# SPIRA — web demo (Next.js)

Interactive browser demo of the SPIRA architecture. **No backend required** — the
full v2 model runs client-side:

- `src/lib/spira/` — TypeScript port of the Python prototype: same FNV-1a-64 /
  splitmix64 / mulberry32 specs, reservoir regenerated from the same seed, and the
  **hierarchical slow area** (windowed spike volleys, variance-matched top-down
  fusion with the learned gain β). Loads `public/spira_deploy.json` (~29 MB,
  format v2, exported by `python -m spira.export_deploy`).
- `src/components/spira/` — four tabs:
  - **Tokenizer** — live entropy-patch segmentation with surprisal bars
  - **Generate** — closed-loop generation from the hierarchical spiking cortex
  - **One-shot memory** — teach any text in a single pass + recall with
    0–3 corrupted context slots (CA3-style completion)
  - **Brain activity** — spike-raster playback of the recorded reservoir

## Run

```bash
npm install        # or: bun install
npm run dev        # http://localhost:3000
```

Build for production: `npm run build && npm start`.

## Regenerating the deploy artifact

```bash
cd ../python
python3 -m spira.export_deploy --model artifacts/spira_model.npz \
    --out ../web/public/spira_deploy.json --am-cap 100000 --am-min-count 1
```

The spike raster (`public/spira_raster.json`) is exported from
`artifacts/spira_model.npz.raster.npz` recorded during training.

Hippocampal replay (consolidation) is a Python-side capability — see
`../python/README.md`; the browser demo keeps the one-pass teach + recall loop.