What it does

PyGPT is a desktop application written in Python that brings several model providers into one window. You connect your own API credentials for services such as OpenAI, Claude, Gemini, Grok, DeepSeek, Perplexity and Hugging Face, or run local models through Ollama, which the site says can work without external API keys. Beyond chat, it offers work modes for agents, realtime voice, research, image and video generation, and computer use with supported models. Agent workflows can use a primary agent, an orchestrator or a swarm, and can import portable SKILL.md skills and project-specific AGENTS.md instructions. For your own material it indexes files, web pages, images, audio, video and past conversations through LlamaIndex, with vector stores and separate project indexes. Tools cover Python and system commands with a sandbox or Docker, MCP servers, a task scheduler and plugins for services such as GitHub, Slack and Telegram. The app is free under the MIT license and the source is on GitHub. Paid cloud models may bill you separately through their own providers, and local models need suitable hardware.

Capabilities

Chat with models from many cloud providers or local Ollama models
Agent workflows with orchestrator and swarm modes, SKILL.md skills and AGENTS.md instructions
File, web and conversation indexing with LlamaIndex retrieval
MCP connectors, Python and system tools, and a task scheduler
Vision, realtime voice, speech, and image and video generation
Profiles, presets, plugins and themes for a configurable workspace

When to use it

You want one desktop app for several model providers using your own API keys
You want to run local models through Ollama without sending prompts to a cloud service
You want agents, MCP tools and file search on your own machine