| # smolnalysis App |
|
|
| This app uses `gr.Server` with a custom HTML chat frontend. |
|
|
| ## Local Run |
|
|
| From the repository root: |
|
|
| ```bash |
| uv venv |
| uv sync |
| npm install |
| npm run build:openui-renderer |
| uv run python app/app.py |
| ``` |
|
|
| Open [http://127.0.0.1:7860/](http://127.0.0.1:7860/). |
|
|
| ## Runtime Shape |
|
|
| - `app.py` owns the Gradio server, `/` frontend route, and `/api/chat` backend route. |
| - Backend chat goes through `SmolnalysisMoE`. |
| - The frontend is a dynamic server-rendered page with chat state, training-set example prompts, loading states, and adapter trace display. |
| - Auto mode uses the learned router, not the old hardcoded CKAN-first heuristic. |
| - Router outputs select `general_agent`, `ckan_retrieval`, or `openui_translator`. |
| - CKAN-style retrieval output is passed with the user question to the `openui_translator` adapter. |
| - OpenUI-Lang responses are rendered server-side into the custom chat UI. |
|
|
| The old Blocks chat UI, fullscreen React chat frontend, and streaming route are removed. |
|
|
| ## Frontend Assets |
|
|
| The app no longer needs a bundled chat frontend. Only the embedded OpenUI renderer remains: |
|
|
| - `app/frontend/openui-renderer.jsx` |
| - `app/static/openui-renderer.js` |
|
|
| Rebuild it after editing the renderer source: |
|
|
| ```bash |
| npm run build:openui-renderer |
| ``` |
|
|
| ## Useful Tests |
|
|
| ```bash |
| uv run python -m unittest tests.test_smolnalysis_model_wrapper |
| uv run python -m unittest tests.test_openui_adapter_demo |
| ``` |
|
|
