finds.dev← search

// the find

szczyglis-dev/py-gpt

★ 1,851 · Python · NOASSERTION · updated Feb 2026

Desktop AI Assistant powered by GPT-5, GPT-4, o1, o3, Gemini, Claude, Ollama, DeepSeek, Perplexity, Grok, Bielik, chat, vision, voice, RAG, image and video generation, agents, tools, MCP, plugins, speech synthesis and recognition, web search, memory, presets, assistants,and more. Linux, Windows, Mac

PyGPT is a desktop GUI app (PySide6/Qt) that wraps virtually every major LLM provider — OpenAI, Anthropic, Google, xAI, Ollama, DeepSeek, and more — into a single interface with chat, RAG, agents, image/video generation, voice, and a plugin system. It's for power users who want a local ChatGPT-style frontend without locking into one provider or running everything in the browser. One maintainer, 1,800 stars, actively shipped.

The provider breadth is real, not just a config flag — it ships native SDKs for Anthropic, Google GenAI, and xAI alongside the OpenAI SDK, so you're not faking it with a compatibility shim. LlamaIndex RAG is properly integrated, supporting 15+ file types and a dozen web sources including GitHub repos and Google Drive, with selectable vector backends. The accessibility work is non-trivial: keyboard shortcuts, voice control, and TTS narration of on-screen actions are first-class features, not afterthoughts. MCP support and the visual node-based agent builder are things most competing apps don't have yet.

The kitchen-sink scope is the main risk: 11 modes, 30+ plugins, 6 vector stores, 4 TTS services — any one of these could be half-baked and you wouldn't know until you tried it. The troubleshooting section alone runs several hundred lines, covering xcb plugin failures, GLIBC version minimums, FFmpeg requirements on Windows, VC++ redistributables, and AppArmor profile editing; that's not a sign of a smooth install experience. It's a one-maintainer project, which means provider API changes (and these APIs change constantly) create a backlog that only one person can clear. The config model stores everything in local JSON files, which is fine until you have 40+ settings screens worth of state to migrate across versions.

View on GitHub → Homepage ↗

// want more like this?

We dig through GitHub every week and send a few repos picked for what you actually care about — each with an honest take like this one.

Get finds in your inbox → Search again →