---
id: engines/sparrow
about:
  label: SPARROW
  description: >-
    Open-source, solar-powered edge device from Microsoft's AI for Good Lab that
    runs camera-trap and bioacoustic AI models on site and sends the results out
    by satellite or GSM, for biodiversity monitoring in remote places.
  tags:
    - biodiversity
    - camera trap
    - bioacoustics
    - edge ai
    - satellite
    - wildlife monitoring
tool:
  homepage: 'https://microsoft.github.io/SPARROW/'
  form: open-source hardware and software platform
  licence: MIT
  cost: free software; hardware per the published bill of materials
provenance:
  origin: agent
  sources:
    - title: >-
        SPARROW: Solar-Powered Edge AI for Biodiversity Monitoring | Microsoft
        AI for Good
      url: 'https://microsoft.github.io/SPARROW/'
    - title: GitHub - microsoft/SPARROW
      url: 'https://github.com/microsoft/SPARROW'
    - title: Project SPARROW and the Future of Conservation Technology
      url: >-
        https://www.microsoft.com/en-us/research/publication/project-sparrow-and-the-future-of-conservation-technology/
    - title: >-
        How AI helps gather remote biodiversity data: Interview with Microsoft’s
        Juan Lavista Ferres
      url: >-
        https://news.mongabay.com/2026/08/how-ai-helps-gather-remote-biodiversity-data-interview-with-microsofts-juan-lavista-ferres/
  created: '2026-10-17'
  reviewed: '2026-10-17'
  reviewed_by: 'Claude, from the sources listed'
  status: draft
import:
  source: openai-linkedin-scrape 2026-10-17
  id: DT049
  kind: monitoring_system
  family: observations_data_and_reconstruction
  priority: high
  verified: true
  posts:
    - P142
  posters:
    - Amy Rosenthal
    - Rhett Ayers Butler
domains:
  - ecology
---
SPARROW (Solar-Powered Acoustic and Remote Recording Observation Watch) is a field unit from Microsoft's AI for Good Lab that collects images from up to 150 solar wifi camera traps, audio from an AudioMoth recorder and readings from small temperature, humidity, pressure and gas sensors, then runs detection and species models on the device itself. Only the summarised results leave the site, over a Starlink Mini satellite link or a GSM connection; when there is no link, data is kept locally and synced later. Images of people are removed on the device before upload.

The lab's June 2026 paper reports deployments in Colombia, Peru, Tanzania and the United States across tropical, temperate and montane ecosystems, running around the clock and gathering more than two million images and recordings in the first 190 days without on-site intervention. An August 2026 Mongabay interview with lab director Juan Lavista Ferres puts it in eleven countries on five continents, with early sites in Colombia's Magdalena Medio and Los Amigos in the Peruvian Amazon, and credits it with helping Chester Zoo and Uganda's wildlife authority locate an endangered giant pangolin.

The software is on GitHub under the MIT licence, packaged as Docker containers, with a bill of materials and assembly guide. The models are PyTorch-Wildlife exports run through ONNX: MegaDetector v6, an Amazon species classifier and a bird detector. The repository now targets a Raspberry Pi 5 with an NVMe drive, powered by two 100 W panels, an MPPT controller and a 24 V LiFePO4 battery; earlier units used an NVIDIA Jetson, and the documentation site still describes that configuration.

Operational in the field since at least 2025; the published hardware guides are dated 2025 and the documentation was updated in May 2026.