PyGPT
by Marcin Szczygliński (independent open-source project)
An open-source desktop AI assistant that runs every major model — cloud or local — from one window
PyGPT is a free, open-source desktop AI assistant for Windows, macOS and Linux that puts almost every major model behind a single interface. It supports OpenAI, Anthropic, Google, xAI, DeepSeek, Perplexity and Mistral APIs alongside local models via Ollama or LM Studio, adding agents, vision, voice, RAG over local files and a Python interpreter for people who want their data to stay on their own machine.
PyGPT is a free, open-source desktop AI assistant for Windows, macOS and Linux, created and maintained by Polish developer Marcin Szczygliński since April 2023. Its distinguishing claim is breadth of model coverage from a single local window: it drives OpenAI (GPT-5, GPT-4, o1, o3), Anthropic Claude, Google Gemini, xAI Grok, DeepSeek, Perplexity and Mistral through their APIs, and runs models entirely locally through Ollama or LM Studio using GGUF weights, so a user can switch between a private local model and a frontier cloud model without changing tools. The application ships twelve distinct modes of operation — chat, vision, agents, research, computer use, completions, image and video generation among them — plus speech synthesis and recognition via Whisper, Azure, Google and ElevenLabs, a real-time Python code interpreter, filesystem operations, web search across DuckDuckGo, Google and Bing, retrieval-augmented generation over local documents through LlamaIndex, Model Context Protocol support, and a plugin system for extending any of it. There is no subscription: local inference is unlimited and free, and cloud calls are billed by the user's own API keys at provider rates with no markup, which makes the total cost of ownership legible in a way subscription assistants are not. The GitHub repository carries roughly 1,880 stars and 334 forks with active releases — version 2.7.12 landed in February 2026 adding xAI video generation and eighteen new interface languages. It occupies the niche between LM Studio, which is polished but not open source, and general chat clients, trading a denser interface for far more capability and full data ownership.
A developer, data scientist or privacy-constrained professional who wants one desktop client across every model vendor, with the option to keep inference and documents entirely on their own hardware.
Unlimited local-model use and unmarked-up cloud API access from a single open-source app, with no subscription and no data leaving the machine unless the user chooses a cloud model.
At a Glance
- Category
- Developer Tools
- Pricing
- Free
- Target Market
- Enterprise Developers, Data Scientists, AI/ML Engineers, Security-conscious professionals, Researchers
- Deployment
- Open-source, Self-hosted, Edge-first
- Founded
- 2023
- Headquarters
- Poland
Key Features
- ✓Multi-provider model support
Drives OpenAI, Anthropic, Google, xAI, DeepSeek, Perplexity, Mistral and Ollama from one window, so switching vendors costs nothing
- ✓Local-first inference
Runs GGUF models through Ollama or LM Studio entirely offline, keeping prompts and documents on the user's own hardware
- ✓Twelve operating modes
Chat, vision, agents, research, computer use, completions and image or video generation are separate first-class modes rather than bolted-on toggles
- ✓RAG over local files
Indexes local document collections through LlamaIndex so the assistant answers from your own material without a hosted vector service
- ✓Real-time Python interpreter and filesystem tools
Executes generated Python and reads or writes files directly, turning the assistant into an actual local automation surface
- ✓Full voice pipeline
Speech recognition and synthesis via Whisper, Azure, Google and ElevenLabs supports hands-free operation and accessibility use
- ✓MCP and plugin system
Model Context Protocol support plus a plugin architecture let users attach web search, custom tools and internal services
Capabilities
Use Cases
- •Offline model evaluation
An engineer benchmarks a local Llama variant against GPT-5 and Claude on the same prompts inside one client
- •Confidential document analysis
A professional indexes client files locally and queries them with an on-device model so nothing reaches a third-party API
- •Desktop task automation
A developer chains the Python interpreter and filesystem tools so the assistant edits files and runs scripts directly
- •Cost-controlled AI usage
A solo consultant drops per-seat subscriptions and pays only metered provider API rates through personal keys
- •Accessible voice-driven computing
A user operates the assistant hands-free through Whisper speech recognition and text-to-speech across supported interface languages
Ideal For
Best For
- ✓Comparing outputs across OpenAI, Anthropic, Google, xAI and local models without maintaining a separate client for each
- ✓Privacy-sensitive work where documents and prompts must never leave the local machine, using Ollama or LM Studio backends
- ✓RAG over a local document collection via LlamaIndex without standing up a server or vector database service
- ✓Developers who want a Python interpreter, filesystem access and MCP tools attached to the model in one desktop app
- ✓Avoiding per-seat AI subscriptions by paying provider API rates directly through your own keys
Not Ideal For
- ✗Non-technical users who want a zero-configuration assistant — PyGPT expects familiarity with API keys, model quantisation and local runtimes
- ✗Teams needing shared workspaces, seat management, audit logs or vendor compliance attestations, none of which a single-maintainer desktop app provides
- ✗Mobile-first workflows, since there is no iOS or Android client and the product is desktop-only
- ✗Laptops without a dedicated GPU, where 30B-plus local models are impractically slow
Integrations
Deployment
Market Analysis
Pros
- ✓Unmatched breadth of supported models and providers for a desktop client, cloud and local alike
- ✓Fully open source and free, with cloud API costs passed through at provider rates rather than marked up
- ✓Actively maintained with a steady release cadence — v2.7.12 shipped February 2026 with xAI video generation and eighteen new interface languages
- ✓Local-only operation means prompts and indexed documents never leave the machine, which is decisive for privacy-constrained work
Cons
- ✗Effectively a one-person project, which is a real bus-factor and support risk for anything business-critical
- ✗Setup demands technical fluency — API key configuration and an understanding of model quantisation are assumed, not guided
- ✗Desktop-only with no mobile client and no shared team workspace, seat management or audit logging
- ✗Large local models (30B and above) are unusable on laptops without a dedicated GPU, so the local-first promise is hardware-gated
- ✗The dense, mode-heavy interface is harder to approach than browser-based assistants, and carries no vendor security attestations for procurement
Pricing
Open source
$0
- ✓Full application, no feature gating
- ✓Unlimited local model use via Ollama or LM Studio
- ✓Windows, macOS, Linux, AppImage and pip install
- ✓Cloud models billed directly by the provider on your own API key
The application itself is free and open source with no paid tier, no seat licence and no feature gating — every mode ships in the public build. Running local models through Ollama or LM Studio costs nothing beyond your own hardware and electricity. Cloud models are billed by the provider against your own API key at their list rates, with PyGPT passing costs through without markup, so total spend is whatever OpenAI, Anthropic, Google or xAI charge for the tokens consumed. The practical cost is therefore hardware rather than software: a machine without a dedicated GPU cannot run 30B-plus models at usable speed, which pushes serious local users toward a GPU purchase. The project site states an MIT licence, though GitHub's own metadata classifies the repository licence as 'Other' rather than a recognised SPDX identifier, so anyone with a strict legal review should read the licence file directly.
Security & Compliance
Connect
Sources
This page was written from 5 sources, 4 on domains other than pygpt.net.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Confident AI
Hosted LLM evaluation, observability and red teaming built on the open-source DeepEval framework
Respan
Observability, evals and an LLM gateway for AI agents in one control plane
Meta Muse Code
Meta's terminal coding agent for large repositories, with persistent background agents and the most aggressive token pricing in the category
Niteshift
The full-stack cloud for coding agents — real environments, verified pull requests