Zosma Cowork AI Assistant: Desktop AI Workspace Setup Guide

Key Takeaways

• Zosma Cowork turns your computer into a private AI workspace in under five minutes
• Your data stays on your machine. Anthropic, OpenAI, Google, and Meta cannot see it because the model runs locally
• Local AI costs far less than cloud subscriptions. You only pay for electricity
• Cowork works with Ollama, LM Studio, and any OpenAI-compatible local model

Why a Desktop AI Assistant in 2026

Knowledge workers spend 60% of their time on administrative work. Emails, file sorting, report reformatting, and meeting follow-ups eat the productive hours you were actually hired to do (Asana Anatomy of Work Index, 2025). A desktop AI assistant changes that equation.

Unlike cloud AI tools that require you to describe your problem from a browser window, a desktop agent lives on your machine. It opens your files, reads your documents, and produces finished output without you copy-pasting context between apps.

In 2026, building a local AI assistant is no longer an engineering project. Open-source models running on consumer hardware produce output quality that matches paid APIs for document work, drafting, and analysis. The tooling to run them takes fifteen minutes to set up.

The privacy advantage

When you send a document to ChatGPT, Claude, or Gemini, it lives on a third-party server. Depending on the provider's current policy, it may be logged, reviewed, or used for training. For professional work with client documents, unreleased code, or financial data, that is a real risk.

Zosma Cowork runs entirely on your hardware. Your documents never leave your machine. Anthropic, OpenAI, Google, Meta, or Chinese AI labs cannot see your data because there is no network call. You can share almost anything with zero risk.

Cost is just electricity

Cloud AI subscriptions stack up fast. A team of five might pay $50 to $200 per month per tool across ChatGPT, Claude, Zapier, and custom GPT integrations. A small business easily spends $2,000 to $5,000 per month on AI tool subscriptions and API costs (VRLA Tech, 2026).

With a local model, the monthly cost is zero. You only pay for electricity. The hardware you buy once pays for itself in months if you are replacing cloud subscriptions.

System Requirements

Cowork runs on most modern computers. The exact requirements depend on which model you choose.

Minimum setup

• macOS: Apple Silicon (M1 or later), 16 GB RAM
• Windows: Any GPU with 8 GB VRAM (RTX 3060, RTX 4060, or AMD equivalent), 16 GB RAM
• Linux: GPU with 8 GB VRAM, 16 GB RAM

Recommended setup

• macOS: M2 Pro or M3 Pro with 24 GB to 36 GB unified memory
• Windows/Linux: GPU with 12 GB+ VRAM, 32 GB RAM

Model sizing guide

Step 1: Download and Install Cowork

Head to the Zosma Cowork download page and grab the installer for your operating system. The package is a single executable — no Docker, no CLI, no manual model downloads.

Windows

Download the installer, run it, and Cowork launches automatically. The first run wizard guides you through model selection.

macOS

Download the .dmg, drag Cowork to your Applications folder, and open it. macOS may prompt you to allow local network access and microphone permissions if you plan to use voice features.

Linux

The app ships as an AppImage. Make it executable and run it. Cowork handles the rest.

Step 2: Choose Your Local Model

Cowork ships with built-in model support through Ollama. On first launch, you will see a model selection screen.

Quick recommendations

• Llama 4 8B — Best all-rounder for 8 GB systems. Handles drafting, summarization, and document Q&A well.
• Qwen 3 14B — Strong across general tasks. Apache 2.0 licensed, good for RAG workflows.
• Phi-4 14B — Leads on reasoning and multimodal tasks. Good choice if your work involves complex analysis.

The differences between these models sit inside the noise floor for most daily tasks. Pick the largest one that fits your hardware and move on.

Model download time

Models download automatically on first launch. A 14B quantized model is roughly 8 GB and downloads in minutes on a standard broadband connection. You only download each model once.

Step 3: Connect Your Documents

This is what makes Cowork useful. A model that answers general questions is a local ChatGPT. What makes it an assistant is giving it context about your specific work.

Built-in document support

Cowork reads PDFs, Word documents, CSV files, Markdown, and plain text out of the box. Point it at a folder and it indexes everything locally.

The vector embeddings run on your machine. No document content leaves your disk.

First task: document search

Start with something simple. Point Cowork at a folder of client documents and ask it to summarize the key terms. The model reads the files, extracts the relevant information, and returns a clean summary.

This is a real job that usually takes 30 to 45 minutes manually. Cowork does it in seconds.

Step 4: Set Up Your First Automation

Cowork includes pre-built workflows for common tasks. You do not need to write code or configure complex pipelines.

Common workflows

Document processing — Extract key data from a batch of PDFs or invoices. Cowork reads each file, pulls out the relevant fields, and produces a structured output.

Report generation — Describe the report you need. Cowork connects to your documents, analyzes the data, and produces a formatted output.

Research summaries — Drop a folder of research papers or reports and ask questions about the content. Cowork retrieves the relevant context and answers locally.

Custom workflows

Cowork supports custom skills for edge cases. Define what you need in plain language and the agent executes it. New capabilities get added over time without requiring configuration.

How Cowork Compares to DIY Stacks

You could build the same thing manually with Ollama, Open WebUI, and Chroma. The result works, but it requires hours of setup and ongoing maintenance.

Cowork packages all of that into a single application. Model management, document indexing, RAG, and the chat interface come pre-configured. You are operational in under five minutes instead of several hours.

What happens behind the scenes

When we tested Cowork against a manual Ollama + Open WebUI + Chroma stack, the core components were identical. Cowork just handled the integration work automatically. Model selection, embedding configuration, vector database setup — all done for you.

The quality is the same because the models are the same. The difference is setup time and ongoing maintenance.

Data Privacy in Practice

Let's be specific about what "private" means with Cowork.

Your data never leaves your machine

All inference runs locally. All document processing runs locally. All embeddings are stored in your local database. There is no Cowork cloud storing your files, conversations, or document contents.

Contrast with cloud providers

When you use ChatGPT, Claude, Gemini, or any cloud AI service, your data hits their servers. Even with strict privacy policies, you are trusting a third party with sensitive information. With Cowork, there is no third party. The model runs on your hardware and the answer comes from your hardware.

Compliance

For businesses handling sensitive data, running AI on-premise eliminates the need for data processing agreements with AI vendors. There is no third-party processor to add to privacy notices. Your audit trail stays clean.

Getting Started: Your First Five Minutes

Here is the fastest path from zero to useful.

1. Download Cowork and install it
2. Launch the app and select a model that fits your hardware
3. Wait for the model to download
4. Point Cowork at a folder of documents
5. Ask it a question about those documents

That is it. The first correct, private answer generated entirely on your hardware is when the real value becomes obvious.

Frequently Asked Questions

Can I use Cowork completely offline?

Yes. After the initial model download, Cowork works entirely offline. No internet connection is required for inference, document processing, or any core feature.

What models does Cowork support?

Cowork works with any model available through Ollama, including Llama, Qwen, Phi, Mistral, and Hermes. You can also connect custom models through the OpenAI-compatible API endpoint.

Do I need a dedicated GPU?

A GPU makes a real difference. With 8 GB VRAM, you get 20 to 40 tokens per second, which feels like real-time conversation. CPU-only runs at 3 to 8 tokens per second, which is usable for occasional queries but frustrating for daily use. If you plan to use Cowork daily, a GPU with at least 8 GB VRAM is the practical minimum.

How is Cowork different from other desktop AI tools?

Cowork is a complete workspace, not just a chat interface. It includes document management, RAG, pre-built workflows, and a desktop-native experience. Unlike tools that are thin wrappers around cloud APIs, Cowork runs everything locally on your machine.

Can Cowork handle multiple users on the same machine?

Cowork is designed as a personal workspace. Each user gets their own instance with separate model configurations, document indexes, and conversation history. For team-wide deployment with multiple users, check out OpenZosma.

How Zosma Helps with Your AI Workspace

Zosma's Cowork transforms your desktop into a private AI workspace by handling the entire setup process automatically.

• Instant setup: Download and start working in under five minutes. No model downloads to manage, no Docker containers, no API keys.
• Document AI: Point Cowork at any folder and ask questions. Built-in RAG processes PDFs, Word documents, CSVs, and more entirely on your machine.
• Pre-built workflows: Ready-to-run automations for document processing, report generation, and research summaries.
• Complete privacy: Anthropic, OpenAI, Google, and Meta cannot see your data. The model runs on your PC and your context stays there.

Cowork is a free desktop harness that keeps your data private. Pay as little as ₹500/month for the AI brain. Your context stays on your PC.

Get Zosma Cowork