# dbt Information Schema

Available in v2

The dbt Information Schema is a contracted interface into the metadata for all of the resources in your dbt project. It takes the form of [Parquet](https://parquet.apache.org/) artifacts, which are more performant than JSON artifacts. For example, a project with a `manifest.json` and `catalog.json` that total \~70 MB has an Information Schema of only \~5 MB. It also includes intermediate views that you can query with `dbt show`. You can also query the Information Schema from SQL you write in your checks.

When you use the [`--generate-info-schema`](#generating-the-information-schema) flag, dbt writes the Information Schema to `target/info_schema/` in a versioned subdirectory (for example, `target/info_schema/v1/`) as standard Parquet files. The metadata available in the schema grows with each step: parsing produces basic metadata, compiling adds column types and column-level lineage (with `--static-analysis strict`), and running or building populates runtime results.

For the full list of tables and their descriptions, refer to the [Information Schema tables](../../reference/info-schema.md).

Rather than parsing `manifest.json`, you can query your project metadata using SQL. Use [`dbt show --inline`](#querying-with-dbt-show) to run SQL queries against the Information Schema directly from the CLI, or [`dbt show --info`](#querying-with-dbt-show) to query a view by name. You can also point any Parquet-compatible tool (for example, Pandas or Polars) directly at the files.

## Generating the Information Schema

Use the `--generate-info-schema` flag with `dbt build`, `dbt run`, `dbt compile`, or `dbt parse`.

* To populate column types and column-level lineage in `dbt.node_columns` and `dbt.column_lineage`, combine [`dbt build`](../../reference/commands/build.md), [`dbt run`](../../reference/commands/run.md), or [`dbt compile`](../../reference/commands/compile.md) with [`--static-analysis strict`](./about-static-analysis.md). Without it, `dbt.node_columns` and `dbt.column_lineage` contain no column types and no lineage.

  ```shell
  dbt build --generate-info-schema --static-analysis strict
  ```

* For [`dbt parse`](../../reference/commands/parse.md), the Information Schema contains no column types, no lineage, and no runtime results, because `dbt parse` doesn't connect to your warehouse.

### Overriding the output directory

Use `--info-schema-dir` to write the Information Schema to a custom directory. The versioned subdirectory (`v1/`) is still appended under whatever directory you set.

```shell
dbt build --generate-info-schema --info-schema-dir /tmp/my_schema
# writes to /tmp/my_schema/v1/
```

## Querying the Information Schema

### Querying with `dbt show`

Use `dbt show --info <view>` to query a specific Information Schema view directly from the CLI:

```shell
dbt show --info models
dbt show --info models --format json --limit 20
```

`--info <view>` is equivalent to `--inline "select * from {{ info_schema('<view>') }}"` and queries `target/info_schema/`. It does not connect to your warehouse.

You can also use `--inline` SQL that calls `{{ info_schema() }}` directly:

```shell
dbt show --inline "select name from {{ info_schema('models') }} order by name"
```

### Querying with external tools

You can query the Parquet files with any Parquet-compatible tool.

**Parquet-compatible tools:** Point your tool directly at the Parquet files in `target/info_schema/v1/`. For example, with pandas:

```python
import pandas as pd
models = pd.read_parquet("target/info_schema/v1/dbt.models.parquet")
```

## Related docs

* [Information Schema tables](../../reference/info-schema.md)
* [`dbt build`](../../reference/commands/build.md)
* [`dbt run`](../../reference/commands/run.md)
* [`dbt compile`](../../reference/commands/compile.md)
* [`dbt parse`](../../reference/commands/parse.md)