Maneek21/Deft

What it solves

Deft addresses the fragmentation of modern work by unifying chat, task management, knowledge documentation, and AI agent interactions into a single, self-hostable workspace. It prevents the common issue where AI agents operate in silos, disconnected from the actual context of team discussions and project history.

How it works

Deft functions as a centralized platform where human-to-human communication (chat) and human-to-AI collaboration coexist. It uses a "work record" approach where conversations, tasks, and knowledge are linked. Users can configure AI providers (such as OpenAI, Anthropic, or local Ollama instances) to interact with this workspace via native tools or the Model Context Protocol (MCP). The system includes built-in governance, allowing teams to set approval policies for agent actions, ensuring that AI-proposed tasks or modifications are reviewed before they are finalized.

Who it’s for

This project is designed for technical teams and organizations that prioritize data sovereignty and want to integrate AI agents directly into their internal workflows without relying on third-party SaaS platforms to manage their sensitive project data.

Highlights

• Unified Context: Links chat, tasks, and knowledge so agents have access to relevant team history.
• Governed Agent Actions: Features approval tiers and activity receipts to ensure AI-driven changes are transparent and controlled.
• Provider-Neutral: Supports multiple AI backends, including local models via Ollama or hosted APIs like Claude and ChatGPT.
• Self-Hostable: Open-source (AGPL-3.0) architecture that allows teams to own their data and infrastructure.
• Extensible: Supports custom modules and apps to add domain-specific functionality to the workspace.