Before we build a command center for a client, we run on one ourselves. This is the hub the consultancy operates from: one browser tab that holds the task board, the CRM, the projects, and a log of everything that changed.
It is built for one person working alongside several AI terminals at once, so its job is to keep the human and the AI pointed at the same list.
Business: Mitchell B Consulting
Users: One consultant and his AI terminals
Runs: Locally, on a laptop
Our role: Built it, and use it every day
One person, many AI sessions, one source of truth
Running several AI terminals in parallel is fast, until two of them edit the same thing or nobody remembers what was decided yesterday.
The hub exists so every session starts from the same board, every change is logged, and nothing a person or an AI saves can overwrite the other's work.
The business views
The hub also holds private views that are not shown here. These are the ones the consultancy runs on.
Today: The start of the day: what needs a decision, what is waiting on the owner, and the open milestones across every project.
Tasks: Milestones grouped into lanes, where each lane is one AI terminal. Each milestone generates its own copy-and-paste session prompt.
Backlog: A full ticket backlog inside the same hub: epics, sprints, story points, and drag-to-rank, with the AI as an assignee.
MBC CRM: The consultancy's own CRM: leads, accounts, contacts, deals, and activities, with a dashboard that flags deals missing a next step.
Deals: Deals as a board or a table. Every move between stages is kept in a stage history, so the path of a deal is never lost.
Operations: Engagements with their milestones and hours, the invoices sent, and the payments received.
Products & Pricing: The services catalog, pricing models, packages, and pricing rules, in the same place the proposals are written from.
Reports: Pipeline by stage, source, and industry, plus activity over time. Every table exports to CSV.
Projects: Each project with its progress, why it matters, and the next action.
Log: A change log of everything that was built or decided, with a weekly view and a drafted executive summary.
Library: Every file in the workspace, indexed and searchable, opening in a viewer without leaving the hub.
What makes it work
Milestones in lanes: Work is organised as milestones, grouped into lanes. Each lane is one AI terminal, and a lane with two active milestones gets a warning.
A prompt for every milestone: Each milestone carries what done means, why it matters, and its steps, and generates a ready-to-paste prompt to start the session.
Who does what, step by step: Every step is marked as the person's (decide, send, call, approve) or the AI's (build, draft, analyze, research). The AI drafts; the person sends.
A backlog like a product team's: Epics, sprints, story points, and drag-to-rank, with the AI as an assignee next to the human.
Saving that never loses work: Each save is merged with whatever changed on disk since the screen loaded, so an AI terminal and an open browser tab can both write without clobbering each other.
A CRM with a memory: Leads, accounts, contacts, deals, and activities. Every stage change is kept, and deals with no next step are flagged.
A lead qualification scorer: Weighted criteria, hard disqualifiers, and a live score, so the decision to pursue a lead is made the same way every time.
A change log that reports upward: Each entry records what changed and why it matters, with its benefit marked as measured, estimated, or projected. A weekly executive summary is drafted from it.
A day plan: One command writes the day's plan and a kickoff prompt for each lane.
A closing routine: Ending a session updates the board and the instructions, logs the work, and writes a dated wrap-up.
Rules the hub enforces
The AI drafts into a drafts folder. A person sends.
Nothing posts or emails on its own. Reports are written to an outbox for a person to send.
Benefits marked measured, estimated, and projected are never added together.
Board changes go through one locked path, so two writers cannot corrupt it.
Under the hood
A single HTML file
Python standard library server
JSON files with backups
Claude Code skills
No framework, no build step
This is the smallest version of what we build for clients: one screen, one source of truth, and AI that works inside your rules.