Skip to content
Branch: master
Find file Copy path
Find file Copy path
180 lines (129 sloc) 8.09 KB

Search with autocomplete

If you are building an end-user application, you can use the /autocomplete endpoint alongside /search to enable real-time feedback. This type-ahead functionality helps users find what they are looking for, without requiring them to fully specify their search term. Typically, the user starts typing and a drop-down list appears where they can choose the term from the list below.

To build a query with autocomplete, you need a text parameter, representing what a user has typed into your application so far. Optionally, you can specify a geographic point where the search is focused, this will allow users to see more local places in the results.

User experience guidelines

There are several user experience pitfalls to watch out for when implementing a client-side typeahead solution:

Requests must be throttled

Since autocomplete requests generally correspond directly to user input, it's important to account for fast typers and throttle requests. Some devices and networks (for example, mobile phones on a slow connection) may respond poorly when too many requests are sent too quickly, so be sure to do some testing on your own. Learn more in this interactive demo.

Our testing shows that between 5 and 10 requests per second is the range with the best balance of resource usage and autocomplete responsiveness.

Many Pelias services also enforce hard per-second rate limits, so setting a client-side throttle can help you avoid exceeding those limits. It's better to send fewer requests on your own terms than rely on a server's rate limiting to decide which requests will receive a complete response.

Account for asynchronous, out of order responses

You cannot be sure responses will be returned in the same order they were requested. If you were to send two queries synchronously, first 'Lo' then 'London', you may find the 'London' response would arrive before the 'Lo' response. This will result in a quick flash of 'London' results followed by the results for 'Lo', which can confuse the user.

Autocomplete requests with more characters typed will often return faster, since the search space of the query is smaller, so this is not an edge case.

Use search even with autocomplete

While the autocomplete endpoint is designed specifically for use with user-entered inputs, the search endpoint can still be useful in certain situations. A common paradigm is to send an autocomplete request on key presses (throttling appropriately as described earlier), but sending a search request when the user hits the enter key or a submit button.

This approach allows for the speed and partial-input handling of the autocomplete endpoint to be used when needed, and the accuracy and additional functionality of the search endpoint to be used when possible.

In our experience, most users have been trained by other websites that submit buttons paired with an autocomplete interface will trigger a "more in depth" search, and so will naturally use this ability.

Use a pre-written client library if possible

If you are already using Leaflet, we recommend using the Nextzen (previously Mapzen) leaflet-geocoder plugin. This plugin follows all the autocomplete guidelines listed here and has been well vetted by many members of our community.

Global scope, local focus

To focus your search based upon a geographical area, such as the center of the user's map or at the device's GPS location, supply the parameters and focus.point.lon. This boosts locally relevant results higher. For example, if you search for Union Square:

From San Francisco:

/v1/autocomplete? square

1)	Union Square, San Francisco County, CA
2)	Union Square, New York County, NY

From New York City:

/v1/autocomplete? square

1)	Union Square, New York County, NY
2)	Union Square, San Francisco County, CA

The /autocomplete endpoint can promote nearby results to the top of the list, while still allowing important matches from farther away to be visible. For example, searching hard rock cafe with a focus on Berlin:

/v1/autocomplete? rock cafe

with focus.point you will find the Berlin restaurant first:

1)	Hard Rock Cafe Berlin, Berlin, Germany
2)	Hard Rock Café, San Giljan, Malta

without focus.point you will find the most popular restaurants first:

1)	Hard Rock Cafe, Pune, Maharashtra
2)	Hard Rock Café, San Giljan, Malta


You can filter the results in several ways: the original data source and/or the type of record.


The sources parameter allows you to specify from which data sources you'd like to receive results. The sources are as follows

  • openstreetmap or osm
  • openaddresses or oa
  • geonames or gn
  • whosonfirst or wof


with sources=openaddresses you will only find addresses on Pennsylvania Ave or Street:

1) 8 R Pennsylvania Avenue, Amity, PA, USA
2) 7 Pennsylvania Avenue, Amity, PA, USA
3) 9 Pennsylvania Avenue, Cherry, PA, USA

without sources=openaddresses you will find the most popular Pennsylvanias first:

1) Pennsylvania, USA
2) Pennsylvania Avenue Heights, Washington, DC, USA
3) Pennsylvania, Satsuma, AL, USA


The type of record is referred to as its layer. All records are indexed into the following layers:

layer description
venue points of interest, businesses, things with walls
address places with a street address
street streets,roads,highways
country places that issue passports, nations, nation-states
macroregion a related group of regions. Mostly in Europe
region states and provinces
macrocounty a related group of counties. Mostly in Europe.
county official governmental area; usually bigger than a locality, almost always smaller than a region
locality towns, hamlets, cities
localadmin local administrative boundaries
borough a local administrative boundary, currently only used for New York City
neighbourhood social communities, neighbourhoods
coarse alias for simultaneously using all administrative layers (everything except venue and address)
postalcode postal code used by mail services


with layers=coarse you will see only administrative areas with names containing Starbuck

1) Starbuckville, NY, USA
2) Starbuck, MN, USA
3) Starbuck, WA, USA

with layers=venue you will see only the venues by that name

1) Starbucks, Braunschweig, Germany
2) Starbucks, Islip, NY, USA
3) Starbucks, Austin, TX, USA


Searching in a country

Sometimes your work might require that all the search results be from a particular country. To do this, you can set the parameter value to the alpha-2 or alpha-3 ISO-3166 country code.

Available autocomplete parameters

Parameter Type Required Default Example
text string yes none Union Square floating point number no none 48.581755
focus.point.lon floating point number no none 7.745843
boundary.rect.min_lon floating point number no none 139.2794
boundary.rect.max_lon floating point number no none 140.1471
boundary.rect.min_lat floating point number no none 35.53308
boundary.rect.max_lat floating point number no none 35.81346
sources string no all sources: osm,oa,gn,wof openstreetmap,wof
layers string no all layers: address,venue,neighbourhood,locality,borough,localadmin,county,macrocounty,region,macroregion,country,coarse,postalcode address,venue string no none 'GBR'
boundary.gid Pelias gid no none whosonfirst:locality:101748355
You can’t perform that action at this time.