---
title: Montaigne OpenWebUI
emoji: 🪶
colorFrom: yellow
colorTo: gray
sdk: docker
app_port: 8080
pinned: false
---

# Montaigne OpenWebUI

**This Space is the production user interface of the Montaigne chatbot.**

`https://ask-montaigne.pages.dev` redirects here; the admin dashboard (backend
user management) lives at `https://ask-montaigne.pages.dev/admin.html` and uses
its own cookie/CSRF auth.

The Space connects to the Montaigne FastAPI backend through an
OpenAI-compatible API exposed at `/v1`.

## Auth Model

OpenWebUI runs with authentication **disabled** (guest-only, `features.auth=false`):

- Anonymous visitors are mapped to per-IP backend guest accounts with a
  **5-message limit** per IP.
- At the limit, the UI patch shows a "Create free account" modal
  (`POST /auth/claim-guest`), which creates a backend account and sets its
  password at claim time.
- The account token rides on `/v1` requests as `variables.montaigne_token`;
  conversation continuity is keyed by `X-OpenWebUI-Chat-Id` / browser
  fingerprint.
- There are **no OpenWebUI accounts, no billing, no conversation sharing, and
  no OAuth buttons** in the UI. The backend `/auth/profile` and `/billing/*`
  endpoints stay dormant.

## UI Policy (Minimal Consumer Interface)

Standard users see a stripped-down chat-only interface.  The model selector,
Controls panel, advanced parameters, voice input/output, Settings menu,
and all power-user surfaces are hidden.  Only admins retain full access.

This is enforced in two layers:

1. **Persistent config** — the container entrypoint syncs
   `config/ui_policy.json` into OpenWebUI's database on every boot,
   ensuring permissions and feature flags always match the repo state.
2. **Runtime DOM patch** — `config/minimal_ui_patch.js` is copied into the
   built OpenWebUI app and injected into `index.html` at container startup.
   It hides the model selector, settings entry, archived chats, keyboard
   shortcuts, and other power-user affordances from the public UI, and
   implements the guest-limit claim modal.

### Updating the policy

Edit `config/ui_policy.json` and redeploy.  The entrypoint handles first
boot (bootstrap `config.json`) and subsequent boots (update the `config`
row in `webui.db`).

Key fields:

| Key | Purpose |
|-----|---------|
| `default_models` | Model pre-selected for every new chat |
| `default_pinned_models` | Model pinned in the selector |
| `permissions.chat.*` | Per-user chat capabilities |
| `permissions.features.*` | Per-user feature access |
| `permissions.workspace.*` | Per-user workspace access |
| `features.*` | Global OpenWebUI feature toggles |

### Minimal UI patch

`config/minimal_ui_patch.js` is intentionally lightweight: it uses a small
MutationObserver-based DOM patch instead of rebuilding OpenWebUI from source.
That keeps the Space repo small, avoids binary asset issues on Hugging Face,
and still removes the public-facing controls we do not want to expose.

If you upgrade OpenWebUI and the selectors stop matching, adjust only
`config/minimal_ui_patch.js` and redeploy.

## Conversation Starters

Configured in `config/prompt_suggestions.json`.  These Montaigne-style
prompts appear on the empty chat screen.

## Environment Variables

| Variable | Description |
|----------|-------------|
| `OPENAI_API_KEY` | Your API key for the backend LLM provider |
| `OPENAI_API_BASE` | Backend URL: `https://tgrumb-michel.hf.space/v1` |
| `DEFAULT_PROMPT_SUGGESTIONS` | Optional override for the repo-managed prompt suggestions |