Open-source alternatives to Gemini

Proton Lumo provides conversational chat, document analysis, web search, image generation, projects, and Proton integrations. Its published mobile clients are GPL-licensed, but the complete hosted service and AI infrastructure are not a fully open-source stack.

Pricing: Hosted privacy-focused service; complete self-hosting is not available.

• Strong privacy focus and encrypted chat storage
• Document analysis, web search, image generation, and Proton Drive integration
• Open-source mobile clients
• Hosted infrastructure is not fully open source
• Cannot self-host the complete service
• Less model flexibility than self-hosted platforms

Open WebUI provides a unified interface for Ollama, OpenAI-compatible APIs, Anthropic, and other providers, with document RAG, web search, tools, multimodal input, user management, and extensions. Current code uses the Open WebUI License with branding restrictions.

Pricing: Free to self-host; current license includes branding restrictions beyond traditional permissive licenses.

• Excellent self-hosting and Ollama integration
• Local and cloud models, RAG, web search, tools, and multimodal support
• Multi-user support and large community
• Current license includes branding restrictions
• Current releases are not purely BSD-3-Clause
• Team deployments require infrastructure

LobeChat, now part of the LobeHub platform, combines multi-model chat, AI agents, knowledge bases, plugins, tools, and orchestration. Its community license is based on Apache-2.0 but adds restrictions on commercial derivative works.

Pricing: Self-hostable source-available community edition; license contains additional commercial restrictions.

• Multiple providers, agents, knowledge bases, and tools
• Model switching and self-hosting
• Strong interface and agent workflows
• Not standard Apache-2.0
• Commercial restrictions apply
• More complex than a simple Claude replacement

AnythingLLM is an MIT-licensed local-first and self-hostable AI workspace with local or cloud models, document ingestion, vector search, agents, multi-user workspaces, and a developer API.

Pricing: Free and open source under MIT with desktop and self-hosted deployments.

• MIT licensed and no mandatory account
• Desktop, local models, document chat, RAG, agents, and API
• Multi-user workspaces
• More focused on document workflows than general Claude use
• Local inference requires suitable hardware
• Setup is more involved than a hosted chatbot

Cherry Studio is an AGPL-3.0 Community Edition desktop AI assistant for Windows, macOS, and Linux. It combines model selection, chat, knowledge bases, agents, MCP servers, documents, model comparison, and local conversation storage.

Pricing: Free and open source Community Edition; external model API costs may apply.

• Large provider and model selection
• BYOK, local models, agents, MCP, knowledge bases, and document analysis
• Cross-platform local conversation storage
• Desktop-focused
• Many models require external APIs
• Large feature set can be more complex than Claude

LibreChat is an MIT-licensed self-hostable chat interface supporting Anthropic, OpenAI, Gemini, Azure, local models, custom agents, files, tools, web search, code execution, model comparison, and multi-user access.

Pricing: Free and open source under MIT; API provider and infrastructure costs may apply.

• MIT licensed and multi-provider
• Self-hosted chat, agents, files, tools, and model comparison
• Familiar interface with multi-user support
• Requires API configuration
• Local models need separate infrastructure
• Administration is more involved than using Claude directly

Jan is an Apache-2.0 desktop AI assistant designed to run models such as Llama, Qwen, Mistral, and DeepSeek locally while also supporting cloud providers.

Pricing: Free and open source under Apache-2.0; cloud model API costs may apply.

• Runs locally and offline
• Apache-2.0 with local model management
• Multiple cloud providers, desktop workflow, and strong privacy
• Performance depends on hardware
• Large models require significant RAM or GPU resources
• Cloud capabilities still require third-party APIs

Fullmoon is an MIT-licensed local AI assistant that runs compatible language models entirely on iPhone, iPad, Mac, and Vision Pro using Apple's hardware acceleration.

Pricing: Free and open source under MIT; compatible local models run on supported Apple hardware.

• Runs completely locally and offline
• Private conversations and MIT license
• Optimized for iPhone, iPad, Mac, and Apple Silicon
• Limited to Apple platforms
• Smaller models than cloud assistants
• Less capable for complex reasoning and smaller ecosystem

Zola is an Apache-2.0 web chat interface for OpenAI, Anthropic, Gemini, Mistral, OpenRouter, and local Ollama models, with self-hosting and bring-your-own-key support.

Pricing: Free and open source under Apache-2.0; provider and infrastructure costs may apply.

• Apache-2.0 and self-hosted
• Multi-model, Claude, OpenRouter, and Ollama support
• Clean web interface with BYOK
• Primarily an interface
• Smaller feature set than larger platforms
• Advanced features may require additional services