Earth Swarm
Data and API

The data, for science and for business

Everything on this site comes from a public JSON API. Researchers use it free; companies pay for site monitoring.

Access tiers

No key

Public

  • What this website uses
  • Confirmed finds, 6 hours or more after they were made, each saying how it was confirmed
  • Positions to a 0.01 degree grid
  • 2,000 requests a day per address
Free

Research

  • For universities and research institutions, with an institutional email address
  • Non-commercial use only; cite Earth Swarm
  • Positions to a 0.001 degree grid
  • 10,000 requests a day

Apply for research access

Paid

Commercial

  • Site level monitoring of your own land
  • Change metrics between seasons and years
  • Bulk export and higher limits
  • A service level agreement

Tell us about your sites

Under agreement

Verifier

  • For accredited validation and verification bodies
  • Full precision, every validation status
  • The signed receipt of every find, to check against published Merkle roots

Sensitive species are never available at better than a 0.2 degree grid through the public, research or commercial tiers, and their photos are never published. Full precision goes only to verifiers under agreement.

Licences and citation

Earth Swarm records shared with GBIF carry CC BY-NC 4.0: free for science, while commercial users come through the API rather than reselling the raw records.

GBIF context records shown on this site keep their own licence, either CC0 1.0 or CC BY 4.0, and every one carries its GBIF occurrence link, dataset and rights holder in the API. We import no records under any other licence.

Please cite: Earth Swarm (year). Earth Swarm confirmed species finds. https://earthswarm.ai, accessed on the date you accessed it.

Use Earth Swarm in an AI assistant

Earth Swarm has a public MCP server, so an AI assistant can answer wildlife questions from this data, with the same privacy as this website. No key or sign-in is needed for any of the public tools. A step by step guide for everyone is on AI assistants.

Server address: https://earthswarm.ai/mcp

The tools

Your own record, if you sign in

If you record with the Earth Swarm app, you can let your assistant read your own record. The first time you ask it about your own record, your assistant asks you to sign in to Earth Swarm: use the email you use in the Earth Swarm app (the sign-in page is on stringer.run, which Earth Swarm runs on). Your assistant then sees only what the app shows you, with places named, never pinpointed, and can change nothing. To stop it, remove the connector in your assistant, or disconnect it in the Stringer dashboard at stringer.run/dashboard, under Settings, Connected apps. Either way it stops at once.

Every answer says what kind of claim it makes (confirmed on Earth Swarm, recorded nearby in national records, background), and carries a link back here.

Terms

Free display of limited, attributed results, with a link back to Earth Swarm; no bulk extraction or resale. Records from the NBN Atlas and GBIF keep their own upstream terms, and their credit travels with every answer that uses them. For bulk data, use the API above.

Earth Swarm runs on Stringer, which handles real-world tasks of every kind: https://stringer.run

Endpoint reference

Base address https://earthswarm.ai/v1. Send a key, if you have one, as Authorization: Bearer es_.... Every response is JSON. Errors look like {"error":"E_RATE_LIMITED","message":"..."} with a stable code and a sentence for a person. Each response carries x-ratelimit-limit, x-ratelimit-remaining and x-earthswarm-tier.

Coverage by grid cell

Counts of confirmed finds and distinct species per cell, for a year and season. GBIF context records come back separately, in context. bbox is minLng,minLat,maxLng,maxLat; season is spring, summer, autumn, winter or all; taxon and cell are optional.

curl -s 'https://earthswarm.ai/v1/coverage?year=2026&season=autumn&bbox=-6,49.8,2,56'

Observations

Individual confirmed finds, coarsened for your tier. Each carries confirmation: model_agreed, reviewed, community or expert. Filter by bbox, taxon, from and to (YYYY-MM-DD) and source (earthswarm, gbif or all). Page with limit and offset.

curl -s 'https://earthswarm.ai/v1/observations?bbox=-3.6,50.6,-3.4,50.8&taxon=gbif:2492462&from=2026-01-01&to=2026-12-31' \
  -H 'Authorization: Bearer es_your_key'

Species

curl -s 'https://earthswarm.ai/v1/taxa?q=robin'
curl -s 'https://earthswarm.ai/v1/taxa/gbif:9515886'

Projects, registry and monitoring periods

curl -s 'https://earthswarm.ai/v1/projects'
curl -s 'https://earthswarm.ai/v1/projects/prj_example'
curl -s 'https://earthswarm.ai/v1/registry?q=Acme'
curl -s 'https://earthswarm.ai/v1/periods/per_example'

A closed monitoring period publishes the Merkle root over the receipt hashes of every find made in it, and the leaves, so anyone can check a single find against a number fixed at the time.

Recently confirmed

curl -s 'https://earthswarm.ai/v1/recent?limit=12'

The latest confirmed Earth Swarm finds, newest first (at most 50), and running totals: finds, species, named places and firsts. Each find carries its species and English name, date, confirmation, photo, its coarsened point and cell, the named place it was made in and whether it holds a first (earthswarm, the first Earth Swarm record of the species; place, the first there). A sensitive species never names its place. The same publication rules as Observations.

Places

curl -s 'https://earthswarm.ai/v1/places?q=bushy'
curl -s 'https://earthswarm.ai/v1/places/plc_...'

Search the places register (parks, commons, woods and nature reserves from OpenStreetMap) by name, then read one place: its name, kind, centre and area, its latest confirmed finds and its totals, in the shape of Recently confirmed.

What the NBN Atlas records near a place

curl -s 'https://earthswarm.ai/v1/places/plc_.../nbn'

Species the NBN Atlas holds records of within a radius of the place (0.5 to 3 km, by its size) over the last ten years, with record counts. Open-licence records only (CC0, CC-BY, Open Government Licence), never a species on Earth Swarm's sensitive list, names and counts only. Asked at most once a month per place. These are national records for context, not Earth Swarm finds.

What a place is part of

curl -s 'https://earthswarm.ai/v1/places/plc_.../designations'

The protected areas at the place's centre, from Natural England's open data: National and Local Nature Reserves, Sites of Special Scientific Interest, Special Areas of Conservation, Special Protection Areas, Ramsar wetlands, National Parks and National Landscapes. England only. Asked at most every 90 days per place. Open Government Licence.

A place's species

curl -s 'https://earthswarm.ai/v1/places/plc_.../species'

The species recorded in one place of the places register, each with how many finds of it there are. Only confirmed finds count; sensitive species are never listed, nothing made in the last 6 hours is counted, and no date, position or surveyor is returned.

BioBlitz days

curl -s 'https://earthswarm.ai/v1/bioblitz?lat=51.45&lng=-0.31'

BioBlitz days within 30 km, today and the next fortnight: a place, a date, and on the day, everyone's shared count so far, as a number of confirmed finds and of species. Never who, never which species, never where inside the place. The app sends its position rounded to 0.01 degrees.

Photos

curl -s -o find.jpg 'https://earthswarm.ai/v1/photos/f_...'

Only for confirmed, published finds of species that are not sensitive. Every photo served here is a copy with all of its metadata removed: no EXIF, no GPS position, no capture time, no camera details. Only verifiers under agreement receive the original file.

Some finds are short video clips rather than photos. Clips are never published and never available through the public, research or commercial tiers; they are kept for validators and verifiers only.

For the Earth Swarm app

curl -s 'https://earthswarm.ai/v1/keys/evidence'

The public key finds are sealed to. The upload format is fixed by the Earth Swarm find protocol, version 1.1: POST /v1/ingest/find carries the signed receipt, the sealed find, and exactly one sealed piece of media, either sealed_photo (a JPEG of at most 2 MiB) or sealed_clip (an MP4 or MOV of at most 30 seconds and 25 MiB sealed). The media keeps the place and time it was taken inside the sealed wrapper, and Earth Swarm checks them against the signed find: the position within 25 metres or the fix's own accuracy, whichever is larger, and the time within two minutes.

curl -s -X POST 'https://earthswarm.ai/v1/finds/status' \
  -H 'content-type: application/json' \
  -d '{"receipt_hashes":["<64 hex characters>"]}'

Where a surveyor's own finds have got to, for the app's My finds list. Send up to 100 receipt hashes. Each one Earth Swarm holds comes back as {"receipt_hash","status","taxon"}, where status is received, pending, validated or rejected and taxon is the confirmed gbif: identifier, only once validated. Once the species model has looked at the find, the line also carries suggestions, its top three guesses. Once the find is confirmed for play, it carries play: the identification tier, the species, the named place the find was made in (fixed when it was first confirmed, or null) and any firsts it holds (earthswarm, the first confirmed Earth Swarm record of that species; place, the first in that place). A hash Earth Swarm does not hold is left out. Nothing else is ever returned: no position, photo, surveyor, note or validator. A receipt hash cannot be guessed and only the phone that made the find holds it. Metered like every public request.