Last Larch · ARIA

Operations Manual

A step-by-step guide to setting up and using ARIA, from downloading the file to generating your first sector-specific CAP alert descriptions.

Getting started · 1

Download and open ARIA

ARIA is a single HTML file, available in two versions. There is nothing to install. It runs in your web browser.

1
Download the file

Go to lastlarch.com and click Download tool →. Choose one of the two versions.

FileSizeChoose it when
ARIA.htmlAbout 0.3 MBYou have an internet connection each time you open ARIA. It loads its map, database and file-reading libraries from public servers when it opens.
ARIA-offline.htmlAbout 2 MBYou need ARIA to open without a connection, for example in the field or on a restricted network. Those libraries and the fonts are built into the file, which makes it about six times larger.
Both versions work the same way once they are open. Save the file somewhere you can find it again, such as your Desktop or a dedicated folder like Tools/ARIA/.
2
Open it in your browser

Double-click the file you saved. It opens in your default web browser. Chrome, Edge, Firefox, and Safari all work. The standard version needs an internet connection to open. The offline version does not.

Even in the offline version, three things need internet: the map background (Esri), online data sources such as ArcGIS, WFS and URLs, and AI providers. Without a connection the map background stays grey and ARIA shows a short notice. You can still draw or paste a polygon, query local files, load ARIA-Lex and use Library only mode.
3
You are ready

You will see a map on the right and a sidebar on the left. A sample dataset (European city points) is preloaded so you can test the tool immediately.

If your browser shows a security warning about opening local files, click Allow or Open anyway. ARIA does not send your data anywhere unless you choose an online source or an AI provider.

Getting started · 2

How ARIA works

ARIA is a three-step wizard. Each step builds on the last.

→
Step 1: Query impact area

Load your GIS data sources, then draw or paste a polygon representing the hazard footprint. ARIA queries every active source and shows what is inside: schools, health facilities, roads, or whatever else your data contains.

→
Step 2: Review and configure

Choose which layers to include, set the hazard type and severity, and connect your ARIA-Lex library and AI model. This is where ARIA grounds the output in validated language before the AI adds local specifics.

→
Step 3: Alert descriptions

ARIA produces sector-specific <headline> and <description> blocks ready to paste into your CAP editor. Each block is tagged to show whether it came from your library or was newly generated.

→
Moving between steps

Every step has a ← Back button so you can return and adjust at any time. You can change your polygon, add a source, or change the hazard settings without losing your work. The step indicators at the top show where you are and which steps are complete.


Step 1: Data sources · 3

Load a locally saved file

Use this for GeoJSON files or shapefiles stored on your computer. Your data never leaves your machine.

1
Drag and drop

The easiest way. Drag a .geojson, .json, or .zip (shapefile) file from your file explorer and drop it anywhere on the ARIA window. It loads instantly and appears in the sources list.

2
Or use the Add source form

Click + Add source in the sidebar. Enter a name, select Shapefile (.zip) from the type dropdown, click File, and browse to your file. Click Add source.

3
The source appears in the list

A toggle switch shows it is active. The badge shows the format. If you have already drawn a polygon, the query runs automatically.

Prefer GeoJSON over shapefiles where you have the choice. It loads faster and handles special characters in place names better.

Step 1: Data sources · 4

Connect an ArcGIS layer

ARIA can query ArcGIS Online FeatureServer layers directly. It also accepts web map URLs and extracts all layers automatically.

1
Click + Add source and select ArcGIS REST / Web Map
2
Paste your URL

ARIA accepts three URL formats. Paste whichever you have:

FormatExample
FeatureServer Recommended …/FeatureServer/0
FeatureServer (no index) Auto-fixed …/FeatureServer → ARIA appends /0
Web Map Viewer URL Auto-extracts all layers arcgis.com/home/item.html?id=…
If you paste a web map URL, ARIA fetches the map definition and adds every queryable layer as a separate source automatically. Public maps only. Private maps require login.
3
Click Add source

ArcGIS sources are queried live when you draw a polygon. The features inside your polygon are fetched directly from the service. No data is cached locally.

If the query returns an error, the layer may require authentication or have CORS restrictions. Try opening the URL directly in your browser first to confirm it is publicly accessible.

Step 1: Data sources · 5

Connect a WFS endpoint

WFS (Web Feature Service) is a standard used by many national spatial data infrastructures and open data portals.

1
Find your WFS GetFeature URL

A WFS URL typically looks like:
https://example.com/geoserver/ows?service=WFS&version=2.0.0&typeName=layername

2
Select WFS endpoint and paste the URL

ARIA appends a bounding box filter derived from your polygon when querying. You do not need to add BBOX parameters yourself.

WFS endpoints hosted on different domains sometimes block browser requests (CORS). If a WFS source returns an error, your IT team may need to configure the server to allow cross-origin requests, or you may need to use the data as a locally downloaded file instead.

Step 1: Data sources · 6

Load a GeoJSON from a URL

If your data is hosted online as a GeoJSON file, ARIA can fetch it directly.

1
Click + Add source, select GeoJSON (URL)

Paste a direct URL to a .geojson or .json file, for example a GitHub raw file URL or an HDX download link.

2
Click Add source

ARIA fetches and caches the file in memory. It shows a green success message with the source name. The data is now available for polygon queries without a network request each time.

HDX datasets (data.humdata.org) often offer GeoJSON downloads. For shapefiles from HDX, download the .zip and drag and drop it instead. ARIA parses shapefiles natively.

Step 1: Polygon · 7

Draw a polygon on the map

Use this when you want to define the hazard footprint manually or when exploring data coverage.

1
Navigate to your area of interest

Pan and zoom the map. If you have local data loaded, click ⊕ Fit map to data to jump straight to the right area.

2
Click the polygon tool

In the top-left of the map, click the pentagon-shaped draw icon. Your cursor changes to a crosshair.

3
Click to place vertices

Click once for each corner of your polygon. Click the first point again (or double-click) to close the shape. The query runs automatically as soon as the polygon is closed.

You can also use the rectangle tool for a quick bounding box query.
4
Edit if needed

Click the edit (pencil) icon in the map toolbar to drag vertices. The query re-runs automatically when you save the edit.


Step 1: Polygon · 8

Paste a polygon from a warning

Use this when you have coordinates from a meteorological service, a flood model output, or any other external source. ARIA accepts three formats.

1
Find the Paste polygon box in the sidebar

It is below the data sources list. Click into the text area.

2
Paste your coordinates in any of these formats
FormatExample
Lat/lng pairs -1.94, 30.06
-1.94, 30.14
-2.01, 30.14

One pair per line. Comma or space separated.
GeoJSON {"type":"Polygon","coordinates":[[...]]}
Feature, FeatureCollection, or geometry object.
WKT POLYGON((30.06 -1.94, 30.14 -1.94, ...))
Standard Well-Known Text format.
3
Click ⬡ Draw from paste

ARIA draws the polygon on the map, zooms to it, and runs the query automatically.

ARIA detects coordinate order automatically. If the first value is greater than 90, it assumes longitude first. Otherwise it assumes latitude first. This covers most met service formats.

Step 2: Configure · 9

Load ARIA-Lex

ARIA-Lex is the seed library of pre-validated early action messages. Loading it means the AI uses approved language as its foundation instead of generating everything from scratch.

1
Download ARIA-Lex

Go to lastlarch.com and download ARIA-Lex.db. Save it alongside ARIA.html or in a known location.

2
Click the library status indicator

In the top-right corner of ARIA, a small indicator says No library loaded. Click it to open the library settings panel.

3
Drop or browse to ARIA-Lex.db

Drag ARIA-Lex.db onto the drop zone, or click it to browse. The library loads in seconds. The indicator turns green and shows the message count.

You can also drag a .db file directly onto the main ARIA window at any time. It is sent to the library loader automatically.
4
The library stays loaded for your session

You need to reload it each time you open ARIA, because no data is stored between sessions. This keeps your data local and private.

What the library covers now. ARIA-Lex holds 1,513 active messages for seven hazards (cyclone, flood, heatwave, wildfire, tsunami, cold wave and disease outbreak) across seven sectors. Two gaps remain. There are no earthquake messages yet, so for earthquake the AI writes every sector from your GIS data and tags it Synthesised, and Library only mode finds nothing. Messages for Food Systems & Agriculture are included but switched off until they are reviewed. All messages are marked provisional and should be reviewed by domain experts before operational use.

Step 2: Configure · 10

Set hazard type, severity and authority level

These three settings control which library messages are retrieved and how the AI frames the output.

Hazard type

OptionDB codeUse for
Cyclone / HurricanecyTropical cyclones, hurricanes, typhoons
FloodflRiver flood, flash flood, urban flooding
HeatwavehwExtreme heat events
WildfirewfForest fire, bushfire
EarthquakeeqSeismic events. No library messages yet, so all output is AI-generated.
TsunamitsCoastal wave events
Cold wave / FreezecwExtreme cold, frost, snowstorm
Disease outbreakdoEpidemic, pandemic risk areas

Severity

LevelCAP urgencyUse when
ExtremeImmediateThreat to life, take action now
SevereExpectedSignificant impact likely, prepare now
ModerateFutureImpact possible, monitor and prepare
MinorFutureLimited impact expected

Authority level

OptionUse when
Auto (infer from data)Let the AI decide based on the scale and type of your GIS data. Good default.
NationalAlert is being issued by a national agency, for national distribution
Regional / ProvincialAlert targets a province, region, or state
Local / OperationalAlert targets a district, municipality, or specific facilities
Individual / HouseholdAlert language is directed at members of the public
Once you set hazard and severity, ARIA shows a preview below the form indicating how many library messages exist for that combination and which sectors they cover. This tells you, before generating, what the output will be grounded in.

Step 2: Configure · 11

Connect an AI model

ARIA works with any OpenAI-compatible API, including a model running on your own computer or network. You provide the API key when one is needed. It is stored only in your browser and is never sent to Last Larch.

1
Choose a provider preset or set a custom endpoint
PresetProviderGet API key at
ClaudeAnthropicconsole.anthropic.com
GPT-4oOpenAIplatform.openai.com/api-keys
GeminiGoogleaistudio.google.com
CustomAny OpenAI-compatible API on an allowed hostEnter the endpoint URL and model name. See the note below.
2
Paste your API key

Click the preset button for your provider. The endpoint and model name fill in automatically. Paste your API key into the key field.

ARIA saves your key per provider in your browser's localStorage. You only need to enter it once. It is there the next time you open ARIA on the same computer and browser.
3
Advanced: a hosted custom endpoint

Select Custom and enter the endpoint URL and model name of an OpenAI-compatible API, for example OpenRouter at https://openrouter.ai/api/v1/chat/completions. For an endpoint reachable on the open internet, ARIA sends the request through its relay worker, and the worker only forwards to api.anthropic.com, api.openai.com, generativelanguage.googleapis.com and openrouter.ai. A public endpoint on any other host is refused.

4
Advanced: a local model on your own computer or network

Select Custom and enter the address of a model running on your own computer or local network, for example http://localhost:11434/v1/chat/completions for Ollama. ARIA recognises addresses like localhost, 127.0.0.1 and private network ranges (10.x, 172.16–31.x, 192.168.x) and calls them directly from your browser. The request never reaches the relay worker or the open internet. The field below the endpoint says which mode is active.

A local server usually needs no API key. Leave the key field blank unless yours requires one.
Your local server must allow the request. Browsers block a page from reading a response unless the server explicitly permits it (CORS), and ARIA opened as a file has no address of its own, so the server must allow it specifically.
ServerHow to allow it
OllamaSet the environment variable OLLAMA_ORIGINS=* before starting it, for example OLLAMA_ORIGINS=* ollama serve. Set it permanently in your system's environment variables to avoid repeating this.
LM StudioIn the local server tab, turn on Enable CORS before starting the server.
llama.cpp server, text-generation-webui, and similarStart the server with its CORS or --allow-origin flag set to allow all origins, or place a reverse proxy in front of it that adds an Access-Control-Allow-Origin: * header.
If the connection is refused, ARIA shows a plain error naming the address and CORS. See Troubleshooting.
When generating descriptions, a summary of your GIS data is sent to the AI provider: feature counts, attribute names, and up to 40 sample rows per layer. No raw geometry coordinates are transmitted. For a hosted provider, review its data policy if you are handling sensitive infrastructure data. For a local model, this summary never leaves your computer or network.

Step 3: Output · 12

Generate and read the output

Click Generate alert descriptions → and ARIA does the rest. You can also use Library-only mode to skip the AI entirely.

1
Library-only mode (no AI required)

If you have ARIA-Lex loaded and want to work on an air-gapped machine (use the offline version) or without any AI API, tick the Library only checkbox next to the Generate button. ARIA looks up matching pre-validated messages by hazard, severity and authority level and renders them directly. Library only mode makes no network calls. With the offline version it works with no internet connection at all. With the standard version, a connection is needed only to open the tool.

All output in library-only mode is tagged Library. There is no Contextualised or Synthesised output. You get the validated text as written, nothing more.
Library-only mode cannot slot in specific place names or feature counts from your GIS data, because that contextualisation step requires the AI. Choose the mode that suits your operational context.
2
AI mode: wait for the loading message

ARIA first queries your library, then calls the AI. The loading message tells you what is happening, for example: "Grounding against 24 library messages then contextualising…"

3
Read the source tags on each sector card

Every sector card has a tag in the top-left telling you where the language came from:

Library The pre-validated library message was used as written, with local names and counts filled in. This is the highest-confidence output because the core language has been approved.
Contextualised A library message was the starting point and was adapted with specific local data, such as named facilities, populations, or conditions from your GIS layers. Review before publishing.
Synthesised No library message existed for this sector and hazard combination. The AI generated the language from your GIS data alone. Treat with more caution and review carefully before publishing.
4
Check the library messages used section

On Library and Contextualised cards, a green panel shows the exact library messages the AI used as its source. This lets you verify the AI stayed true to your validated language.


Step 3: Output · 13

Copy and export descriptions

1
Export a validation CSV from Step 1

Once you have query results in Step 1, a green ↓ CSV button appears in the results panel. Click it to download a flat CSV of every feature found across all layers, with source name, geometry type, coordinates, and all properties. Use this file to confirm what the query found before you generate alert language.

This CSV requires no AI and no library. It is the raw GIS query output, useful for record keeping or for sharing with colleagues who do not need the full alert text.
2
Copy a single sector description

Click the Copy button on any sector card. Both the <headline> and <description> are copied to your clipboard, formatted for pasting directly into a CAP editor.

3
Download all as .txt

Click Download .txt at the bottom of the results. You get a plain text file with all sector descriptions, each labelled with its source tag ([LIBRARY], [CONTEXTUALISED], or [SYNTHESISED]) and the hazard and severity.


Reference · 14

Troubleshooting

?
The map is not loading

The map background (Esri World Street Map) needs an internet connection in both versions. When it cannot load, ARIA shows a short notice and switches to a simplified offline basemap: country borders and a coordinate grid, built from Natural Earth data and included in the file. It is for orientation, not precise measurement. Drawing, local files, the spatial query and Library only mode are unaffected either way, since they work from your data's coordinates, not from the map image.

?
ARIA.html opens as a blank or broken page

The standard version loads its libraries from public servers when it opens. If you have no connection, or your network blocks those servers, the page cannot start. Use ARIA-offline.html instead. It has the libraries built in.

?
A local model will not connect

ARIA shows an error naming the address and mentioning CORS. This means your local model server is either not running or has not been told to accept requests from ARIA. Confirm the server is running and reachable at the address you entered, then see Connect an AI model for how to enable CORS in Ollama, LM Studio, and similar tools.

?
A source is showing an error

Open the URL in your browser directly. If you see a login page, the layer is private. If you see JSON, the issue is likely CORS, meaning the server is blocking browser requests. Try downloading the data as a file and loading it locally instead.

?
No features found in my polygon

Check that the source's toggle is switched on (green). Click ⊕ Fit map to data to confirm the source data covers the area where you drew your polygon.

?
The AI is returning an error

The most common causes are a wrong API key, a wrong endpoint URL, or a misspelled model name. Check all three. For Claude, the key starts with sk-ant-. For OpenAI, it starts with sk-.

?
The library preview says "No library messages for this combination"

The current hazard/severity combination may not have entries in your library file. Check the library preview in Step 2. The AI still generates descriptions, and they are tagged Synthesised.

?
Library-only mode gives no results

Check that ARIA-Lex is loaded (the indicator in the top-right should be green). Then confirm that the hazard and severity combination you selected has entries in the library. The preview in Step 2 shows this. If it says "No library messages for this combination", try a different severity level or check that the correct .db file is loaded.

?
Still stuck?

Email fysh@lastlarch.com with a description of what you were doing and what happened. Screenshots help.


Reference · 15

Running the local data server

If you have GIS files stored on your computer and want ARIA to load them automatically, without dragging and dropping them each time, you can run a small local server that watches a folder and serves those files to ARIA.

1
Make sure Python is installed

Open a terminal (Command Prompt on Windows, Terminal on Mac/Linux) and type:

python --version

If you see a version number of 3.7 or higher, you are ready. If not, download Python from python.org. The free installer takes two minutes. On Mac and Linux, the command is usually python3, so use python3 --version here and python3 aria-server.py in step 4.

2
Create your library folder

Create a folder anywhere on your computer, for example Documents/ARIA-Library/. Put your GeoJSON files in it. ARIA loads everything in this folder automatically.

Files must be GeoJSON (.geojson or .json). Shapefiles should be converted to GeoJSON first, or loaded via drag-and-drop instead.
3
Save this Python script as aria-server.py

Copy and save the following into a file called aria-server.py inside your library folder:

# ARIA Local Data Server (Last Larch)
# Save this file in your GeoJSON library folder and run it.
# ARIA will connect to it automatically on localhost:8765

import http.server, json, os, urllib.parse

FOLDER = os.path.dirname(os.path.abspath(__file__))
PORT   = 8765

def layer_files():
    return sorted(f for f in os.listdir(FOLDER)
                  if f.lower().endswith(('.geojson', '.json')) and not f.startswith('.'))

class Handler(http.server.BaseHTTPRequestHandler):
    def respond(self, code, body=b''):
        self.send_response(code)
        self.send_header('Access-Control-Allow-Origin', '*')
        self.send_header('Content-Type', 'application/json')
        self.send_header('Content-Length', str(len(body)))
        self.end_headers()
        self.wfile.write(body)

    def do_GET(self):
        path = urllib.parse.unquote(urllib.parse.urlparse(self.path).path)
        if path == '/layers':
            layers = [{'name': os.path.splitext(f)[0], 'filename': f} for f in layer_files()]
            self.respond(200, json.dumps(layers).encode())
        elif path.startswith('/file/') and path[6:] in layer_files():
            with open(os.path.join(FOLDER, path[6:]), 'rb') as f:
                self.respond(200, f.read())
        else:
            self.respond(404)

    def do_OPTIONS(self):
        self.send_response(204)
        self.send_header('Access-Control-Allow-Origin', '*')
        self.send_header('Access-Control-Allow-Methods', 'GET')
        self.send_header('Access-Control-Allow-Headers', '*')
        self.send_header('Content-Length', '0')
        self.end_headers()

    def log_message(self, *args): pass  # suppress terminal output

print(f'ARIA library server running. Watching {FOLDER}')
print('Open ARIA and click Connect in the sources panel.')
http.server.ThreadingHTTPServer(('localhost', PORT), Handler).serve_forever()
4
Run the server

In your terminal, navigate to the folder and run:

python aria-server.py

A confirmation message appears. Leave the terminal window open. Closing it stops the server.

5
Connect from ARIA

Open ARIA in your browser. At the top of the sources panel, Local data server not connected appears. Click Connect. ARIA loads all GeoJSON files from your folder and checks for changes every 10 seconds, so you can drop new files into the folder while ARIA is open and they appear automatically.

The server only accepts connections from your own computer (localhost) and only serves the GeoJSON files in your library folder.

Reference · 16

License

ARIA is free and open source under the MIT License. You may use, adapt, and share it, including for commercial purposes. The one condition is that the copyright and license notice stay with any copy.

MIT License · Copyright Last Larch

You are free to:

  • Use and share: copy, publish, and distribute the tool in any medium or format
  • Adapt: modify, merge, and build upon the tool for any purpose, including commercially, and sublicense or sell it

Under the following terms:

  • Notice: include the copyright notice and the MIT permission notice in all copies or substantial portions of the tool
  • No warranty: the tool is provided "as is", without warranty of any kind
→
How to credit ARIA

When you use or adapt ARIA, keep the copyright and license notice that ships with it. A credit line such as the following is also welcome:

ARIA by Last Larch (lastlarch.com) · MIT License
→
Full license text

The full text is in the LICENSE file in the ARIA repository. You can also read the MIT License at opensource.org/license/mit.

→
Questions or custom implementations

If you are deploying ARIA within a national authority, UN agency, or research institution and would like support, customisation, or a formal partnership, contact Last Larch at fysh@lastlarch.com.