| # Flask Demo Interface |
|
|
| **Source files:** `app_flask.py`, `templates/index.html` |
|
|
| --- |
|
|
| ## Overview |
|
|
| `app_flask.py` is the demo application. It runs a Flask server on port 7860 and serves a single-page application with 10 tabs. The frontend uses Chart.js for reward/metric plots and vis-network for the cascade dependency graph visualization. There is no Gradio in the stack — `app.py` is a legacy file; `app_flask.py` is the production demo. |
|
|
| --- |
|
|
| ## Architecture |
|
|
| Flask handles all HTTP routes. The frontend is a full HTML/JS application in `templates/index.html`. Tabs communicate with the backend via `/api/*` endpoints that return JSON. The vis-network cascade animation calls `core/cascade_utils.py:animate_cascade()` to get frame-by-frame cascade data. |
|
|
| --- |
|
|
| ## Module-level singletons |
|
|
| These are instantiated at startup: |
|
|
| | Variable | Type | Purpose | |
| |----------|------|---------| |
| | `agent` | `LifeStackAgent` | GRPO model + Groq fallback | |
| | `memory` | `LifeStackMemory` | ChromaDB trajectory + feedback store | |
| | `MODEL_REGISTRY` | dict | Maps model label → HF repo ID for v1–v4 | |
| | `_load_grpo_model` | function | Lazy-loads a specific LoRA adapter | |
|
|
| `MODEL_REGISTRY` allows the demo to switch between v1, v3, and v4 adapters on the fly via a dropdown in the UI. |
|
|
| --- |
|
|
| ## Key API routes |
|
|
| | Route | Purpose | |
| |-------|---------| |
| | `GET /` | Serves `templates/index.html` | |
| | `POST /api/run_demo` | Runs the agent on a preset conflict, returns step-by-step cascade frames and action JSON | |
| | `POST /api/custom_run` | Runs the agent on a user-typed situation | |
| | `POST /api/submit_feedback` | Stores `OutcomeFeedback` in ChromaDB memory | |
| | `GET /api/cascade_animate` | Returns frame list from `animate_cascade()` for the vis-network graph | |
| | `GET /api/memory_stats` | Returns `LifeStackMemory.get_stats()` | |
| | `GET /api/health` | Basic health check | |
|
|
| --- |
|
|
| ## Running |
|
|
| ```bash |
| python app_flask.py # serves on http://localhost:7860 |
| ``` |
|
|
| In Docker, `start.sh` runs this as the foreground process. HuggingFace Spaces health-checks port 7860 — Flask must stay alive for the Space to remain healthy. |
|
|
| --- |
|
|
| ## Related files |
|
|
| - `templates/index.html` — full frontend (Chart.js, vis-network) |
| - `core/cascade_utils.py` — `animate_cascade()` drives the dependency graph visualization |
| - `agent/agent.py` — `LifeStackAgent` with GRPO model |
| - `agent/memory.py` — `LifeStackMemory` |
| - `start.sh` — starts this as foreground service on port 7860 |
|
|
