Incognide

Explore the unknown and build the future.

Download for Linux, macOS, and Windows

---

Incognide unifies chat, code, documents, web browsing, and media into a tileable workspace with intelligent context and composable automations.

Built for seamless workflows, Incognide eliminates distractions and context switching. A path-based organization keeps your work structured naturally, while auto-saving workspaces let you resume any project exactly where you left off. No more juggling desktops, drowning in browser tabs, or hunting for scattered files.

Highlights

• Write and run code, use terminals, build reusable workflows and tools that chain together natural language and templateable code through jinja execution templates.
• Browse the web, read and annotate PDFs, view 3D STL models, analyze data and create dashboards, compile LaTeX.
• Edit DOCX, XLSX, PPTX, MAPX.
• Arrange chats, editors, PDFs, browsers, terminals, 3D viewers as your work evolves — each tab maintains independent state.
• Manage agents, have them run on scheduled jobs, edit team context, integrate with MCP Servers, approve or reject suggested memories, prune and evolve knowledge graphs.
• Schedule automated memory extraction, knowledge graph evolution, and context compression.
• Fine-tune your own image and text models using curated data from your conversations and memories.
• Built-in Pomodoro timer with programmable schedules and break enforcement.

---

Setup

1. Install

Download the installer for your platform from enpisi.com/incognide, run it, and launch Incognide. Linux, macOS, and Windows are supported.

2. First launch — setup wizard

On first launch the wizard walks you through five steps:

1. Welcome — intro screen.
2. Preferences — theme (dark/light) and the data directory for teams, models, and configs (default ~/.npcsh/incognide).
3. Defaults — default web search engine, default terminal shell (auto-detected per platform: system/zsh/bash on mac/linux, powershell on Windows), and activity tracking opt-in (local only, used for next-action predictions).
4. Usage path — pick one of:
   • No AI — workspace only (files, code, browsers, terminals, docs, maps). No model calls.
   • Cloud AI — OpenAI / Anthropic / Gemini / etc. via API keys.
   • Local AI — Ollama / LM Studio / llama.cpp / oMLX running on your machine.
5. Cloud keys (cloud-ai path) or Local models (local-ai path). See step 3 below for details.

The wizard writes preferences to localStorage (theme, fonts, shortcuts, UI defaults) and settings to ~/.incogniderc .

3. Connect a model provider

Local providers

The setup wizard (Local Models step) and the in-app Model Manager both probe these endpoints and binary locations:

Provider	Server port	Binary / app checks
Ollama	11434	ollama in PATH; /Applications/Ollama.app on macOS
LM Studio	1234	lms in PATH; /Applications/LM Studio.app on macOS
llama.cpp	8080	llama-server, llama-cli, or koboldcpp in PATH
oMLX	8000	omlx in PATH; /Applications/oMLX.app (macOS only)

Each tile shows one of three states: Running (server responds), Installed (not running) (binary or app found but port is dead), or Not found. The Model Manager has Start/Stop buttons when the binary is installed:

• Ollama: open -a Ollama (macOS) or ollama serve (Linux). Stop via pkill -f 'ollama serve' (or kill from the Windows system tray).
• LM Studio: lms server start / lms server stop if the lms CLI is present; otherwise the app launches and you start the server from the Developer tab.
• llama.cpp: requires a model path to start, so Incognide prints the exact command (llama-server -m <model.gguf> --port 8080) for you to run in a terminal. Stop via pkill -f llama-server.
• oMLX: open -a oMLX launches the menu-bar app; stop/start the server from its menu-bar icon.

GGUF / GGML model files can be loaded directly without a server, but only if llama.cpp or koboldcpp is installed — the Model Manager's GGUF tab depends on that engine and shows "Not found" otherwise.

Cloud providers

Add API keys on the Cloud keys step of the wizard, or later in Settings → Global Settings or Team Management → API keys. Keys are stored in ~/.npcshrc as export <PROVIDER>_API_KEY=... and are read by npcsh and by the in-app provider clients.

4. Local setup for fine-tuning

Inference with LLMs is routed through the bundled backend (including calls to locally running models with llama.cpp, omlx, etc), but users can also fine-tune models within incognide. To accomplish this, you need to specify a Python virtual environment with the heavy packages (torch+ transformers etc.). For such calls, Incognide shells out to the specified venv instead of including these dependencies in the bundled backend to keep the packaged executable small.

1. Open Team Management (Users icon in the right sidebar or the settings screen).

2. Go to the Python Env tab.

3. Choose an environment for the current workspace:

   • Detected: list of venvs/pyenv/conda/uv environments found under the workspace, homedir, and common paths.
   • Create new venv: creates <workspace>/.venv (or the name you pick) with the system Python.
   • Custom path: point at any existing Python interpreter.

4. With an environment selected, click Install packages and pick one of the bundles:

   • PyTorch (CPU) — torch torchvision torchaudio
   • PyTorch (CUDA) — same plus the CUDA index URL for GPU builds
   • Diffusers (Image Gen) — diffusers transformers accelerate safetensors
   • Transformers (LLM) — transformers accelerate safetensors sentencepiece
   • Whisper (Speech) — openai-whisper

   Or install any specific package by name.

5. Data directory

Incognide stores teams, NPCs, jinxes, memories, knowledge graphs, and model configs under the data directory you picked in step 2 (default ~/.incognide). Changing it in Settings → Global Settings → Default Directory updates ~/.incogniderc's INCOGNIDE_DATA_DIRECTORY.

6. Troubleshooting

• Backend unhealthy indicator in the status bar — right-click the npcpy icon in the status bar for Restart / View Logs. Logs live in ~/Library/Logs/Incognide/ (macOS), ~/.config/Incognide/logs/ (Linux), or %APPDATA%\Incognide\logs\ (Windows).
• Tutorial didn't highlight anything — the tutorial opens the Help pane before it starts so the workspace highlight has a target. If it runs before any pane is open you'll see an un-highlighted step; re-run it from Settings → Replay Tutorial.

---

Table of Contents

• Office & Productivity
• Development
• 3D & Media
• Research & Knowledge Management
• Model Training & Fine-tuning
• AI Chat & Agents
• Image Tools (Vixynt)
• Focus & Productivity
• Settings & Customization
• Keyboard Shortcuts
• Setup
• Development Setup
• Community
• License

---

Office & Productivity

Document Editing

Create and edit Office documents directly in Incognide without needing external applications or cloud services.

Word Documents (DOCX) - Full rich text editing with formatting, tables, and images:

Spreadsheets (XLSX & CSV) - Edit data with formula support and cell formatting:

Presentations (PPTX) - View and edit PowerPoint presentations:

PDF Management

Read, annotate, and analyze PDF documents with AI assistance.

Highlight & Annotate - Mark up PDFs with highlights that persist across sessions:

PDF Library - Browse and organize your PDF collection:

File Management

Folder Explorer - Drag any folder from the sidebar into a pane to open a dedicated file browser:

Disk Usage Analyzer - Visualize what's taking up space on your drives:

Web Browsing

Browse the web alongside your documents and chat with AI about what you're viewing.

Integrated Browser - No need to switch to a separate browser window:

Tileable configuration - Browse while viewing PDF with a terminal open and a chat window:

---

Development

Code Editing

Write code with syntax highlighting, run scripts, and compile documents.

Code Editor and script execution - Syntax highlighting for Python, JavaScript, TypeScript, and more. Vim, Emacs, and Nano keybinding modes with a toggleable cheat sheet. Run Python scripts directly and see output inline:

LaTeX Compilation - Write and compile LaTeX documents with PDF generation launching a pane.

AI-Assisted Development

File Analysis - Select files and ask AI to analyze, explain, or refactor code:

Database Tools

Connect to databases, explore schemas, and run queries.

Schema Viewer and SQL Querying - Investigate your database structure, write queries manually or with natural language.

Manipulate table results and plot data directly for quick analysis:

Git Integration

Git Manager - Stage, commit, and manage branches without leaving Incognide:

---

STL Viewer

View 3D models directly in Incognide.

Features:

• Orbit, pan, and zoom with mouse controls
• Wireframe, axes, and grid toggles in the pane header
• Quick axis views (X, Y, Z) for front/side/top perspectives
• Adjustable mesh color and opacity
• Screenshot export of the current viewport
• Model info: triangle count, vertex count, and bounding box dimensions

---

Research & Knowledge Management

Data Analysis

Build dashboards and visualizations from your data.

Data Dashboard - Composable widgets for analytics and visualization:

Knowledge Graphs

Explore connections between concepts and entities.

Graph Explorer - Navigate and edit knowledge graphs built from your conversations:

Memory & Context

Memory Management - Review, edit, and organize what your agents remember:

Agent Memories - See what context agents have learned from conversations:

---

Model Training & Fine-tuning

Train custom AI models using your own data—no coding required. Incognide provides first-class UI for curating training data from your conversations and memories, then fine-tuning models locally.

Image Model Training

Fine-tune image generation models with your own images using LoRA training.

Training Interface - Select reference images, configure training parameters, and monitor progress:

Features:

• Drag-and-drop image selection for training datasets
• Automatic captioning with AI assistance
• LoRA training with configurable rank, learning rate, and epochs
• Real-time training progress and loss visualization
• Export trained adapters for use in generation

Text Model Training

Fine-tune language models using curated subsets of your AI interactions and memories.

Data Curation - Select specific conversations, memories, and agent interactions to use as training data:

Features:

• Browse and filter your conversation history
• Select individual messages or entire conversations for training
• Export agent memories as training examples
• Preview and edit training pairs before export
• Quality scoring to identify high-value training examples

Training Pipeline - Fine-tune models locally with LoRA/QLoRA:

Features:

• Support for Llama, Mistral, Qwen, and other popular architectures
• LoRA and QLoRA for efficient training on consumer hardware
• Configurable hyperparameters with sensible defaults
• Training metrics and loss curves
• Merge adapters or use them standalone

Why Fine-tune?

• Personalization - Train models that understand your terminology, preferences, and domain
• Privacy - Keep sensitive data local; train on private conversations without uploading anywhere
• Specialization - Create expert models for specific tasks from your best interactions
• Cost savings - Run fine-tuned smaller models instead of expensive API calls

---

AI Chat & Agents

Conversations

Chat Interface - Clean, focused conversations with AI:

Model Selection - Choose from local models (Ollama, llama.cpp) or cloud providers (OpenAI, Anthropic, Gemini):

Aggregate Conversations - Select multiple conversations and combine them for context:

File Attachments - Include files directly in your conversations:

Tool Use & MCP

Agentic Tool Use - Enable agents to use tools from MCP Servers or local jinxes:

MCP Server Management - Connect to Model Context Protocol servers:

Browser Automation

Agents can control browser panes directly - clicking elements, typing into inputs, extracting page content, and taking screenshots.

Agent Browser Control - Agents can open browser panes and interact with web pages:

Click & Type - Agents click elements by text or CSS selector and fill in forms:

Available browser actions for agents:

• browser_click - Click elements by selector or text content
• browser_type - Type into input fields with optional form submission
• get_browser_content - Extract page text content for context
• browser_screenshot - Capture page screenshots
• browser_eval - Execute JavaScript in page context

Terminal & File Control

Agents can also interact with terminal panes and files directly within Incognide.

Terminal Commands - Agents can run commands in terminal panes:

File Operations - Agents can open, read, and edit files in editor panes:

Agent Management

NPC Editor - Create and customize AI personas with specific directives, models, and capabilities:

Agent History - Track what your agents have done:

Team Management - Manage global and project-specific context for your agent team:

Jinx Workflows

jinxes are reusable automation templates that combine natural language prompts with code execution.

Jinx Editor - Create and edit Jinx workflows:

Jinx Execution - Run jinxes with custom parameters:

SQL Jinx - Create jinxes that query databases:

Agents in SQL - Utilize agents and NPC personas within your SQL models for advanced analyses with native graph computations afforded by SQL engines.

Scheduled Tasks

Cron Jobs - Schedule jinxes and agents to run automatically:

Schedule memory extraction, knowledge graph evolution (sleep/dream), and context compression as automated jobs. Configure guidance context to focus extraction on specific topics. Schedule directly from the Memory Manager or Knowledge Graph Editor.

---

Image Tools (Vixynt)

Photo Browser

Browse and organize your image collection:

AI Image Generation

Generate images using AI with reference images for style and composition:

DarkRoom

Simple photo editing with cropping, filters, and adjustments:

---

Focus & Productivity

Pomodoro Timer

A built-in Pomodoro timer in the top bar helps maintain focus with enforced break periods.

Features:

• Configurable work and break durations (right-click to configure)
• Full-screen break overlay that locks the UI during breaks
• Persists across window refreshes — active timers survive reload
• Programmable schedule: set specific days of the week and times for sessions to auto-start
• Visual status: red during work, green during break

Backend Health Monitor

The status bar shows Python backend health with a colored indicator. Right-click to restart if the backend becomes unresponsive.

---

Settings & Customization

The Settings panel provides comprehensive configuration across multiple tabs: Global Settings, Theme, Keyboard Shortcuts, Model Management, Voice/TTS, Custom Providers, Passwords, Python Environment, and Account.

Cloud Sync & Account

Sign In - Create an account to sync your conversations, settings, and workspace state across devices.

End-to-End Encryption - All synced data is encrypted with your passphrase before leaving your device. Your passphrase never leaves your machine.

Multi-Device - Work on your desktop, pick up on your laptop. Your conversations, memories, and workspace layout stay in sync.

macOS Permissions

On macOS, manage permissions for camera, microphone, and screen capture from Settings. Required for voice input and screenshot features.

Theme

Light and dark modes with full color customization:

Sidebar

Collapsible Sidebar - Manage files, conversations, and navigation:

Model Management

Configure, download, and manage models from multiple sources:

HuggingFace Integration - Download GGUF models directly:

Custom Providers - Connect to custom OpenAI-compatible APIs:

Voice & Audio

Configure TTS and STT settings:

Python Environment

Configure Python environments per project:

Keyboard Shortcuts

View and customize keyboard shortcuts:

---

Development Setup

Incognide is an Electron + React frontend with a Python Flask backend powered by npcpy. The UI uses npcts, a React component library.

Prerequisites

• npcpy - Core Python library
• npcsh - Shell interface (starts the backend)
• npcts - React component library (installed via npm)
• Node.js 16+ and npm
• Ollama (optional, for local models)

Setup

Option 1: Manual setup

git clone https://github.com/npc-worldwide/incognide.git
cd incognide
npm install

Option 2: Via npcsh (installs to ~/.npcsh/incognide)

npcsh> /incognidev

Running

python incognide_serve.py   # Backend
npm run dev                   # Frontend (Vite)
npm start                     # Electron

---

Community

• Discord: Join us
• Issues & Bugs: GitHub Issues
• Discussions: GitHub Discussions
• NPC Ecosystem: npcpy | npcsh | npcts

---

License

Incognide is licensed under the MIT License. See the LICENSE file for details.