But the story was even wilder than what I learned at the time, and the downstream influence of the company's employees still reverberates in Silicon Valley today. There was even a movie made about them, grabs of which I've used here under fair use. That prompted me to get the DataRover back out again, still peacefully charging under the desk with its games and software intact, to see what we could do to make Magic Cap and my trusty old Rover relevant again in a modern world.
My original intention was to do a brief introduction to the units and then talk more about how you can program one, but there's so much history here and so much we can say about these unique devices that first I'm going to spend this entire entry talking about the hardware (the first and last entries in the line), the technology and the company. In an upcoming post we're going to sit down and program one. Buckle up, because it's time for quite a tale.
General Magic was founded in July 1990 with a lot of hoopla and a lot of mystery, and in particular Apple's backing, who was a minority investor and the fledgling company's largest corporate shareholder. Apple CEO John Sculley even took a position on the board of directors, and its three founders entirely came from Apple, including Mac luminaries Bill Atkinson (developer of such minor, inconsequential things as MacPaint, QuickDraw and HyperCard) as chairman, Andy Hertzfeld (utterly unknown for his contributions to the Mac OS and Toolbox) as "software wizard" and senior VP, and another former Apple employee, Marc Porat, as president.
Apple even crafted their press release, in which they stated their intention to design and develop a new class of "personal intelligent communicator" for a future connected world, but was deliberately vague about further plans. Apple's spokesman did make reference to a project called Paradigm intended to underlie General Magic's new technology, which Apple chose to spin off because it fell "outside the scope" of Apple's personal computer business (how things have changed). Paradigm and the related miniaturized Pocket Crystal were actually Porat's own research projects at Apple's Advanced Technology Group, developed sometime around 1989 or early 1990 for a portable multifunction device that could "scan, photograph, connect [to a Mac] and type" using a "wand, diskette, compact media [a small CD shown] or smart card [a PC Card static RAM card shown, but also later a floppy-like unit]" with scenarios for checking a to-do list, making a phone call, listening to the radio and "link[ing] information" (some sort of messaging). The device was handheld, with a separate external keyboard, and attachments for a probe, a phone and a scanner. Josh Carter has pictures of the Paradigm concept book.
Most industry observers expected that the new company would produce their own mobile hardware, but what was particularly strange about the new spinoff was that Apple was already doing the same thing, and had been doing so in a parallel project as far back as 1987. That was of course the Newton, a skunkworks project started by engineer Steve Sakoman with R&D VP Jean-Louis Gassée's blessing. Sakoman and Gassée's precipitous departures caused a corporate reckoning and led to Bill Atkinson's invitation to Hertzfeld, Porat, engineer Steve Capps (then assigned to Newton), the now independent designer Susan Kare and Apple CEO John Sculley in March 1990 to discuss future directions. Sculley felt that Newton was closer to his 1987 Knowledge Navigator vision and assigned ATG VP Larry Tesler, formerly Xerox PARC, to run it, permitting the spinoff of Paradigm and Pocket Crystal. Apple's nevertheless enthusiastic support for the new company caused much consternation in the Newton unit, particularly after General Magic's unsuccessful attempt to recruit Capps as well, and the project was further burdened by their original choice of the notoriously buggy AT&T Hobbit as CPU (which would come to haunt Gassée a second time; the early BeBox had the same problem). Excitement circulated around the new company, with echoes of Paradigm in trade journals of the time; InfoWorld reported in December 1991 that the new device would have pen input, a 2.5" CD-ROM, wireless connectivity and a GUI interface — intriguingly similar to Porat's original concept.By 1992, however, Newton was starting to fire on all cylinders: a new CPU (ARM), a smaller form factor (Pocket Newt), and wild new interest in the portables sector after the successful launch of Apple's PowerBook line. Whether this was the original idea all along or not, and the pictures above show hardware in development at various points including a screen and a physical keyboard, in January 1992 InfoWorld's Mark Stevens (as Robert X. Cringely) remarked that General Magic wasn't doing an end-user product at all. Instead, the new company was devising an operating system and reference design that they would license to Sony, IBM, French Telecom ... and Apple. In February 1993 this was formalized at a press event as the General Magic Alliance, with Apple, AT&T, Matsushita, Philips, Motorola and Sony all taking equity states in the new venture and declaring their interest to incorporate the core technologies into their future products.
Of these, AT&T's infrastructure role as what was dubbed the Consumer Messaging Service was arguably the most crucial; while the hardware OEMs were a big deal, the software and user experience vision covered more than just the interface they ran. Potentially the most consequential of General Magic's technologies was Telescript, devised by Jim White and Rich Miller, a programming language for developing "agents" that could remotely move from system to system and perform tasks. In an era where linkages between systems were much lower bandwidth, the concept allowed a client to send an agent to do work locally on a destination service and push back just the desired outputs rather than having the agent running on the client and trying to run requests from afar. Agents could be transmitted through a variety of means, even messages (so-called "smart E-mail"), containing a complete execution bundle with the stack, variables and code frozen and ready to run. Security for untrusted foreign code thus became a fundamental part of the execution environment: Telescript ran in a bytecode interpreter rather than compiled on the metal (also allowing multiple architectures to run the same agent), RSA signatures identified the agent and whom they were operating on behalf of ("Telenames"), explicit permissions were required to resources and operations to limit agents' access to only what they needed ("permits"), and Telescript objects carried limited execution budgets ("Teleclicks") to ensure they would not overwhelm a busy system. Agents could even spawn their own sub-agents to complete tasks in parallel. But they all had to have somewhere to go, and AT&T was to provide that environment as well as connectivity for more mundane communications like E-mail, fax and voice. White was hopeful that Telescript's open specification and crossplatform engine would allow Telescript agents to run virtually anywhere.
Signs of what was to come were present if you knew where to look. The phone shown above was a 1993 mockup showing the envisioned new operating system (from a BBC TV demo). There was no keyboard or even any keys except possibly for a large red button. Instead, the phone was dominated by a single large screen showing icons you could press and a road scene with buildings representing services. The 1993 Jurassic Park also had a prototype Motorola device prominently placed on Dennis Nedry's desk (possibly through Michael Crichton's influence, who pushed Steven Spielberg to include computing technology in the film; Macintoshes are prominently shown, as are Silicon Graphics hardware — both used, at different times, to develop Magic Cap). Like the phone, there are no obvious buttons, just one big display. Motorola planned to create an array of wireless devices for the ARDIS radio network they then owned jointly with IBM and bridge the radio network to cellular for voice and data. The underlying hardware would be portable to multiple operating systems, including Magic Cap, Microsoft At Work and Apple Newton. Overall the prototype's appearance is suggestive of the 1995 Motorola Envoy.What struck me most from the videos I scrubbed through from those early days of development (and, yes, the movie) is how much the leadership and their developers — termed "magicians" — believed in their mission. They had some big names, too. Some got bigger later.
These grabs alone (from a 1993 update video), from the very looks on their faces and the shine in their eyes, show how much they were into it. Details mattered — the user interface considered sound effects as important as graphics — but they never lost sight of the big picture that enjoyment, however that's defined, is as equally essential to the user experience as utility. Really, it was the magicians' enthusiasm for the task, to design a device that would not only transform how people accessed and interacted with information but also a device that they would truly love and enjoy using, that made the company incredibly fascinating to me. You saw that kind of optimistic vision a lot in the earlier days of computing. You don't see it as much now, and there are some very dark patterns today that go exactly in the opposite direction. And that rabbit logo? That was Bowser, magician Phil Goldman's grey dwarf bunny who freely roamed the office and left "gifts" for other magicians (just another office perk). He was immortalized in topiary outside the building in Mountain View. Eventually the time came to put up or shut up and in January 1994 the Magic Cap operating system was officially unveiled at a press event for the first time. While much of the demonstration was "virtual" using a Macintosh runtime (naturally, early implementations were actually mockups in HyperCard), attendees were treated to a front row seat at the literal Magic Cap desktop, and an extensive trip downtown past multiple services. There was also a demonstration of Telescript, shown on the Mac emulator. A simple agent was E-mailed in a "frozen" state from Andy to Bill with the task of mining Bill's "name cards" (contacts) to find a dentist. The frozen state of the agent contained its code, the current state of the call stack and all of its variables in a self-contained executable package. Bill was alerted he had received an agent from Andy (described by Andy on the message), consented to its execution, and allowed it to search his contacts for dentists and E-mail itself back to Andy with the two name cards it found. All of this occurred over the AT&T network.While that demo was virtual, real hardware did appear, albeit without any markings:
The contour of the display matches a Sony HIX-300 (codenamed Bamboo), technically the first Magic Cap device the public would ever see, but never named and never released for commercial sale as such. Other bits of hardware were also visible during the demo, including a handheld device Bill was using and a fax/phone-like unit behind Andy: There were also mockups of future services, here a news service and a virtual travel agent. And at the end, a happy birthday banner to the Mac, and a shout-out to the magicians in the audience. Other than proof that Magic Cap wasn't vapourware, the other notable output was publicly revealing minimum specifications for Magic Cap devices: a built-in two-way modem at 2400bps or better, speaker and mic, minimum 1MB of RAM and 3MB of ROM, a high-speed serial port (the "Magic Bus") and at least one PCMCIA Type II slot.While the Sony and Motorola devices were supposed to emerge in roughly similar time frames, Sony got there first. Rumours swirled repeatedly around the new device during its long gestation: in 1991, they said it would be $700 to $800 with a later price point of $500 "as soon as possible" and InfoWorld in December 1993 even reported Sony's new wonderpad was supposed to be in colour. The latter was definitely not true of the HIX-300, which also apparently didn't have enough ROM, horsepower or battery life to be up to the task of Magic Cap 1.0, and neither was true of the device that rose from Bamboo's ashes: Cedar, sold as the PIC-1000 Magic Link.
At $950 for its MSRP (but actually sold at $995, or about $1950 today) it was probably as cheap as it could be while still meeting the minimum specification. The CPU was a Motorola 68349 "Dragon I," a microcontroller based on the Motorola 68030 (as many of the Macs used to develop Magic Cap were), specifically the XC68349FT16V variant at the lowest supported clock speed of 16MHz. The 68349 includes two DMA and two serial channels, 4K of on-chip SRAM configurable as four individually mappable single 1K blocks, and a "configurable instruction cache" with four independent blocks usable as either 1K of L1 I-cache or an additional 2K of SRAM. It isn't clear how these are configured on the Magic Link. The twin 2MB ROMs (total of 4MB) are on a detachable daughtercard and in theory could be upgraded, though at least for the PIC-1000 they never were. In the box was the Magic Link itself and a delightful little vinyl case it snapped into and closed with Velcro. It also came with an AC adapter, a phone cord, a CR123A backup battery, a battery pack and various paraphernalia. What was not in the case was a rechargeable battery — the battery pack uses six AAAs — nor any kind of sync cable for connecting it to a PC or Mac. A 10-hour rechargeable battery pack (Sony NP-500H, but NP-500s were also okay) was a separate option.The case is hard plastic but with a soft-touch coating. At the time this was very well-received but it turns out it slowly degenerates and tends to stick to things (like the fuzz inside the case, dust particles, you, etc.). This very minimally used unit has already started to mottle slightly almost entirely from age.
Again, meeting the minimum spec at the price point was the order of the day, so there was only a single PC Card slot. This was mostly intended for memory cards (definitely static RAM up to 1MB but never tried larger capacities or linear flash), though you could also install a Sony-branded wireless card that used the SkyTel paging network for connectivity. SkyTel for Magic Link primarily provided messaging but also had voice mail, fax, news headlines and stock quotes as options. At least initially, however, its services were downlink-only (no sending!), and if you had to ask how much it cost, you already couldn't afford it. Memory cards do not expand the internal RAM of the device per se; rather, they serve as removable storage, more or less like floppy disks would have. No other cards were ever supported in this device.The pen (stylus) was on the same side as the card slot, slid and locked into a spring-loaded chamber, and popped out with a quick press. There was no handwriting recognition, but given that the Newton had already become the butt of jokes for its own foibles this was probably a wise omission. Eventually Aha Software Corp. sold a tool for this called InkWriter for Magic Cap ($129), but Microsoft bought them out in 1996 and killed the product. Otherwise, an on-screen keyboard was built into the operating system you could tap with the pen.
Apart from that, the only other visible controls are two functionally equivalent OPTION buttons. The option button can be treated as a key in its own right but primarily works like a right click for the pen. For example, if you tap on the dump truck, you see its contents with an button to empty it, but if you option-tap the dump truck, it empties immediately. Another useful shortcut is option-tapping the magic lamp, which instead of bringing up special options as a usual tap would, gives you a volume slider and a shortcut button to the other system settings.
The phone jack was for the built-in modem at a — cheap! — dismal bare minimum speed of 2400bps (fax supported at 9600bps). Other ports included an IR receiver/blaster (beams to other Magic Cap devices at 38.4kbps, but not IrDA), power, headphones, an option port and a serial sync port ("Magic Bus"). The serial sync port supported a $130 external keyboard, which was (you guessed it) optional, as well as the (optional) computer connector, but the only device known to connect to the option port was a headset for phone calls (also optional, $80). However, it did have a built-in speaker and microphone, so the base unit could still make and receive phone calls as long as everyone else didn't mind eavesdropping, and some users found they could use a much less expensive Y-adaptor to make and receive calls instead of using Sony's bespoke phoneset. In the box were offers from America Online with ten hours free, AT&T PersonaLink, and SkyTel. These services had built-in support in the interface as buildings downtown you could enter along with their baseline messaging connectivity; AT&T hawked their Market Square virtual shopping storefront while America Online touted stock quotes, headline news and tech support. All three offered messaging, though again, if you were wire-free you couldn't actually send anything. Notably, America Online conspicuously advertised their system's connectivity with other Internet-enabled E-mail services such as MCI Mail and SprintMail, but what AOL did not conspicuously advertise was Telescript support, and neither did SkyTel. We'll come back to this. Sony left the desktop connectivity piece to third-party providers, though I only know of one that actually delivered a product. IntelliLink's Magic Xchange (announced at $80, sold for $100) furnished import/export, backup, printing, software installation and even using your expensive Mac or PC as a keyboard. The necessary serial sync cable came in the package.There were two other places Sony cheaped out, and these were big ones. In broad strokes Magic Cap divides onboard RAM into program and storage space (much as other mobile OSes of the time like NewtonOS and, later, Palm OS). The minimum amount of supported RAM for Magic Cap is 1MB and that's exactly what you get in the PIC-1000. Roughly half of that megabyte goes to program space, variables, stack, etc. — that's 512KB, for those of you keeping score at home — and the rest stores your data and any additional apps you load. Even in 1994 that wasn't a lot.
But what uniformly got the worst ratings from reviewers and users was the screen. Although generously sized at 480x320 and four shades of grey at a time when the Newton was just 320x240 and strictly black and white, the Sharp-manufactured LCD has hideous contrast and no backlight at all. It's so bad it makes my Data General One's horrid screen look decent and even in moderate indirect sunlight shown here it's badly washed-out. Although I shot at an angle to reduce reflections, I wasn't trying to make it look bad in these photos: it just is. On the other hand, besides the standard Magic Cap desk set (phone, Rolodex, E-mail, calculator, calendar, clock, notepad/drawpad, file cabinet, storeroom and library) Sony provided a number of pack-in apps including a couple games, Intuit Pocket Quicken, PenWare's PenCell (a spreadsheet compatible with Microsoft Excel and Lotus 1-2-3) and an IR remote control (!). All of these are built into ROM and come ready to use. I had the most interest in the IR remote control but it's clearly a product of its time. It only talks to Sony devices, of course, and it only supports TVs, VCRs (up to three, because if you could afford a Magic Link, you could afford three VCRs), camcorders, CD players, MiniDisc players, LaserDisc players, and DAT and cassette players. Both my Sony A/V receiver and Sony DVD player completely ignored it. But I found the PenCell spreadsheet to be excellent. While limited to 128 rows by 64 columns because of the pathetic amount of working memory, it supports a rather comprehensive set of formulas, exports and imports Excel and 1-2-3, and even does charting.It did all of that rather slowly, however, consistent with the other big complaint from reviewers: the speed. All that interactivity and animation came at a cost and a system that was effectively the equivalent of a lowest-specced Macintosh II just couldn't keep up with it.
Yet for all its warts the Magic Link was still a competitive product with other contemporary handhelds like the HP 100LX, Psion Series 3a, Tandy Zoomer and, yes, Apple's Newton MessagePad 110. While many were cheaper, most didn't come with even the basic communication options the PIC-1000 had standard and suffered from their own memory constraints, the PIC's built-in software was more than comparable in function, and although there were more compact options none had a screen as large as the Magic Link's no matter how bad it was. The built-in connectivity and intuitive interface proved very popular with (patient) users, as Steve Wozniak effused here at length in 1994.Meanwhile, despite plans in 1994 to be first to market, Motorola's own wireless unit was lagging badly. In January 1994 internal sources claimed it would have not one but two wireless modems and two PCMCIA Type II slots, plus everything that the PIC-1000 had and a faster landline modem too, to emerge mid-1994 at the eye-watering MSRP of $1500 ($2800 in 2022 dollars) — plus wireless service. In fact, Motorola was so "all in" on wireless that they bought ARDIS partner IBM out completely in July. Even at that premium price, however, it still offered just the single megabyte of RAM and had the same underpowered 68349 CPU. In July 1994 Motorola announced it would ship by "at least the end of 1994"; it did not emerge until February 1995. The Envoy 100 took a step back in at least one respect by actually putting the contrast control in software (even the Magic Link gave you a thumbwheel), but the 4MB ROM was flash and upgradeable, it came with a hardcase, and the ARDIS service was great for being on the go until you actually got the bill. Although the radio modem ran at 4800bps, the landline ended up just 2400bps after all and told you what Motorola's priorities really were. The unit was still greyscale despite some screenshots allegedly being in colour, and early Envoys were plagued by an odd bug after 48 days of uptime causing the device to slow down drastically. If you had the radio modem, the fix was pushed out to you. If you didn't, you got to call a 1-800 number.
Theoretically Magic Cap could run on any platform, but it was developed on 68K, the first two devices were 68K, and the only third-party development tool to support it was CodeWarrior which at that time compiled 68K. Metrowerks released Magic/MPW in March 1995, using the Macintosh Programmer's Workshop as the front end but the Metrowerks C compiler instead of Apple's. The Macintosh runtime that demonstrated Magic Cap in 1994 evolved into the Magic Cap simulator, but apps ran on the Mac as "native" 68K code (however, because the simulator required an FPU and the PowerPC Mac 68K emulator emulated a 68LC040, you had to install an FPU emulator to make it run on a Power Mac). The simulator included a built-in Inspector for looking at objects "live" in their "natural habitat." Enhancements for MacsBug and the MPW Shell were included as well as the ObjectMaker object compiler and the Bowser Pro (there's that wascally wabbit again) class and source code browser for examining classes, creating templates and seeing class hierarchies. An additional debugger for Telescript, Telebug, was mentioned on the package but essentially required the optional Telebug Kit to debug packages on a connected device. The compiler came with a pretty manual which served only to document the blindingly obvious.Later that year Metrowerks evolved their offerings to use their own CodeWarrior IDE, familiar to Macintosh programmers of a certain age everywhere. A "lite" version that could not add new source files appeared on the CD-ROM for the only third-party manual on programming Magic Cap I know of, Barry Boone's Magic Cap Programmer's Cookbook from October 1995. Furthermore, it only supported running packages in the simulator and did not have facilities for downloading them to a connected device.
In the forthcoming post on Magic Cap programming I'll go into this in more detail, but even though Magic Cap is object-oriented, it's written in C, not C++. (The later Rosemary SDK actually builds from C++, but the object model is still C-based.) Objects aren't even structs; you can't just take a generic ObjectID or Reference, cast it to a pointer and start calling methods — those aren't actually pointers and the compiler won't let you. Classes are defined in a separate specific syntax that is compiled by a special tool to a C header file, making function dispatch more or less a combination of autogenerated glue code and preprocessor macros. You can define specific object instances either written out by hand as a text file, or within Magic Cap itself using the Magic Hat (option-tap the Stamper's title bar) to visually place them and then dump what's there. In fact, using the Magic Hat in construction mode is the standard way you would build any scene or user interface in Magic Cap.
Now with the first hardware iteration out the door and the beginnings of an ecosystem, the obvious question was ... what's next?
Getting the form factor smaller had always been part of the idea. In August 1995 Andy Hertzfeld demonstrated a rough Magic Cap phone prototype. Though built into a bulky shell and connected by a long ribbon cable, the device was real, and showed a dramatically smaller desk scene that could be styled into different appearances, such as this button skin: While this was the phone's only known appearance, the small-screen Magic Cap concept survived long enough to be implemented in a a clamshell prototype called Zodiac using a Sharp Zaurus case and custom electronics.Making a bigger Magic Cap was a possibility too: magician Steve Perlman, whose interest in combining computing and television started in high school when he hacked a TV to be his text-based home computer's graphic display and later with his work on QuickTime, wanted to build an interactive TV set using Magic Cap as the basis. Indeed, at that point it seemed like anything could be a possible candidate to run it. (Perlman left Magic Cap in 1994, and in July 1995 joined fellow magicians Bruce Leak and Phil Goldman and his bunny in a new startup to commercialize the concept: WebTV.)
But further iterations on the small or big screen concepts never saw the light of day as General Magic proceeded with a different frontier for Magic Cap: the desktop. Strictly speaking Magic Cap already existed for the Macintosh in the form of the simulator, but it was oriented almost exclusively towards developer debugging and was hardly suitable as an end-user application. Nevertheless, even though GM wanted to bring Magic Cap to Microsoft Windows (being developed by an outside contractor) because of the potentially large office userbase, GM had already signed an agreement with Apple in 1991 that the Macintosh version would be first.
Somehow or another Apple was placated, because the only commercially sold desktop version of Magic Cap was indeed for Windows 95 (the above is from an old eBay auction; I don't have a physical package). A pre-release emerged in 1995. Magic Cap was officially sold as a product for Windows 95, but will run on Windows 3.x with Win32s. Despite the larger screen real estate, however, the desk is still a fixed 480x360 and you can't resize the window, and while officially colour for the first and only time solely appears in dithered shades (Virtual PC was running this in Windows 98 on an emulated 1024x768 display with 24-bit colour). I did like the colours in the hallway, though. Inside the AT&T PersonaLink building downtown. Although first and foremost sold as a self-contained complete messaging solution, Windows Magic Cap was nevertheless a full runtime and could even load packages (presumably compiled for x86), though I'm not aware of any that were ever released nor any publicly available desktop tools that supported building for it. But on top of all that, it only worked with a modem. This didn't seem like a good time in a corporate environment where LANs were starting to become more common — and corporate customers would likely have been more interested in the product than a home user even though General Magic priced it competitively at $49. Finally hitting the market in October 1996, it fizzled like an expired Alka-Seltzer.General Magic was also working on an updated operating system and hardware, also targetted for 1996. Apple had begun the transition to the Power Mac and industry observers expected the new system would run on PowerPC as well, though the company advised second-generation applications might no longer be compatible. New vendors interested in developing their own hardware were told to wait, but General Magic's timelines kept slipping, and inquirers got cold feet as other early PDA manufacturers (notoriously EO) started to fold.
In the meantime an incremental release as Magic Cap 1.5 was released with faster suspend and a snappier interface, which Sony added to their new PIC-2000 Magic Link ("Oak") in November 1995 along with an extra megabyte of memory, a second PC Card slot, an upgraded 14.4Kbps fax/data modem and (at last!) a backlight. It was rather better received but failed to sell in numbers and the PIC-2000 was the last Magic Cap device Sony would make. Motorola's update was more modest, offering Magic Cap 1.5 to current Envoy owners for $29, and releasing the Envoy 150 with a better-reflecting screen instead of a backlight plus support for Microsoft Mail and Lotus CC:Mail. The ARDIS service was still seen as more costly than it was worth, however, and while Motorola wouldn't release sales figures analysts estimated only around 25,000 Envoys had been sold at the time the Envoy 150 came out in April 1996. Motorola discontinued the Magic Cap Envoys in December and sold off ARDIS entirely in 1998, becoming today's DataTAC, which is still used for legacy paging systems due to its more forgiving lower speeds and better penetration in buildings. The Envoy brand name itself was later recycled for other products.
Things weren't much better in the C-suite: General Magic had gone public as GMGC in February 1995 and its stock nearly doubled the same day, but just four months later in June the company was posting a $6.5 million loss on sales of only $1.9 million. In addition to the delays with the next generation hardware and the Windows version, AT&T got out of the PersonaLink business in June 1996, eliminating the solitary commercial service where Telescript was supported. Bill Atkinson, burned out, went on leave in 1995 and never came back; Andy Hertzfeld sold his shares in 1996 and quit as well. The company started shifting more towards software as the hardware side floundered, including marketing Active Paper's Presto Mail client and Presto Links Web browser to facilitate connecting devices to corporate intranets in April 1996, shipping a software modem implementation (SoftModem) for Pentium MMX and various embedded chipsets like MIPS in July 1996, and releasing the Magic Internet Kit in September, an SDK update enabling TCP/IP, PPP and Ethernet support for free download or as part of CodeWarrior 10 Gold, which still supported Magic Cap development. Telescript was repositioned into a Java-like product that now could run as an agent on Web sites ("Tabriz"), but there was still little interest in infrastructure for code moving from system to system, and Java's explicit write-once-run-anywhere mentality and heavy support from Sun eventually eclipsed the older platform.
Under new CEO Steve Markman, replacing Marc Porat, the company started doing more work with voice interfaces and telephony. Their initial product, code-named Serengeti, acted as a voice-driven assistant over the phone — eventually, the user's own 1-800 number. It was planned to synchronize with Outlook, Internet Explorer and Netscape Communicator as well as Windows CE and WebTV devices, though the Serengeti screenshot above clearly shows a mockup file running from a developer's hard disk. After posting another $12 million loss in October 1996, General Magic cut staff by over half and spun off the SoftModem product as AltoCom in March 1997, which notched wins in the Philips Nino, Compaq C-Series and Samsung InfoMobile handheld devices and later in set-top boxes.During this chaotic period, development on the next generation device and the next major release of Magic Cap continued. In December 1996 the company gathered developers to discuss Rosemary, the next release of Magic Cap. General Magic wouldn't say who was developing the hardware or who would sell it, only that the new device would be MIPS-based using a 32-bit R3000-class processor. It would have a built-in software modem and TCP/IP support (by incorporating the Magic Internet Kit) including E-mail and a web browser, but because Rosemary would move to C++ with a new API, applications wouldn't be backwards compatible without porting work and a recompile. (As it happens, Rosemary and compatible apps are written in C++, but the Magic Cap object model used is still C-based.)
In reality there was no company interested in building the next overpriced doorstop, so General Magic decided to do it themselves using Oki as the OEM. The new device started life as Sputnik, crammed into PIC-1000 shells painted pink with the same hideous screen, then into the purple Apollo prototype, and finally the finished product and the last publicly released Magic Cap device, the DataRover 840 announced in December 1997 running Magic Cap 3.1.
The DataRover 840 was the only Magic Cap device to bear the General Magic name explicitly, though with a new logo. Where the Magic Link at least had pretenses of being a consumer device, however, the DR840 was all business, intended for vertical applications and field work. I mean, just look at the stinking box: the back of it shows a form for entering trouble tickets on light poles, and the front has a form for writing traffic citations!It's also notable what the box doesn't mention: Telescript. By now it was good and dead. Not even a single parenthetical reference exists in the 340-page user guide.
A smaller DataRover was also developed in a clamshell form factor with the same-sized screen (to avoid the software issues with Zodiac), but no slots. This was the DataRover 440, codenamed Gemini, but it had many design flaws and hardware bugs and ended up unreleased while General Magic was haemorrhaging cash. The 840 was the only member of the DataRover family to make it to commercial sale.
This boxed unit is one of the four DR840s I own (two used, two new). While the outer part of the box was thrashed when I got it, almost everything inside was pristine. It came with the device itself, a CR123A backup battery, the main Li-ion battery, a spare stylus, a telephone cord, the AC adapter and two manuals. I also keep my spare sync cable in this box even though it didn't originally come with one.Despite Oki being the manufacturer, the included rechargeable Li-ion battery is still a Sony camcorder battery. Sony badged it as the "GMB001" (General Magic Battery, presumably), but it's really just a regular NP-520 (7.2V/1350mAh), and in fact many Sony and Sony-knockoff batteries will fit in the chamber as long as they're 7.2V and have the same charging points. I currently use a Wasabi Power clone of the NP-F550 (their part# is BTR-F550-JWP) which is a tight fit but nearly doubles battery life at 2600mAh. Although General Magic billed the original battery as good for "over 8 hours continuous use," obviously that use didn't figure in the backlight.
The DataRover itself, with Bowser still badged on the corner. Let's size up its contemporaries. The HP LX series remained strong sellers at this time (I myself had a 95LX before I got my Palm m505) but were best considered very small DOS laptops rather than PDAs, and other than the unique GEOS-based OmniGo 100 did not use pen input. In the more traditional form factor, the original Pilot 1000 and 5000 had come out in 1996 but the Palm III would not emerge until March 1998, and they still used a 16MHz 68030 derivative like the Magic Link did in 1994 (I'll talk about the DR840's CPU in a moment). Microsoft themselves had just launched the Palm-size PC refurb (Windows CE 2.0) after the relatively tepid response to the 1996 "Pegasus" systems, but the first of the new line wouldn't emerge until around mid-year. So that left really only one device in the DR840's size and power class at the time of its release, and naturally that was the Newton, newly (November 1997) available as the top-of-the-line MessagePad 2100. Like the DataRover would be, it was the last of its kind. And really, the devices are pretty comparable physically: both of them have flip-up screen covers and large touch backlit LCDs, and while the DR840 is not quite as long as the MP2100 it's a bit thicker and a little wider, although the MP2100 is also slightly heavier (1.4lb/0.64kg versus 1.15lb/0.52kg). As far as overall portability goes the comparison is probably a wash. They also both had two PC Card slots and a selection of ports. In the DR840's case, those ports included a new Magic Bus serial port on the same side as the cards (incompatible with previous sync cables; an optional serial keyboard — actually a Palm device with a converter — connected here), as well as the built-in phone jack and an IrDA-compatible infrared transceiver in the display cover's hinge notch. The modem was now V.32terbo 19.2kbps with 9600bps fax capability. (Remember how I mentioned General Magic had developed a software modem? Hold that thought.) Headphones and the AC adaptor also connected on the back and the backlight was controlled with a dedicated physical switch you could flip with a fingernail. The power switch and contrast control are on the other edge not shown here. The screen is still 480x320 and four grey shades; the MP2100, meanwhile, had upgraded to the same resolution but now sported 16 greys. On the other hand the DR840's screen contrast is excellent and its backlight is way better than any Newton's, including the flagship MP2100's — at least, for as long as it works (read on). General Magic (and later Icras, but we'll talk about them later on) progressively advertised a variety of peripherals and accessories as compatible with the DR840 such as barcode scanners, modems, Ethernet cards, printers, mag-stripe readers, WiFi (but no encryption, not even WEP), a serial keyboard, a vehicle mount, a soft case and a digital camera that wrote images to ATA flash cards.I only have one of these accessories to show off, and that's the embossed leather case with a shoulder strap perfect for, uh, writing parking tickets or scaling streetlights, though kidding aside it's rather a nice case for carrying it around. However, it has two big deficiencies: there's no obvious place to put the stylus (I slip it under the strap on the back), and while there are cutouts for the microphone, speaker, power adapter, IR, headphones, phone jack, contrast and power switch, and even a Velcro door for the Magic Bus and card slots, the option buttons are completely covered up.
What was not a supported General Magic option was the miniature gooseneck lamp I clipped onto the side: the backlights on these things, at least earlier units, burn out faster than they ought to. The unit in the case is my 2004 original and its backlight went out completely long ago, while the other used unit's backlight is down to about 60%. That's a problem because with their reflective screens it's hard to view it without the backlight in many lighting conditions and the gooseneck lamp is only a very imperfect mitigation.
One other trick the DR840 could do that the MP2100 couldn't was fold into its own stand, as the boxed unit demonstrates. The back of the unit has a flip-down face that the hinged front cover can hook into and works great on a desk. If you wanted a Newton like that, you probably wanted an eMate 300 instead.
Oh, yes: what about specs? Both the MP2100 and the DR840 have 8MB of ROM and 4MB of RAM, though the MP2100 also has an additional 4MB of on-board flash for user storage while the DR840 needs a card. On the other hand, while both have IrDA, serial and the two card slots, the MP2100 lacked a modem or rechargeable battery out of the box.That leaves the CPU, which General Magic still refused to describe as anything other than a "MIPS R3000 RISC." Neither the box nor the data sheet says exactly which R3000 it is, nor anything about its speed or cache; the MP2100, by contrast, is the fastest Newton with a very efficient StrongARM SA-110 at 162MHz with 16K cache each for instruction and data. Let's find out what the DR840 is actually packing.
Apollo's specification is described as three main chips, Dino (the CPU we're looking for along with a serial port controller [one to IR, one to the Magic Bus], MMU, RTC and LCD controller), Glacier (PC Card controller) and Betty (touchscreen digitizer, modem and sound input and output). With the exception of Glacier, however, Apollo was actually based on an off-the-shelf design. The codenames Betty and Dino (i.e., the Flintstones) are explained in an abandoned 1997 patent application from a small UK company called the Eden Group that sought to use the same chipset in a desktop system. In this diagram and in Eden's patent documentation, Betty is specified as a Philips UCB1100 "single chip integrated mixed signal audio and telecom codec" intended to be wired directly into the phone jack, mic and speaker, while Dino is specified as a 32-bit 40MHz R3000A-derived Philips PR31100 "Poseidon" core with 4K of instruction and 1K of data cache running the softmodem and operating system (but don't stop reading yet). Philips announced sales of the two-chip Betty and Dino system as early as May 1996. Our goal will be to identify all three of these chips as we go through the board. Fortunately these machines are very straightforward to work on, especially being portable devices. You can open it easily by just removing screws and very little of the innards is glued in. Take that, Apple (even Palm used adhesive with their batteries). The backside of the logic board is here, dominated by the PC Card cage and a layer of insulating material under it. The INV-TH-205(1) part in the upper left with a date code of 5th week 1998 is the inverter for the NEC-manufactured electroluminescent backlight. However, most of the rest of what we see are discrete components and nothing here looks obviously like the CPU, so we'll need to continue the hunt and turn the logic board over. To do that we'll remove some more screws as well as the negative wire pole of the backup battery (which is attached to the screw at the top left), disconnect the speaker cable (the red-black cable going under the backup battery compartment) and pull the connector out of the power daughterboard (between the backup and main battery compartments). We also disconnect the small ribbon cable at the lower left along the right edge of the main battery compartment so that the board can flip up and over to the right and not disturb the cables connected there. Turned over, we see two RAM chips here (both Hitachi 51W16160TT-6 16-bit 1MW units, yielding our four megabytes), two chips with Bowser silkscreened on them, and then a smaller chip at the top left. This smaller chip is Betty, labeled as a Philips UCB1100BE. The "Bowser chips" are actually twin Glaciers, each labeled © GMI JAPAN GLACIER-01 F840276. These were custom designs General Magic did themselves. There is one for each slot.That covers Betty and Glacier, but where on earth is Dino? There's only one place to look left, and that's under the PC Card cage. We release it on this side of the board by removing the four screws (don't lose the washers under them!).
With the screws out, we turn the board back over and pop the small edge connectors for the card cage out with a nylon spudger. That leaves the insulating material tacked onto the board, which we peel away. Removing the insulating material reveals three new chips. Two are the separate halves of the system ROM, here ordinary 4MB mask ROMs labeled OKI PIC31H — it's still a PIC — and OKI PIC31L. There are two DataRover 840 variants, the 840F and the 840, with the specific variant marked on the plate on the bottom (the 840s also lack the words "General Magic" under the Bowser logo on the front, but the name still appears on the backplate). The 840F is actually the earlier of the two based on serial number comparison and both of my used units, notably with the fading backlights, are 840Fs. In the 840F (F for "flash") the system ROM is stored in flash memory and can be overwritten, though since they both run the last Magic Cap version 3.1.2j there's nothing to upgrade to. My two new units including the board here are the later "just 840" in which the ROM is fixed.The third chip is a big 208-pin Toshiba QFP labeled TMPR3902U with a date code of 4th week 1998. Hey, we've finally found Dino!
Dino, at least as it appears in the DR840, is most definitely not a Philips core. Toshiba announced their own R3000-derived core, the R3900, in February 1995 with the first-cut "Southern Cross" TMPR3901F topping out at 400mW at 50MHz. Also with 4K instruction and 1K data cache, the TMPR3902U here is actually downclocked to 36.864MHz (from the 9.216MHz oscillator above it which the chip multiplies by 4), presumably for power consumption reasons.
Other than the ISA they implement, the R3900 and PR31100 cores have nothing in common. In fact, Toshiba claimed performance on par with the desktop R4000 in part from their proprietary fast multiply-add, which brings us back to General Magic and their MIPS-compatible SoftModem product. Naturally, GMGC used SoftModem for telephony in the DR840 and the custom TMPR3902U Toshiba fabricated for General Magic as a drop-in replacement enabled the DSP routines to run reliably at 19.2kpbs whereas the Philips core could only manage 14.4kbps. The PR31100 overall went poorly for Philips, who simply decided to abandon the effort and license the R3900 from Toshiba for their PR31500 and PR31700 cores which (Betty in tow) were much more popular with early Windows CE OEMs.
Nevertheless, a 36MHz R3000 — even a fast R3000 — doesn't sound all that hot compared to a StrongARM clocked over four times faster, so I can see why General Magic chose not to talk about it to avoid direct comparison. And, well, given that both had MSRPs of around $1000 ($1730 in 2022 dollars, $999 vs $1095 for the DR), the DR840 would have come out on the losing end for apparent value. Still, they probably didn't actually compete head-to-head in the market anyway as by now these were both mature product lines that people bought based on ecosystem rather than hardware. In regular use, the DataRover certainly doesn't feel slow. Magic Cap was such an efficient operating system from its several years in 68K purgatory that the interface is very swift on the 840, even though the clock speed was only double and change more as compared to the Dragon.
The DR840 started shipping in February 1998 along with a rapid forms development kit for application support, and vertical market agreements with health care and utility companies secured promising initial sales (expanding later to pharmaceuticals and, yes, public safety). Still, management felt the future lay in telephony and not in tablets, so in October another spinoff was in order: DataRover Mobile Systems, led by hardware division VP Steve Schramm, in which General Magic retained a 49% interest and to whom they licensed their IP and trademarks which still appeared on the individual units, boxes and documentation.
Although few native apps were widely distributed for the DataRover, and understandably given the standard Magic Cap desk set was fully ported and the 840 wasn't really an end-user product even though some end-users bought them, a smattering of games and network clients eventually surfaced including the promised DataRover-specific Web browser (picture taken from my "60% backlight" 840F):
I'm not using the modem here but rather a 3Com EtherLink III PC Card for which DataRover drivers are available. The DataRover browser doesn't do PNGs or SSL, but it does do tables, JPEG and GIF images (including animation), and it even can do old-school '90s-level JavaScript. Pages are actually rendered as self-contained scrollable documents which can be individually discarded into the trash truck. Unfortunately very large pages were not only hard on the modem but also on its use of working memory; Magic Cap 3.1 only boosted it from 512K to 768K. Configurable rules for the browser set page size limits not only to conserve data but also to avoid freaking out the operating system if it runs out of memory (flip the power switch twice to forcibly power it down and once more to trigger a warm start to recover; data is preserved).In 1998, though, this was probably more than adequate for basic work, and the company was proud of it. Every one of my four DataRovers from a factory reset has an inbox message "signed from" Steve Schramm, "VP & General Manager, General Magic" (dated January 1, 1998, prior to the DRMS spinoff) to "Get your Web Browser."
Interestingly, the client was delivered over E-mail from an automated address. The "Getting Started" button in the same message triggered a wizard to step you through opening the Getting Started help book in the Library. After all that work on making E-mail the centre of the Magic Cap universe, they weren't going to step away from it now! The DataRover was well received by users. Pen Computing raved about it in 1998 and said the 840 was "in contention for one of the best PDAs on the market today," citing its dramatically faster speed even over the PIC-2000 and the enhanced stability of the new version of Magic Cap. The Computerworld article above from May 1999 cites its selection over Palm and Windows CE by a Bay Area home builder because the company's inspectors needed a pen-based system that was sufficiently powerful and easy to use. The inspectors entered data into a bespoke application developed with a third party consultant, exactly the kind of usage originally envisioned.In March 2000 DRMS renamed itself to Icras ("EYE-kraz"), allegedly derived from the Latin root for "tomorrow" to represent "the company's rapid expansion to provide leading edge end-to-end wireless data solutions for commercial customers." Doubling down on their vertical markets strategy, Icras started shipping Ricochet and Nortel Merlin modems as out-of-the-box wireless options and emphasized remote data access more strongly, reworking the rapid development tool as the Remote Applications Kit and enlisting new third party middleware authors to create custom DataRover-based clients for customer backends. There was even talk about a future major release of Magic Cap.
However, Palm and handheld Windows CE devices were getting more capable and more available (and Microsoft under Steve Ballmer was getting more aggressive), and the almost exclusive use of DataRovers in highly specialized corporate settings meant virtually no consumer ecosystem to backstop it. Furthermore, because Rosemary devices weren't backwards-compatible, they could run none of the 68K Magic Cap applications that still existed. A U.S. Air Force technical report (PDF) on handheld computing devices for remote access summarized the DataRover as "limited for use in [handheld computing devices] within tight vertical markets in that it has very little third party support," and further observed that "Magic Cap, which had been thought of [sic] one of the best [handheld computing device] operating systems, has almost disappeared from the market." The Air Force report was publicly released right about the time Icras folded, in January 2001.
General Magic itself fared little better. Serengeti became the Portico product, launched August 1998, which supported E-mail, scheduling, contacts and news over voice commands (licensed from Nuance) and the Web. It was modestly successful with multiple vendors and up to 2.5 million users at its peak; Microsoft invested $6 million in the company in return for access to the company's IP. The voice interface to Portico itself, named Mary after Mary McDonald-Lewis who voiced it, became the basis of other products including the first iteration of General Motors' OnStar (i.e., the other GM). However, GMGC's stock price plunged again on weaker earnings by 1999 and most of the Portico magicians subsequently left or were laid off, leaving only a skeletal team to do further development. The company ceased operations in September 2002. Of the three major pieces of General Magic only AltoCom, the 1997 software modem spinoff, survived long enough to be acquired by Broadcom in August 1999 for $170 million.
The technologies and ideas introduced throughout the entire line were far ahead of their time — how many of these concepts, as radical as they were in the late 1990s, are now commonplace? Sadly, Magic Cap will probably never be open-sourced, meaning things like abandonware copies of the Windows runtime and the simulator, and of course any remaining real devices, are the only way you'll experience it now. As part of GMGC's liquidation in 2004 the IP portfolio (Icras never owned the marks or patents; they were always licensed) was sold primarily to former Microsoft CTO Nathan Myhrvold for his patent troll racket Intellectual Ventures. Myhrvold was chiefly interested in the patents underlying Telescript agents, but despite Andy Hertzfeld's best efforts, Myhrvold placed so many restrictions on Magic Cap's eventual usage that Andy gave up. As a poor substitute, let me regale you with a serenade instead:
Magicians went far and wide from General Magic throughout all of technology and one of them even ended up in a U.S. Presidential administration. Pop quiz: can you name these people? What else did they do and where are they now?
In our upcoming second Magic Cap post, we'll do some platform programming, and maybe even see what we can do to punch up the DataRover's modern day Internet mojo. Stay tuned.
My understanding is the 68349 is a 68000 derivative. It certainly lacks and MMU and FPU.
ReplyDeleteI was probably wrong. I was confusing this with the CPU in the Palm Pilot, which was a 68328.
ReplyDeleteYeah, the 683xx range is a little confusing. The Dragon I 68349 is technically a CPU32, which is closest to the '030 in microarchitecture (Motorola even called it the CPU030 in the datasheet), though you note correctly it has no MMU. However, some of the series, including the DragonBall family, are indeed 68000-derived.
DeleteThis is a remarkable post! Congratulations on collecting all this history in one place. Do you have a copy of this book? https://www.google.com/books/edition/Presenting_Magic_Cap/9S4rPQAACAAJ . If not, let me know with a reply here and I can send you one, I have extras. It describes Magic Cap's Construction Mode, roughly analogous to HyperTalk in HyperCard, but never shipped.
ReplyDeleteScott, I sure don't! Looks very cool. Let me know: ckaiser at floodgap dawt com
DeleteEmail sent!
DeleteDo you happen to have any information on the Datarover 2? Ages ago (I grew up and with all these devices in real time and was crazy about them!), I discovered a company called SET Engineering that put up a site about their work on the Datarover 2 with pictures and everything. But I can't find even a single reference anymore, until today, when I figured I'd try Google Bard and it hinted that https://www.set-engineering.com/datarover-2/
ReplyDeletedid at one time exist! Anyway, it's a mystery that's haunted me for years and years, so it would be cool to finally learn more...
No, sadly, I don't know anything about it. Wayback Machine has a single capture of http://web.archive.org/web/20021124175941/http://www.setengineering.com:80/datarover.htm (note different URL) and it does reference a DataRover II, but the pictures weren't captured so it's not clear if this is just a refurbishment or a completely new model. If the latter, it doesn't seem to have been released, and Josh Carter's site doesn't make reference to it. One more mystery to solve ...
DeleteOh man, thank you so much for finding that!!! You have no idea how much that has sat in the back of my head over the years, and I started to think it was just a dream or mixup with something else. Too bad the pictures are gone, because I can say for sure that what was on the site was a new version of the Datarover with a color screen!
DeleteAlso, it might be worth reaching out to some of the SET management with questions about what ended up happening with this device and if the prototypes still exist somewhere, to close the Magic Cap history loop once and for all!
DeleteDoesn't seem like it's the same company anymore. The colour question is interesting because the only colour version of Magic Cap I've personally seen is Magic Cap 95, and Josh doesn't really make any reference to a colour version on the handhelds. But it sure sounds cool! If I find out anything about it, I'll post a followup.
Delete