:3080. Ask:

What should we reorder, and what could run out first?

The starter is an operations assistant for a fictional home-goods shop. It uses a stock export and saved policies to explain what needs attention. You can inspect the numbers and rules in my-hub/.

init detects a signed-in Claude Code or Codex CLI. Otherwise it offers to save a model API key in the Hub's private .env. Chat needs model authentication; provider charges or subscription limits apply. Choose a runtime →

Help me set up a local Hubzoid agent using the current quickstart:
https://github.com/hubzoid/hubzoid/blob/main/docs/quickstart.md

Use a new folder and a Python 3.11 or 3.12 virtual environment.
Help me choose an available runtime without pasting secrets into chat.
Run the operations sample and show me one result from its stock data.
Then help me adapt it to one task using material I choose.
Show me proposed file changes before replacing sample content.

Local mode is private to your machine. Turn on sign-in before sharing access. For Windows, uv, troubleshooting and source installation, see the quickstart.

Python 3.12 workflows need SQLite 3.42+ or PostgreSQL. hubzoid doctor my-hub checks the environment. This checkout is 2.0.1; until it is published, pip install hubzoid installs the earlier 1.x release. Use the source-install steps to try this revision. Deployments that still need Open WebUI stay on 1.x (pip install "hubzoid<2").

Make it work your way

Try a correction you can verify:

1. Ask the sample: “A customer opened a blender 40 days ago and wants to return it. What can we offer?” Its saved change-of-mind policy is 30 days.
2. Ask Claude Code or Codex to change the change-of-mind window to 45 days in my-hub/knowledge/returns-policy.md, keep the faulty-item rules unchanged, and show you the diff. Or edit the file yourself.
3. Restart the Hub and ask again. The answer should apply the saved 45-day window and the policy's other conditions.

Then replace the fictional policies and stock export with your own material. Only explicit Hub content is shared; a correction in chat alone does not become shared knowledge or automatic memory. Learn to author a Hub →

One Hub, three ways to use it

In 1.3.1, offered reports can run from chat and have personal schedules. My workspace → Schedules & reports shows available reports, recent results and your schedules. Edit inputs and timing, pause or stop. Schedules need the Hub running; email notices are off by default. Reports, permissions and sharing →

For Claude Code, use the connection command printed by hubzoid run, then open /mcp in Claude Code and approve the connection. Other clients have their own setup. Connect your assistant →

Pick a starting point

Each sample includes instructions. Watchtower's Service snapshot walkthrough uses a model-free report; chat still uses your configured runtime. GitZoid starts in preview mode with fictional data; reviews still use your configured model. Its setup guide covers GitHub access, existing PRs, publishing and costs. For a small starting folder, use --template minimal.

Share with your team

Turn on sign-in and set the public address in my-hub/.env, then create your first administrator:

HUBZOID_AUTH=true
HUBZOID_PUBLIC_URL=https://hub.example.com

hubzoid admin create you@example.com my-hub --owner

Open the one-time sign-in link, set a password, and add teammates through Settings → People & access. Grant access to the agents and capabilities they need. Shared knowledge does not mean shared credentials or access to another person's private results.

For a public deployment, configure HTTPS and persistent storage, then restart. Team setup · Deployment · Backup and restore

Optional Claude failover (1.3.2 and later): add up to two backup credentials to keep answering when a Claude credential reaches a confirmed usage limit or its login is rejected. Switching keeps the same model and permissions and never replays work that has started. Existing single-credential setups are unchanged. Configuration, token replacement and limits →

The web app

pip install hubzoid includes the chat and its Settings, built with assistant-ui and bundled with Hubzoid. No separate UI package or Node installation is needed. Hubzoid 2.0 no longer offers Open WebUI; deployments that still need it stay on 1.x. When upgrading an existing deployment, read the upgrade guide first.

Go further

All guides · Website documentation · Release notes

Hubzoid is Apache-2.0, including team controls. Dependencies and brand assets retain their own terms. Licensing. To contribute, start with the contributing guide. Need help with your company's agent? Build with us uses the same open-source product.