Projects

Published

AI-First Company Framework

People and AI. Organized together.

How does a company work when people and AI work together? The AI-First Company Framework describes a provider-independent organizational model that connects tasks, knowledge, decisions, and responsibility.

It grew out of exploring how to build Yamarti. As an openly published reference model, it is intended for people who want to build companies with AI or develop existing organizations further.

Structuring collaboration

What capabilities does a company need, and how do people, AI agents, and software contribute? The framework describes a shared structure for tasks, responsibilities, and collaboration.

Preserving company knowledge

Knowledge, experience, and decisions should remain accessible to the company over time. The model describes how people and AI can build on them, with traceable sources and defined access rights.

Autonomy with clear boundaries

AI agents can work independently within explicitly granted authority. The framework separates technical capabilities from actual permissions and describes how actions remain auditable, limited, and interruptible.

Keeping technology replaceable

Models, providers, and tools continue to evolve. The framework therefore separates the capabilities a company needs from their technical implementation. Switching should remain possible while preserving knowledge, requirements, and responsibilities.

Explore the framework

Architecture, reference design, and technical requirements form three layers that build on one another. You can start with a specific question. The repository also includes a starter prompt for exploring the framework together with an AI assistant.

Explore the framework on GitHub

Version 2.0 · Freely available under CC BY 4.0 · Practical validation is still pending.

Support our work

In development

Yamarti Hub

Your data. Your rules. Any AI.

Private AI Data Hub for Mac

Yamarti Hub is a native Mac app that makes information from local files and connected data sources searchable. Your data stays on your Mac, and you decide which content can be accessed and with what permissions.

Features that run directly on your Mac include indexing, full-text search, semantic vector search, graph search and text recognition. The built-in help assistant runs on Apple's local model. You do not need a Yamarti account, your own server or an external AI provider for these features.

Yamarti Hub will be released as an open-source project under a dual-licensing model. Anyone will be able to use the app for personal use at no cost.

Simple setup with built-in help

Add your data sources and get started with sensible defaults. You can adjust additional options when you want more control. The help assistant explains features, helps with problems and suggests changes for you to confirm.

Search your files locally

You can also use Yamarti Hub simply to search your own data, without starting an AI chat. Enter a search query and view matching content directly in the app. Best Match finds results even without the exact wording and shows connections between content.

Control what AI can see

Use Areas, permissions and the Standard, Restricted and Confidential levels to decide which content agents and AI tools may use. Data can be pseudonymized locally when communicating with external models. The Data Log shows which external AI tools accessed information through the Hub and what content was sent.

Connect the models and agents you choose

For your work, use models from OpenAI, Anthropic Claude and Google Gemini directly in the app through their APIs. You can also use models you host locally.

You can keep working with your familiar AI tools from OpenAI, Claude or Gemini. Through MCP, they access data you have shared in your Yamarti Areas. The roles and permissions you set for each Area still apply.

In testing for macOS. Not yet released.