Proposal: keep LibreChat as the private agent builder and put a separate public chat UI in front of its beta Agents API. A server-side gateway would hold credentials and manage guest sessions. This preserves the agent backend, not LibreChat's UI; the integration is untested.

A public chat frontend without replacing the LibreChat agent

Reviewed September 29, 2026. This is an architecture proposal, not a deployed integration.

Documented foundation

The LibreChat Agents API exposes agents through streaming Chat Completions and Open Responses endpoints. The Chat Completions request identifies the saved agent through its model field. The API is explicitly beta.

Proposed separation

Keep the authenticated LibreChat application private for configuring and testing the agent. Give visitors a separate public chat page that talks only to an application-owned gateway. The gateway calls LibreChat using credentials held on the server.

Proposed request path:

Visitor -> public chat page -> guest-session gateway -> LibreChat Agents API -> model and approved tools

The visitor does not need a LibreChat account or a login screen. The gateway, however, needs a secure way to distinguish guests and associate each request with the correct conversation. This is application work, not a hidden LibreChat guest-mode switch.

For the page, assistant-ui's custom runtime is a candidate: it provides React chat components and adapters for an application-owned backend. The adapter would translate the gateway's stream into UI messages. This does not establish drop-in compatibility with every LibreChat feature.

Tradeoff

This approach retains LibreChat's agent configuration and execution backend without maintaining a fork of its login flow. It does not retain LibreChat's complete frontend. A basic text-chat integration should not be assumed to reproduce native file handling, tool approvals, citations, persistent threads, or resumable streams.

I would pin the LibreChat version and first test two concurrent guest sessions, a multi-turn conversation, one actual agent-tool call, stream cancellation, and denial of requests that attempt to address another guest's conversation. None of these tests has been performed in this exchange.