Gnom-Hub is a local multi-agent control hub for your desktop. You brainstorm freely with AI agents; workers that cost money or change files only run when you press Execute.

Does Gnom-Hub need Docker?

No. Gnom-Hub runs with a Python virtualenv and FastAPI on localhost. No Docker or Kubernetes is required.

How is Gnom-Hub different from LangGraph or autonomous agents?

Gnom-Hub is a visible desk: Send stays dialogue-only; Execute starts workers on purpose. It is not unattended full autonomy and not a LangGraph drop-in framework.

Which LLM does Gnom-Hub use?

DeepSeek (deepseek-v4-flash) by default, optional Ollama for local models. Keys live in User/Key.txt. Optional Tollgate integration for spend control.

Local multi-agent control hub

Brainstorm freely.
Execute only on purpose.

Gnom-Hub is the desktop desk for multi-agent AI work. Send stays dialogue. Workers start when you press Execute — not on every message.

Install locally View on GitHub Documentation

multi-agent hub
local AI agents
no Docker
Python · FastAPI
DeepSeek · Ollama
computer-use safe

“Brainstorm freely. Execute only when you press Execute.” Product rule — exploration stays cheap and reversible

Product

What Gnom-Hub is

A local multi-agent control hub for desktop work: visible agents, honest keys, HTML deliverables, safe tools — without cloud lock-in.

01 · BRAINSTORM

Send = dialogue only

Talk, plan, refine. No silent worker spend until you decide to run.

02 · EXECUTE

Workers on purpose

Cost, files, and side effects start with Execute — cancel and cost stay visible.

03 · DESK

See who works where

Agent cards, work boxes, tools badge, light trace — a desk, not a black box.

How it works

From idea to deliverable

One control plane for planning and intentional execution.

You Chat · tasks · wishes

Gnom-Hub desk Brainstorm → Execute gate

Workers + tools HTML · search · computer-use

Local & portable — User/Key.txt, USB-friendly data, no Docker
Honest auth — placeholders never fake success
Safe computer-use — dry-run until God-Mode
Skills & memory — playbooks, HOT/WARM, optional vector search
Tollgate-ready — optional spend control on the LLM path

Desk UI

Visible multi-agent workspace

Agent cards · work boxes · chat — built for real desktop sessions.

Gnom-Hub multi-agent desktop UI showing agent cards, work boxes and chat panel — Gnom-Hub desk — agent cards, boxes, and conversation

Gnom-Hub tools modal with computer-use and core tool registry — Tools modal — core tools and computer-use controls

Positioning

Not another autonomous agent swarm

	Gnom-Hub	Typical agent frameworks
Core job	Visible desk · intentional Execute	Graph / autonomy plumbing
Send message	Dialogue only	Often starts tools immediately
Deploy model	Local venv · no Docker	Often containers / cloud
Computer control	Dry-run until God-Mode	Varies · often wide open
Best for	Desktop multi-agent ops	Library integration

Install

Run on localhost in minutes

Python ≥ 3.10 · FastAPI · UI at http://127.0.0.1:8080/

1 · Install
./scripts/install.sh
2 · Keys
User/Key.txt with real DEEPSEEK_API_KEY
3 · Start
./scripts/start.sh → open desk
4 · Work
Chat freely · press Execute for workers

Full install guide

quick start

# Local multi-agent hub — no Docker
git clone https://github.com/landjunge/gnom-hub-v1.git
cd gnom-hub-v1
./scripts/install.sh && source .venv/bin/activate
# Key.txt → User/Key.txt with DEEPSEEK_API_KEY
./scripts/start.sh
# → http://127.0.0.1:8080/

FAQ

Common questions

What is Gnom-Hub?

A local multi-agent control hub: brainstorm first, execute workers only when you choose.

Docker required?

No. Python venv + FastAPI on your machine. Portable data folder, USB-friendly.

vs autonomous agents?

Visible desk and intentional Execute — not unattended full autonomy by default.

Open source

On GitHub

Repo: github.com/landjunge/gnom-hub-v1 · Version 3.10.x · Private-use license

★ Star on GitHub Docs index + Tollgate gate Deutsch Tollgate site

Brainstorm freely.Execute only on purpose.