Fitbit, Linux, and you

Fitbit devices are a family of wearable gadgets that track everyday health metrics (such as motion, calorie burn, and sleep cycles) on a minute-by-minute basis. There are certainly other hardware and software solutions to track such data, but most of them are "sportsband" products, focused on monitoring intense training sessions; the upshot of the Fitbit is that it monitors the contribution of everyday activity: walking, climbing stairs, and sitting still behind the keyboard. Since the latter activity consumes a large portion of the average software developer's day, the device has unsurprisingly attracted a following among programmers. But as is often the case, Linux users have some hurdles to clear in order to make use of the product.

Fitbit devices work by logging time-stamped readings from internal sensors; the current models incorporate both accelerometers and altimeters, tracking typical foot traffic. By associating the movement data with some basic facts about the size of the user, secondary statistics like calorie count are calculated, and some logic is used to smooth out the results and determine when a wearer is (for example) driving on the autobahn rather than running. The sensors themselves are not unique — plenty of smartphones ship with similar offerings — but the devices have gained a loyal following thanks to their low power consumption, small size, and relatively rugged construction.

Sync

Unfortunately the company releases support software only for Windows and Mac OS X. Support software is required because the devices are designed to offload their logs periodically to servers running at fitbit.com, which provides a free web service where users can track their individual metrics over time and compare it against other users. The early Fitbit models came bundled with a USB "base station" that connected wirelessly to the Fitbit devices using the ANT protocol; every so often when the device was in range it would synchronize the latest activity readings to the web service.

For that first generation of hardware, there is an open source solution for Linux users. In 2011, Kyle Machulis released libfitbit, a Python tool that supports pulling data off of the first generation Fitbit devices and uploading it to the official web service. In addition to pushing the device's data logs upstream, libfitbit also saves them locally in plain text format. The library does not support the altimeter sensor introduced with the Fitbit "Ultra", however. Machulis has subsequently created an umbrella project for libfitbit and an array of related health-monitoring device projects called OpenYou — he does not appear to be actively developing it, but there are several forked branches on Github that introduce improvements.

In mid 2012, Paul Burton started another Fitbit sync project called fitbitd that re-implemented libfitbit in C and ran the synchronization service as a daemon, with the client connecting to it over D-Bus. Fitbitd also added support for the Fitbit Ultra's altimeter sensor, allowing users to keep a close eye on their stair-climbing prowess.

But in late 2012 the company updated its line of tracker devices again, and switched radio protocols from ANT to Bluetooth Low Energy (LE). On one hand, the switch to Bluetooth should make implementing Linux support a little simpler, since ANT is proprietary. But on the other, only relatively recent kernels support Bluetooth LE, and thus far neither of the existing open source Fitbit sync projects have undertaken the task of implementing Bluetooth support. The new Fitbit devices do ship with a Bluetooth LE adapter for the PC, so the hardware is in place, but it may be a matter of time before the devices are widespread enough in the developer community for an effort at reverse-engineering the Bluetooth traffic to pick up steam.

Nor is there any great hope that the company will provide a Linux synchronization client of its own. Canonical's Jono Bacon reported in January 2012 that he had reached out to Fitbit about writing a Linux driver for the devices, but that after an initial expression of interest, the company stopped replying. Several application developers have asked about Linux support on the Fitbit API development list (most recently in September 2012) and were met with similar silence. In fact, Fitbit has still not introduced an Android synchronization app for the Bluetooth products; only recent Android devices support Bluetooth LE, but the company has released an iOS app that synchronizes with the devices over Bluetooth LE.

Data access

Of course, synchronization from a Linux machine is one aspect of using the device, but it is not the only one. For some people, the notion of uploading one's activity data to a remote service is anathema to begin with, and even more so if the data cannot be retrieved. The company offers an API so that developers can link other applications (such as those for jog/run planning or nutrition tracking) to the same user account. The API can be used to retrieve historical data from the web service in JSON format, but it only provides access to daily totals of the covered metrics.

At the moment, the most robust way to retrieve one's data from fitbit.com is through John McLaughlin's Fitbit for Google Apps script, a JavaScript tool that can extract the data directly into a Google Docs spreadsheet. It requires registering as a Fitbit application developer in order to obtain an authorization key, but Fitbit has recommended the script on the mailing list, so the company is clearly not opposed to the idea.

Through the official API, McLaughlin's script can download all of the daily totals from a user account going all the way back to the account's creation. The retrieval period is configurable, so users can run the script once to grab the complete history, then run shorter updates for subsequent daily or weekly additions. The result is a nicely formatted table of time-series data suitable for any inspection or graphing tool.

Daily summary data is certainly useful, but the minute-by-minute data is far more useful, particularly when it comes to analyzing patterns of activity over the course of the day. Libfitbit allows owners of the ANT models to save their full sensor logs locally, but so far Bluetooth device owners are out of luck. It is conceivable that the company regards high-resolution data as something it can monetize, since the basic data logging services is free of charge (Fitbit currently offers a paid "pro" version of the web service that somehow acts as personal trainer to the user; I was not intrigued enough by that idea to pony up the cash necessary to try it out).

In October 2011, Fitbit announced that it would begin allowing access to minute-by-minute "intraday" data "on a case by case basis to a limited number of developers." It is not clear what happened after that announcement. As of January 3, 2013 the company still has not opened intraday data to the public. The developer wiki mentions a Partner API which still requires that the developer contact Fitbit and ask for access, but the limited number of developers granted that access do not seem keen to discuss it.

Others have inquired on the mailing list about "raw" access to the device's sensor logs before they are uploaded to the web service. This would be even more appealing to the privacy conscious, and if one wants to do data mining personally it would be much faster to simply capture the data locally before it takes a round trip to the server. Here again Fitbit expressed interest in the concept, but has not acted on it subsequently.

Alternative hardware

Fitbit is far from the only player in the health-monitoring hardware market, but none of the other manufacturers are particularly friendly to Linux and free software either. Machulis's OpenYou project hosts a number of other libraries for communicating with personal monitoring devices, such as Nike's training-oriented Fuelband and the BodyBugg armband. Like libfitbit, both are designed to allow extraction of log data before it is uploaded to the relevant web server, but neither is actively developed. Similarly, there was a 2009-era effort to reverse engineer support for personal monitoring devices manufactured by BodyMedia, but it has not been updated in several years.

The "Up" hardware devices from Jawbone offer a slightly better experience. There is no Linux software available, but the full data set can be downloaded from the web service in CSV form. Motorola's motoACTV is an Android-based wristwatch with health monitoring features, and it is possible to root the device and side-load third party applications. Perhaps the strangest entry in the field is the Zeo line of sleep-monitoring headbands. The company released an official library to decrypt the logged data and export it to CSV, and a separate project to access the raw Zeo data in real-time. Although it was initially advertised as "under an open source library" it can currently only be accessed by agreeing to a severely limiting terms and conditions document. This dichotomy is perhaps a simple misunderstanding about licensing, but that does not resolve the situation for free software sticklers.

Another alternative that is frequently proposed is to make use of the position and accelerometer sensors already found in many smartphones. There has been at least one effort to write an open source pedometer application for Android, but both the reviews of the app and issues filed on the bug tracker reveal a key difficulty. Apparently many Android phones do not allow an application to access and record sensor data when the phone is in an inactive state. Since the goal is to track movement over the course of an entire day, this can make the app useless if one's phone manufacturer choose to enable this restriction.

Exactly what Fitbit plans to do with the API and with raw sensor data remains an open question. The company previewed another set of new devices at the Consumer Electronics Show in January 2013, but did not announce changes to the developer program. Consequently, Linux users in possession of the newer devices will need to start capturing and decoding Bluetooth LE traffic in order to make use of their hardware. The good news is that so far all of the Fitbit models successfully reverse engineered appear to use the same data format. The bad news is that capturing and reading Bluetooth traffic logs is such a sedentary activity.