All projects
Build JournalContact
V3 in development · Live AI under testingWynn Orbit · internal business tool

Wynn Orbit AI Agent Hub

One workspace for a small business’s tasks, projects, reports and approvals, with an AI assistant that suggests work but never changes anything without approval.

The problem

Running a small software business means tasks, projects, reports and decisions end up scattered. I wanted an AI assistant that helps me decide what to work on, but which keeps me in control of every change.

What it does

The home screen opens with a question: “What can I help you with today?” From there you can ask the assistant to organise tasks, check projects or prepare a report, or use one of the quick actions: What should I work on today?, Show my unfinished tasks, Summarise my projects and Prepare a weekly report.

Demo

The screenshot above is the only published view of the Agent Hub. It shows the home screen with sample data in Demo Mode.

A short screen recording is planned. It will show only features that work reliably today, using sample data and no real business information.

How it works

  1. Interface: Python and Streamlit. It runs locally and opens in the browser.
  2. Workspaces: each business has its own workspace, chosen from the sidebar, so tasks and projects never mix.
  3. Assistant: in Demo Mode it answers from sample data. A Live AI mode connects it to a language model, and that connection is under testing. Either way, suggested changes go to Activity & Approvals and are never applied without approval.
  4. Planned: stable Live AI conversations and conversation memory.

Feature status

Kept deliberately strict: only what has been checked is listed as working. Updated November 2026.

Implemented

  • A business switcher with a separate workspace for each business.
  • My Tasks, with counts of unfinished and overdue tasks.
  • My Projects, with the percentage of tasks done.
  • Business Reports drafts and a Needs Approval queue.
  • Focus next, ordered by priority and due date.
  • Recent activity and a clear Demo mode label.

Verified with offline tests

  • The V2 build notes record 113 passing automated tests.
  • V3 test results will be added after they are re-run and checked.

Verified with a live API call

  • With Live AI turned on, V3 has returned at least one real response from the language model.

Known bug

  • A later follow-up request in the same conversation failed with RuntimeError: Event loop is closed. This is being investigated.

In development

  • Live AI integration: under testing. It is not yet stable enough for everyday use.
  • Conversation memory.

My role and AI-assisted development

What I did

  • Came up with the product, defined the workflows and set the rule that the AI never changes anything without approval.
  • Designed the features, tested each version and decided what goes into it.

What AI coding assistants did

  • Wrote much of the Python code and the automated tests under my direction.

Design challenges

Keeping a human in control

AI suggestions never act directly. Changes go through Activity & Approvals, and the assistant says so up front: “I never change anything without your approval.”

More than one business

Each business has its own workspace, chosen from the sidebar, so tasks and projects never mix.

Showing what matters first

The dashboard leads with what needs attention: overdue work, a focus list sorted by priority and due date, and anything waiting for approval.

Privacy

This is a private internal tool. The screenshot shows demo data. There is no public demo, and no business data or code is published.