GeoLens

Get a dataset's schema and source trust metadata. Returns columns (name/type/role), geometry type, CRS/SRID, feature count, spatial extent, and the safe `source_origin`. `source_health` is healthy, missing, inaccessible, or unknown: inaccessible means GeoLens could not determine whether the source still exists, while unknown means it was never probed or cannot be probed. `source_health_detail` is a fixed GeoLens reason code, not provider text. `last_checked_at` records the latest probe attempt; `last_refreshed_at` records only the latest successful committed refresh. `source_freshness` is advisory; overdue means two declared update intervals elapsed without a successful refresh. Raw provider URLs, origin pointers, and credentials are excluded. Call this before writing spatial questions so you know both the available columns and whether the source may be stale. Args: dataset_id: Dataset id (e.g. from search_datasets).

Get GeoJSON features for a dataset (bounded). Returns an OGC-API FeatureCollection. Raster datasets have no features and will error. Results are capped by `limit`; page with `offset`. Only the arguments listed below are supported; anything else is rejected. Args: dataset_id: Dataset id. limit: Max features (1-200, default 10). offset: Feature offset for paging. bbox: Optional bounding box filter "minx,miny,maxx,maxy" (WGS84).

Get a summary of one saved map: name, description, visibility, view state (center, zoom, bearing, pitch), basemap, terrain configuration and layer list. Each layer has `id`, `dataset_id`, `dataset_name`, `dataset_table_name`, `dataset_geometry_type`, `display_name`, `visible`, `layer_type` and `filter`. Styling, popups, labels, column info and sample values are not included. `dashboard_widget_count` is the number of dashboard widgets (0 for none). Args: map_id: Map id (e.g. from list_maps).

List saved maps (read-only metadata: id, name, visibility, layer count). Args: search: Optional text filter on map name/description. limit: Max results (1-200, default 50). offset: Result offset for paging.

Run one read-only SQL SELECT against accessible datasets. Executes through the server's hardened SQL sandbox: a single SELECT over `data.*` tables, an allowlisted function set (aggregates, math, string, date, JSON), and a strict server-side budget (a 5 s statement timeout, a repetition cap on self-joins, and `row_limit` rows, at most 1000). Returns `{columns, rows, row_count, truncated}` with rows as lists in column order. Geometry columns come back as hex EWKB; select `ST_AsText(geom)` or `ST_AsGeoJSON(geom)` for readable geometry. `ST_AsGeoJSON(geom, 6)` rounds coordinates to six decimals, which suits display but can make a small polygon invalid. PostGIS functions allowed: ST_Area, ST_AsGeoJSON, ST_AsText, ST_Buffer, ST_Centroid, ST_Collect, ST_Contains, ST_Distance, ST_DWithin, ST_GeometryType, ST_GeomFromGeoJSON, ST_GeomFromText, ST_Intersects, ST_IsValid, ST_Length, ST_MakeEnvelope, ST_MakePoint, ST_NPoints, ST_Point, ST_SetSRID, ST_Transform, ST_Union, ST_Within, ST_X, ST_Y. ST_Collect and ST_Union take two geometries (no aggregate form). The `&&` bounding-box operator works, as do `::geography` / `::geometry` and `geography(geom)`. Requires credentials with AI-chat permission — anonymous configurations cannot use this tool. Workflow: `search_datasets` to find a dataset, then `get_dataset_schema` for its `table_name` and columns, then reference it as `data.<table_name>` in the SQL and list the same `table_name` in `restrict_tables`. Every table the query touches must be listed there; the scope can only narrow what your credentials already see. Writes, other schemas, and unlisted functions are rejected with a short reason that names the refused function. Args: sql: A single SELECT statement over `data.*` tables. restrict_tables: Table names (no `data.` prefix) the query may touch. Required and non-empty. row_limit: Max rows to return (1-1000, default 100).

Search the GeoLens catalog for datasets by free text. Matches title, description, and keywords (semantic ranking is used automatically when the instance has it enabled). Returns a compact FeatureCollection (`numberMatched`, `numberReturned`, `features`) where each feature is a trimmed dataset record: `id`, `bbox`, and properties `title`, `description` (cut at 300 characters), `record_type`, `geometry_type`, `feature_count`, `crs`, `keywords` and `updated`. Use the feature `id` as the dataset_id for the other tools; `get_dataset_schema` has the full detail. Each feature's properties also include `source_origin` (upload, postgis, service, stac, created, or null) and `source_freshness` (fresh, due, overdue, or unknown). Freshness is advisory: `due` means one declared update interval has elapsed, while `overdue` means two have elapsed; neither proves that the content is wrong. Search does not make a detail request per result. When the catalog summary has no health/check/refresh value, `source_health`, `source_health_detail`, `last_checked_at`, and `last_refreshed_at` are null. Call `get_dataset_schema` for populated trust metadata. Raw provider URLs and credentials are never included in these source-state fields. Args: query: Search text. limit: Max results (1-200, default 10). offset: Result offset for paging.