Open source alternatives to Gemini

What Gemini does

The Gemini apps (the web app, the Android and iOS apps and the Mac app) are Google’s AI assistant. They ground answers in Google Search, read uploaded files, offer Canvas, run Deep Research, generate images, video and music, and connect to Google services such as Drive and Photos. Gemini Live adds spoken conversations with video and screen sharing. Gemini can be used without a paid plan; Google AI Plus, Pro and Ultra subscriptions raise limits and unlock more features. Work and school accounts are covered by Google Workspace terms.

Why people look for an alternative

• Training and human review by default. For users aged 18 and over, the Keep Activity setting is on by default. While it is on, Google uses activity to improve its services, including training AI models, and human reviewers read a subset of chats. Turning it off stops future chats being used for training, unless you send feedback.
• Retention. Activity is auto-deleted after 18 months by default (3 or 36 months, or never). Chats picked for human review are kept, unlinked from the account, for up to three years even if you delete your activity. With Keep Activity off, and for temporary chats, conversations are still kept with the account for 72 hours.
• Personal and work accounts differ. For Workspace accounts, chats and uploads are not human-reviewed or used to train models outside the organization without permission.
• Limits that can change. Gemini has five-hour and weekly usage limits, and Google says these may change without notice, including for capacity reasons. Without a paid plan, compute-heavy features such as Deep Research may be unavailable during high demand.
• No self-hosted version. The Gemini apps run only as Google’s service. Google publishes a separate family of open models, Gemma, that can run on your own hardware.

An open source chat app keeps the interface and chat history under your control, but answers are only as private as the model behind it: with a hosted API, prompts still go to that provider. Open-weight models run locally keep prompts on your hardware, but need a capable GPU or plenty of memory and are generally less capable than the top hosted models.

What to compare in an alternative

1. Models. Which local runtimes (Ollama, llama.cpp, a built-in engine) and hosted providers it supports.
2. Where history lives. Browser, desktop or a server database you back up yourself.
3. Tools. The ones you rely on: web search, document chat, image generation, voice, research reports, app connections.
4. Single user or team. A personal desktop app, or a web app with accounts, SSO and roles.
5. Telemetry and license. What the app reports by default, and whether everything is under an OSI-approved license.

Which one to choose

For a self-hosted assistant shared by a team, LibreChat connects to Google (Gemini and Vertex AI) and many other providers and adds agents, MCP, web search and SSO. AnythingLLM focuses on chatting with documents; its MIT-licensed Docker server is the open source edition, and the desktop download has separate terms. For personal use, Jan and Chatbox are open source desktop clients, and Jan can also run models locally. NextChat and Big-AGI are lightweight web clients, Hugging Face’s Chat UI runs HuggingChat, and Khoj is a personal assistant over your own notes and documents. For a chat app that runs models itself, try TextGen on a computer or PocketPal AI on a phone. Other clients can use open models such as Gemma served by Ollama or llama.cpp.