Pandesh 🐼 — Local AI Personal Assistant for Windows

An open-source, local-first AI assistant inspired by the idea of a Jarvis-style desktop companion — powered by Ollama, with explicit controls for real-world actions.

Looking for a personal AI assistant for Windows, an Ollama desktop assistant, or a self-hosted alternative to cloud-only AI assistants? Pandesh aims to combine local AI conversations, reusable tools, tasks, reminders, and governed automation in one approachable Panda interface.

⬇ Download Windows Alpha (Build 144) • 📖 Start here • 🏗️ Architecture • 🗺️ Roadmap • 🐞 Report an issue

Actual Pandesh desktop interface from v0.1.2 Build 143.

What is Pandesh?

Pandesh is an experimental local AI agent and personal assistant for Windows. Rather than being only a chatbot, its goal is to help with everyday work and show what was actually accomplished. It runs local LLMs through Ollama and can integrate additional tools when installed and permitted.

Experience	What you can expect
Chat with local AI	Use models installed on your own Windows computer
Tasks and results	Organize work and review outcomes rather than accept unsupported "Done!" claims
Human-approved actions	Review sensitive/mutating operations before execution
Reminders and briefings	Scheduled assistant workflows, depending on configured services and permissions
Models and integrations	Ollama first; other compatible providers and agent integrations are optional
Developer workflows	VS Code / coding-related capabilities in the evolving shared platform

Important: The public release is an alpha. Not all workflows are reliable or supported on every machine. Gmail, Calendar, external models and web access require their own permissions. Local-first does not mean all configured integrations are offline.

How to get Pandesh running

1. Download and start Ollama for Windows.

2. Open PowerShell, then choose a model:

   ollama pull qwen2.5-coder:3b
   ollama list
   

3. Install Pandesh v0.1.3 Alpha (Build 144) or extract the portable ZIP and run its launcher.

4. Start with "Hi Pandesh. What can you do?", then try a specific read-only request.

For verified download checksums, permissions, troubleshooting and the full installation flow, use the getting-started guide.

Building from Source

Pandesh Core is fully buildable independently:

# 1. Clone the repository
git clone https://github.com/siddharthakki/Pandesh.git
cd Pandesh

# 2. Install dependencies
npm ci

# 3. Build Core bundle
npm run bundle:core

# 4. Run tests
npm test

Explore the interface

See all six screenshots • View release gallery

Build 143 fixes New Conversation navigation and introduces direct Tasks / Results tabs.

Architecture and principles

Architecture diagram ➔ • Privacy and safety ➔

• Local-first AI: choose a reachable local Ollama model rather than requiring a paid model subscription for basic chat.
• Real tool evidence: explicit tool invocation, observed outcome, and recovery status.
• Governed execution: only approved consequential actions; never indiscriminately stop unrelated host processes.
• Durable continuity: task, memory and conversation state should survive restarts.
• Small, purposeful UI: Panda front and center; show panels when a task requires them.

Roadmap

Our immediate focus is reliable conversations, correct model detection, approvals, task results, and useful daily workflows, not a feature checklist.

Read the short, prioritized roadmap.

Source Code & Licensing

Pandesh Core is published under the MIT License.

MIT LICENSE • CONTRIBUTING • SECURITY

Trust and contribution safety

Outside developers are welcome to fork and propose improvements. Pull requests are validated in GitHub-hosted CI runners.

Security policy • Maintainer protection checklist • How contributions work

Help build Pandesh

• Start contributing: fork the project, improve public guides or report real Windows/Ollama behavior.
• Good first issue #2: make first-run Ollama instructions clearer and reproducible.
• Good first issue #3: document actual model compatibility, including what fails.
• Support and community: know where to ask questions, report bugs or help test.

Community and credits

Found a bug or got Pandesh working with your favorite local model? Open an issue with the app build and reproducible behavior—never raw private logs.

Built with and/or integrated with independent upstream projects including Ollama, Pi, Electron, OpenCode and Hermes Agent. See acknowledgements and license notes. No affiliation with the fictional J.A.R.V.I.S. character, Marvel, or those upstream maintainers is implied.

Downloads: Pandesh_Releases • Public Source: Pandesh.