LibreChat

Chatbox answers "I want a good client for myself." LibreChat answers "we need a shared AI environment for several people." 42,017 stars, MIT-licensed.

It is a web application deployed on a server, with an interface close enough to ChatGPT's that nobody needs training. An administrator configures models and keys once, then issues accounts to team members who share the same API allowance — no key per person, no subscription per person. For a small team, a studio, a lab or even a household, the economics are compelling.

Model connectivity is broad: mainstream cloud providers, any OpenAI-compatible endpoint, and local Ollama. Different conversations can use different models in the same interface, and the same question can be sent to several models at once for comparison.

The feature set is complete: full-text search across conversation history, custom assistants with preset roles and parameters, document upload for Q&A, a code interpreter, image generation, and MCP support for tool calling. The admin side shows usage statistics and allows per-user permissions and quotas.

Costs to accept:

• You must deploy it yourself. Official Docker Compose configuration and thorough documentation exist, but a server and basic operational skills are required. For solo use without the hassle, Chatbox is more direct.
• Model costs are separate. The software is free; cloud models you connect bill per token, invoiced by the provider. Running entirely on local Ollama is free, bounded by your hardware.
• Security is your responsibility. As a multi-user web service, HTTPS, login policy and whether to expose it publicly all need deliberate configuration.

Who it suits: organizations that need one AI entry point for a small group without buying commercial seats per person, and without conversation history landing on someone else's servers.