NPC-Worldwide/incognide

Explore the unknown, build the future, own your data.

What it solves

Incognide is a unified workspace designed to eliminate context switching and distractions by integrating chat, code editing, document management, web browsing, and media viewing into a single tileable interface. It streamlines complex workflows by combining productivity tools with AI agents that can interact directly with the workspace's components.

How it works

The application provides a flexible, tiled layout where users can arrange various panes (browsers, terminals, editors, PDF viewers) that maintain their own state. It connects to AI models via cloud APIs (OpenAI, Anthropic, Gemini) or local providers (Ollama, LM Studio, llama.cpp, oMLX).

Beyond simple chat, it implements an agentic system where AI personas (NPCs) can perform actions like controlling the browser (clicking, typing, screenshotting), executing terminal commands, and editing files. It also features "jinxes," which are reusable automation templates that chain natural language prompts with code execution using Jinja templates. For knowledge management, it uses memory extraction and knowledge graphs to evolve agent context over time.

Who it’s for

  • Developers and Researchers who need to manage code, documentation, and AI interactions in one place.
  • Power Users looking for a highly customizable, automated workspace with integrated AI agents.
  • AI Enthusiasts who want to fine-tune local LLMs using their own conversation data without writing code.

Highlights

  • Agentic Control: AI agents can directly interact with the browser, terminal, and file system.
  • Integrated Fine-tuning: Built-in UI to curate conversation data and fine-tune models locally using LoRA/QLoRA.
  • Unified Workspace: Tileable panes for DOCX, XLSX, PPTX, PDFs, 3D STL models, and LaTeX compilation.
  • MCP Integration: Supports the Model Context Protocol for extending agent capabilities via external servers.
  • Knowledge Management: Automated memory extraction, context compression, and knowledge graph evolution.