picaultj · picaultj · Jul 21, 2026 · Jul 21, 2026
diff --git a/.github/huggingface/space_readme_header.md b/.github/huggingface/space_readme_header.md
diff --git a/.github/workflows/sync-hf-space.yml b/.github/workflows/sync-hf-space.yml
diff --git a/.github/workflows/test.yml b/.github/workflows/test.yml
@@ -23,8 +23,8 @@ jobs:
         with:
           enable-cache: true
 
-      - name: Sync dependencies (base + dev + gradio)
-        run: uv sync --extra gradio   # so the Gradio GUI is covered by the suite
+      - name: Sync dependencies (base + dev)
+        run: uv sync
 
       - name: Tests (pytest)
         run: uv run pytest -q

diff --git a/AGENTS.md b/AGENTS.md
@@ -17,19 +17,15 @@ save/load a run). Python 3.13, managed with **uv**.
 
 ```bash
 uv sync                        # base install (Anthropic, OpenAI, LiteLLM + NiceGUI)
-uv sync --extra gradio         # + the Gradio GUI
 uv sync --extra bertopic       # + BERTopic topic backend (heavy)
 
 cp .env.example .env           # add provider key(s); loaded automatically
 uv run conference-analyzer     # NiceGUI GUI on http://localhost:6868
-uv run conflens-gradio         # Gradio GUI on http://localhost:7860
 uv run conference-analyzer --clear-cache   # wipe the on-disk cache
 ```
 
-**Two front-ends, kept in parity:** `app.py` (NiceGUI) and `gradio_app.py`
-(Gradio, deployed to Hugging Face). All non-UI logic lives in `view.py` +
-`pipeline.py`; when you change a user-facing feature, update **both** GUIs. See
-[CLAUDE.md](CLAUDE.md).
+The GUI (`app.py`) is a thin **NiceGUI** presentation layer; all non-UI logic
+lives in `view.py` + `pipeline.py` — put behaviour changes there.
 
 Docker: `docker compose up --build` (see the README).
 
@@ -66,8 +62,8 @@ regexes over heavyweight parsers, small focused modules).
 
 | Area | Module |
 |------|--------|
-| GUIs (feature parity) | `app.py` (NiceGUI), `gradio_app.py` (Gradio) |
-| Shared view logic | `view.py` (filter/sort/highlight/compute_view/exports) |
+| UI (NiceGUI) | `app.py` |
+| View logic | `view.py` (filter/sort/highlight/compute_view/exports) |
 | Exports | `pptx_export.py`, `bibtex.py` |
 | Orchestration | `pipeline.py` (`AnalysisConfig`, `run_analysis`) |
 | Sources | `sources.py` (registry + `make_source`; `IJCAISource`, `OpenReviewSource`, `PSCCSource`, `DBLPSource`), `scraper.py` (`AnthologyScraper` — also serves EMNLP/NAACL) |

diff --git a/CLAUDE.md b/CLAUDE.md
@@ -25,16 +25,9 @@ repository or its GitHub artifacts:
 feature (the `anthropic` provider and `claude-*` model ids in `llm.py`) and
 stays — this policy is about tooling attribution, not the app's providers.
 
-## Two front-ends (persistent — keep both maintained)
+## Front-end (NiceGUI)
 
-The app ships **two GUIs with feature parity**:
-
-- **NiceGUI** — `conflens/app.py` (console script `conference-analyzer`); run
-  locally / in Docker.
-- **Gradio** — `conflens/gradio_app.py` (console script `conflens-gradio`); this
-  is the front-end deployed to Hugging Face Spaces.
-
-All non-UI logic (filtering, sorting, highlighting, the computed view, exports)
-lives in **`conflens/view.py`** and the pipeline — put behaviour changes there so
-both GUIs inherit them. When you add or change a user-facing feature, **update
-both `app.py` and `gradio_app.py`** so they stay in parity.
+The app has a single **NiceGUI** GUI in `conflens/app.py` (console script
+`conference-analyzer`). All non-UI logic (filtering, sorting, highlighting, the
+computed view, exports) lives in **`conflens/view.py`** and the pipeline — put
+behaviour changes there so the UI stays a thin presentation layer.
diff --git a/README.md b/README.md
@@ -12,9 +12,7 @@
 [![uv](https://img.shields.io/badge/managed%20by-uv-DE5FE9.svg?logo=uv&logoColor=white)](https://docs.astral.sh/uv/)
 [![Built with NiceGUI](https://img.shields.io/badge/UI-NiceGUI-2b6cb0.svg)](https://nicegui.io)
 
-A desktop-style web app — with **two interchangeable front-ends**,
-[NiceGUI](https://nicegui.io) (local/Docker) and [Gradio](https://gradio.app)
-(deployed to Hugging Face Spaces) — that:
+A desktop-style web app (built with [NiceGUI](https://nicegui.io)) that:
 
 1. **Browses** papers from a chosen **source** and retrieves their abstracts:
    - the [ACL Anthology](https://aclanthology.org) (default; e.g. `acl-2026`),
@@ -44,7 +42,7 @@ A desktop-style web app — with **two interchangeable front-ends**,
 
 ## Contents
 
-- [Quick start](#quick-start) · [Run with Docker](#run-with-docker) · [Deploy to Hugging Face Spaces](#deploy-to-hugging-face-spaces) · [LLM providers](#llm-providers)
+- [Quick start](#quick-start) · [Run with Docker](#run-with-docker) · [LLM providers](#llm-providers)
 - [Architecture](#architecture) · [How it works](#how-it-works)
 - [Configuration](#configuration-in-the-ui) · [Features](#features)
 - [Caching](#caching) · [Cost](#cost) · [BERTopic](#optional-bertopic)
@@ -57,15 +55,10 @@ Requires Python 3.13+ and [uv](https://docs.astral.sh/uv/).
 ```bash
 uv sync                                   # Claude, OpenAI, LiteLLM work out of the box
 cp .env.example .env                       # then fill in your provider key(s)
-uv run conference-analyzer                # NiceGUI GUI → http://localhost:6868
+uv run conference-analyzer                # or: uv run python run.py
 ```
 
-Prefer **Gradio** (same features; the front-end deployed to Hugging Face)?
-
-```bash
-uv sync --extra gradio
-uv run conflens-gradio                     # Gradio GUI → http://localhost:7860
-```
+Then open <http://localhost:6868>.
 
 Keys are read from `.env` (loaded automatically) or the process environment;
 you can also paste a key into the app's **API key** field at runtime.
@@ -99,44 +92,6 @@ docker run --rm -p 6868:6868 --env-file .env \
 To also build the optional BERTopic backend into the image:
 `docker build --build-arg EXTRAS="--extra bertopic" -t conference-analyzer .`
 
-### Deploy to Hugging Face Spaces
-
-The **Gradio** front-end is deployed to Hugging Face as a **Gradio SDK Space**
-(free CPU tier). HF installs [`requirements.txt`](requirements.txt) and runs
-[`space_app.py`](space_app.py), which serves `conflens.gradio_app`. A GitHub
-Action mirrors `main` to the Space on every push, so it stays in sync.
-
-**One-time setup:**
-
-1. **Create the Space** — on <https://huggingface.co/new-space>, pick **Gradio →
-   Blank**, **CPU basic** hardware (this app is CPU-only; ZeroGPU errors with
-   *"No @spaces.GPU function detected"*), and note its owner + name (e.g.
-   `your-user/conflens`).
-2. **Give GitHub a token** — create a Hugging Face access token with **write**
-   scope (Settings → Access Tokens) and add it to this GitHub repo as a secret
-   named **`HF_TOKEN`** (Settings → Secrets and variables → Actions).
-3. **Point the action at your Space** *(optional)* — the workflow defaults to
-   owner `picault` and space `conflens` (the HF username differs from the GitHub
-   owner). If your Space differs, set repo **variables** `HF_USERNAME` and
-   `HF_SPACE`.
-4. **Add your API keys as *Space* secrets** — in the Space's *Settings →
-   Variables and secrets*, add whatever your provider needs (e.g.
-   `ANTHROPIC_API_KEY`, or `OPENAI_API_KEY` / `OPENAI_BASE_URL`; see
-   [`.env.example`](.env.example)). The app reads them as environment variables.
-
-That's it — push to `main` (or run the **Sync to Hugging Face Space** workflow
-manually) and the Space rebuilds. The Space metadata (`sdk: gradio`,
-`app_file: space_app.py`, …) lives in
-[`.github/huggingface/space_readme_header.md`](.github/huggingface/space_readme_header.md);
-the action prepends it to the README it pushes, so the GitHub README stays clean.
-
-> The Space's filesystem is ephemeral, so the on-disk cache resets on rebuild.
-> For persistent caching, attach Hugging Face **persistent storage** and set
-> `HOME` (or a `--cache-dir`) to point at its `/data` mount.
-
-> The Docker image above still runs the **NiceGUI** app (`0.0.0.0:6868`) for
-> local/self-hosted use; the two front-ends share all analysis logic.
-
 ### LLM providers
 
 The classifier and topic engine work with three providers, chosen in the UI: