Best for: The default interface for local or remote models, for one person or a team.
The most complete self-hosted chat interface, and the de facto default for local models: a polished ChatGPT-like experience, document upload with retrieval, model management, multi-user support and permissions, and connectors for a wide range of backends. It is also comfortable fronting a paid API, which makes it the sensible place to start before committing to local hardware.
AI & LLM
BSD-3-Clause
Beginner
★ 153,674
Replaces: ChatGPT, Claude, Google Gemini
Full review of Open WebUI →
Best for: One interface across many AI providers and models.
The strongest answer for people who use several providers: one interface, one login, many models, with per-user keys and the ability to switch mid-conversation. Useful in exactly the situation most technical users are in — a local model for some work and hosted models for the rest.
AI & LLM
MIT
Intermediate
★ 45,167
Replaces: ChatGPT, Poe, OpenRouter chat
Full review of LibreChat →
Best for: Building AI applications and agent workflows without writing a backend.
Moves up a level from chatting to building: visual workflows for agents and retrieval pipelines, prompt management, and an API to publish what you build. The tool for constructing an application on top of a model rather than having a conversation with one.
AI & LLM
Apache-2.0
Intermediate
★ 157,623
Replaces: OpenAI Assistants, LangChain Cloud, Flowise
Full review of Dify →
Best for: Private code completion in your own editor.
Self-hosted code completion, which is a different product from a chat interface: it plugs into your editor and suggests as you type, with no code leaving your network. The right tool for anyone whose objection to AI coding assistants is where the source code goes.
AI & LLM
Apache-2.0
Intermediate
★ 33,885
Replaces: GitHub Copilot, Codeium, Amazon CodeWhisperer
Full review of Tabby →
Best for: Tracing, evaluating and costing an LLM application.
Observability and evaluation for LLM applications — traces, cost tracking, prompt versions, and scored evaluations. It is the missing piece once you go beyond casual use: without it you cannot tell whether a prompt change improved anything, or what a workflow is actually costing you.
Developer Tools
MIT
Intermediate
★ 35,243
Replaces: LangSmith, Weights & Biases, Helicone
Full review of Langfuse →