# Build the index

> Create and validate the Milvus paper collection.

Canonical: https://keta1930.github.io/icml2026-mcp/en/docs/guides/rebuild-index



Run preflight to see the record count and conservative input limit without calling DashScope:

```bash
uv run icml-papers build
```

Build the collection only after choosing a hard input limit:

```bash
uv run icml-papers build --execute --max-input-tokens 10000000
```

The command inserts complete batches, validates the Milvus schema, indexes, row count, paper IDs, vector dimensions, and one hybrid query, then writes `.data/index/current.json`.

If an ordinary interruption leaves a valid checkpoint, rerun with `--resume`. A provider-usage limit violation must be inspected and started again without `--resume`.