Search ALL ~1150 CSDI datasets locally (English or Chinese), ranked. No keyword = browse.

Chinese and English both work — Chinese terms are expanded with Hong Kong usage (廁所 -> 洗手間, 行人路 -> pedestrian, 電動車 -> electric vehicle charging, …). Returns dataset_id + bilingual title + why it matched. Feed dataset_id to load_hk_data().

Download one CSDI dataset as GeoJSON (WGS84) into the workspace. Returns a SUMMARY, not the data itself — never paste the features into the chat, put them on a map with build_map(). The local copy is reused on later calls; pass refresh=True to re-download.

Read a FEW individual features, so a single value can be verified without opening the GeoJSON.

where is either FIELD=value (case-insensitive match on one field) or bare text searched across all fields. Rows are capped at 20 — this tool exists to check a value, not to dump a dataset; use summarise_hk_data for aggregates and build_map to see the layer.

Answer a question about a dataset in TEXT (and a bar chart), without putting features in the chat.

• group_by empty: an overview — feature count, geometry, which fields are worth grouping by.

• group_by="NAME_OF_DISTRICT_COUNCIL_DISTRICT_EN" with agg="count": how many features per group.

• agg="sum" / "avg" with value_field="MEDIUM_IEC62196_no": totals or averages per group. Returns an aligned text table, the same rows as data, and a chart URL served on localhost.

Build an interactive 2D web map (ArcGIS Maps SDK, keyless basemap) from loaded datasets and serve it on localhost. Open the returned URL in your client's browser or preview pane.

Same datasets, rendered in a 3D scene: terrain relief, tilted camera, layers draped on the ground. Keyless (public imagery + Terrain3D elevation). Use it when relief, height or a walk-through view matters — towers, slopes, trails, coastal context.

Geocode an address/place to coordinates, or reverse-geocode a coordinate to an address. Esri public world geocoder — no key required. Reverse mode: pass lon and lat.

Project, buffer or simplify one GeoJSON geometry with Esri's public Geometry service — no key.

operation: "project" (in_sr -> out_sr, e.g. 2326 -> 4326), "buffer" (distance in unit), "simplify" (max_deviation in the input SR's units).

Ground elevation in metres at a coordinate, from Esri's public Terrain3D service — no key. Handy for hiking/relief context (e.g. trail comfort, slope warnings) alongside CSDI layers.

Render a basemap-only PNG (no key) into the workspace — for reports, decks and docs. Give either bbox ("minx,miny,maxx,maxy" in EPSG:4326) or center + zoom. NOTE: CSDI features are NOT drawn on this image; use build_map for the interactive version, or an ArcGIS Online print service (needs ARCGIS_TOKEN) for data-on-basemap statics.

Write a printable static map WITH the data drawn, as one self-contained SVG page.

Basemap tiles are referenced by URL (they load in any browser with internet), features are drawn as real vector shapes on top — so the result prints cleanly to PDF from the browser's print dialog. No key, no headless browser, no image library.

Call any ArcGIS Online REST endpoint with the configured token.

Examples: path="search", params={"q": 'title:"my layer"', "num": 5} → find items path="content/items//data", params={} → item JSON path="content/items/", params={} → item metadata path="/0/query", params={"where":"1=1","outFields":"*","f":"json","returnCountOnly":"true"} → count analysis: run a GP task under //submitJob (then /jobs/)

Publish a loaded CSDI dataset to ArcGIS Online as a hosted feature layer. Needs ARCGIS_TOKEN. Returns itemId + the new service URL.

Are these datasets still being served, and how stale are they?

No dataset_id -> lists the limit oldest datasets in the catalogue, so you can see drift at a glance. probe=True also calls the CSDI export endpoint to confirm it answers right now (reads 2 KB only, costs a round trip each — capped at 10 per call).

Treat catalogue date and data freshness as SEPARATE things: the date below is CSDI's dct:modified (when the record was revised), not the statistical reference period of the data.