Free · Open source · Apache-2.0
A starting point for running a small team of AI agents inside Claude Code: an orchestrator that delegates, role-based specialists you can rename, a hiring flow for adding new ones, a quality check before work reaches you, and a small local database to keep track of it all.
What it is
Plain-text instructions, agent profiles, standard operating procedures and two small Python scripts that turn Claude Code into a structured team rather than a single assistant. When you open Claude Code in the folder, it reads the instructions and acts as the team’s orchestrator. It is based on the operating model behind KnowledgeDB.io’s own work. No server, no account with us, and you can change anything you like.
Who it is for
Professionals and organisations who want more structure from AI than one long chat:
- Founders and small businesses who want a team of AI specialists with a clear record of who did what
- Developers building AI-assisted applications who want review, memory and task tracking from day one
- Teams exploring how a multi-agent setup can be run with a quality check before anything is delivered
What’s included
An orchestrator
The Claude Code session you type into. It plans and delegates to the right specialist rather than doing the work itself.
Role-based specialists
Five starter team members — HR, Researcher, QA, Knowledge Manager and Engineer — each a plain file with a written profile. Rename them, change their personas, or remove them.
A hiring flow
Ask for a role the team does not have. The Researcher writes a brief on what a skilled human in that role does, and HR turns it into a new team member.
A QA gate
Deliverables are checked before they reach you, with three outcomes: pass, pass with observations, or needs rework.
Daily priority lists
The Daily Seven: your top seven and the team's ranked list, chosen the night before and worked in order, #1 before #2.
Inboxes, journals and SQLite
Folders for your requests and the team's replies, a journal per member, and a small local SQLite database for tasks, reminders, QA reviews and the activity log.
How to get it
Option A — download a ZIP (no git needed). On the GitHub page, click Code → Download ZIP, or download the v0.1.0 release.
Option B — clone it with git:
git clone https://github.com/knowledgedb-io/ai-application-building-framework.git my-ai-team
cd my-ai-team
python3 scripts/init_db.pyThen run claude in the folder and follow the README.
Requirements
- Python 3, version 3.8 or newer
- Claude Code, with a Claude subscription or API access — billed by Anthropic or your cloud provider, not by us
- git (optional — only needed for the clone route)
- macOS, Linux or Windows
Works with other AI tools?
It is built for Claude Code. Because it is all plain files, tools that read AGENTS.md or GEMINI.md, such as OpenAI Codex and Google’s Antigravity CLI, can use it too. Expect some features to be less complete outside Claude Code; the README explains what differs.
Frequently asked questions
Book a free intro call
Tell us a little about what you want to build and we will set up a short call to walk you through the framework. No charge, no obligation. The call is general information, not professional advice.
Intro calls are intended for businesses and professionals. Any paid work is agreed separately, in writing, before it starts.
Using the contact form? Mention “framework intro call” in your message.
The AI Application Building Framework is an independent project. It is not affiliated with, sponsored by or endorsed by Anthropic, Google or OpenAI. Claude, Claude Code, Antigravity, OpenAI and Codex are trademarks of their respective owners and are used here only to describe compatibility.