Web: Streamlit Exports — vLLM, llama.cpp, MLX

The Planner page's bottom row turns any computed plan into a deploy-ready runtime config. One click per backend — vLLM, llama.cpp, or MLX — emits the matching CLI / JSON artefact.

What you'll see

Narrative beats:

1. Planner page ready with a plan already computed; we scroll down to the Export Configuration row.
2. Export row in view: three buttons side-by-side.
3. Export as vLLM clicked — JSON payload with --model, --max-model-len, --max-num-seqs rendered in a code block.
4. Export as llama.cpp clicked — CLI arg string (-m, -c, -ngl) for the same plan.
5. Export as MLX clicked — Apple Silicon deploy config serialised as JSON.

All three emit from the same hwledger-ffi export functions that power the CLI's hwledger export subcommand, so the configs are identical regardless of client.

Journey not yet recorded.

Run the journey recorder to capture interactions:

./apps/macos/HwLedgerUITests/scripts/run-journeys.sh

Reproduce

cd apps/streamlit/journeys
bun install
bash scripts/record-all.sh
STREAMLIT_URL=http://127.0.0.1:8599 bunx playwright test specs/exports.spec.ts