What is it

The Sharp PT-9000 (development / code name "Bullet") is a PC/XT compatible tablet computer, first announced in late 1993 and becoming available as early as May 1994. (1) Marketed by Sharp at the industrial / vertical market, instead of consumers and offices, by all accounts it was a commercial failure and shipped very few units. (2)

The PT-9000 hardware is built around the Vadem VG230 (sometimes stylized as VG-230) system-on-chip, packaging practically an entire XT-compatible system into a single chip, including PCMCIA support and APM power management. (3)

Why is it so rare?

The PT-9000 was never available for sale to the public. Sharp attempted to market and sell it almost exclusively through industrial channels and vertical markets (1.4). I would speculate that at least some of the units circulating in the hands of end-users from 1994 through 2000 or so were originally purchased by software developers working on the GeoWorks platform.

Inexplicably, the retail-looking packaging that I received mine in focuses on the GeoWorks desktop applications and Palm PIM features. These would have been relevant features for end-users of the day looking for an enhanced PDA experience, but not relevant for vertical markets, who would likely be writing and installing their own custom software.

No one knows exactly how many units were made and shipped. My unit's serial number ends in 0011. The Computer History Museum has a PT-9000 in their collection with a serial number ending in 0131. Based on this and other anecdotal reports, it seems probable that there were less than 1,000 sold in total. (2)

How was it used?

It's not known if there were any units actually sold through vertical channels. The form factor lends itself to a variety of potential use cases:

• The screen cover folds into a stand, making it possible to place on a desk and use with the pen or an external keyboard.
• The case has anchor points on two of the corners where a lanyard or shoulder/neck strap could be attached. This would facilitate simple use in the field for a variety of industries.
• With 6 AA batteries installed, the unit weighs 1.1kg (2.4 pounds). Though lighter than other similar devices of the day, holding the device unaided with one arm is not feasible for extended periods of time.

Photos

Repair and Restoration Journal

• November 2025
• January 2026
• 5 February 2026

November 2025

I acquired this machine along with a number of other PDAs from a local collector's estate. It's in excellent external condition - not a scratch or scuff on the casing or screen. The rubber covers over the keyboard and power ports are still pliable and haven't deterioriated - remarkable for a 30 year old computer. It included the original box, printed manuals, registration cards and other pamphlets, an extra stylus, and a still-wrapped 3.5" floppy disk containing GeoHost 1.0. The serial number on the unit and the box match - 49010011N.

After installing fresh batteries, the machine boots DOS and then automatically launches GeoWorks, landing on the pen calibration screen. Some initial testing uncovered some issues:

1. Half of one column of pixels on the LCD stays off.
2. The resistive touchscreen refuses to pass calibration. Well, I got it to pass once after retrying the calibration routine for 10 minutes, but then it was obvious that the coordinates being read are not correct - and occasionally register touches on the screen when none have been made.

At this point, I disconnected the flex cable for the resistive touch screen, cleaned it (though there was no evidence of corrosion) and reconnected it. Unfortunately, that didn't improve the situation. I dug deeper into the hardware and managed to disassemble to the point where I could remove the mainboard and its attached daughterboards - one containing the ROMs, and the other one an unknown function - just to photograph them. But at this point, I had no XT-compatible keyboard to use to otherwise make any other inputs, so I left the machine alone over the holiday season.

January 2026

I returned to the machine in mid-January determined to get past the pen calibration issue. My plan was to interrupt the boot, modify the GeoWorks configuration to skip the pen calibration routine, and configure a serial mouse as a pointing device. At least then, I'd be able to use the built-in software in some way. Ultimately, I'd like to dump the entire ROM.

Not having an XT compatible keyboard since about 1995, after some research I found the ps2pico project, for interfacing a USB keyboard to an XT or AT keyboard interface using a Raspberry Pi Pico. Thankfully, I had all of the parts on-hand and was able to whip one up quickly, and it works perfectly.

add photo here

With a working keyboard, I could interrupt AUTOEXEC.BAT with a quick Control-C, but I also discovered that GeoWorks intercepts the Ctrl-Alt-Del keystroke, and quits back to DOS. So I could boot the machine all the way to the GeoWorks pen calibration, and then quit cleanly (relatively) back to DOS. Success!

Buoyed by this success, I set about poking around the system and performing a deep recall of my DOS knowledge from my youth, collecting utilities, and performing research about GeoWorks. My first IBM compatible computer was an XT clone that I used for many years, so I'm well versed in the basics of the platform. That said, the PT-9000 being a ROM based machine means that its boot sequence and storage is organized in an interesting (to me) way:

• C: presents as a small (~ 1MB) FAT disk from ROM containing Datalight ROM-DOS 3.31 (very stripped down), PC Card Services drivers, tools for formatting flash cards, the GeoWorks loader and initial configuration, and scripts for setting up the RAM Disk.
• The AUTOEXEC and CONFIG files on C: automatically initialize Card Services, check the RAMdisk on D: (formatting it if necessary), and then execute D:\AUTOEXEC.BAT.
• D: is the RAM disk. When the machine is reset, the RAM disk is formatted, and basic files are copied from C: to D: - a short AUTOEXEC.BAT that starts GeoWorks, and a GEOWORKS folder containing LOADER.EXE, a few drivers (one of which is BFS.GEO), and a short GEOS.INI that initially invokes the first-run setup. The rest of GeoWorks is configured from an included INI file stored on the C: ROM.
• E: through L: are assigned to drives inserted as PC Cards, depending on their type.

I've been able to use a 16MB CompactFlash card with a PCMCIA adapter to shuffle files back and forth from my modern machines, which is necessary since the included ROM-DOS comes with very few utilities - not even an editor.

I discovered has two daughtercards on the underside. One definitely has the IPL and main ROM chips, and the other one has quite a few electrolytic capacitors on it. I haven't removed the board to look at them closely, but with a 32 year old machine I'm highly suspicious of any electrolytic. With a keyboard, most of GeoWorks can be navigated. Except for the Preferences application - which has big buttons to launch each specific preference panel, which cannot be selected using the Tab or Enter key. Thankfully, the pen cooperated enough one day that I was able to configure the second mouse pointer. I plugged in my old MS serial mouse clone and ... nothing.

I checked the settings, I checked what was written to GEOS.INI, I verified that the mouse still worked on my Linux machine using a USB-Serial adapter, and it does. Very confused, I downloaded the Cutemouse driver and its mouse-testing tools, specifically COMTEST and PROTOCOL. Running COMTEST gave almost-expected results: one serial port detected, standard port and IRQ, but most of the time it thinks there's a modem attached when there definitely isn't. And running PROTOCOL, which reads mouse data packets and prints them to screen, I discovered that the serial port was returning absolute trash data - with or without the mouse connected. As a final confirmation, I loaded up some DOS terminal software (good ol' Telix!), and connected a null modem cable from the PT-9000 to my USB-serial adapter on my Linux machine. The Rx light on the adapter immediately started blinking before I did anything else, and I was unable to send or receive any data.

So add a third issue to the list - the serial port isn't working either.

But remember that there are two daughterboards inside? One clearly had the BIOS ROM and the 6MB ROM containing ROM-DOS and GeoWorks. I can't see all of the components of the other boards without removing it from the mainboard, but peeking inside the board sandwich, it has a large number of electrolytic capacitors on it. My working theory is that this second board performs a bunch of power filtering and conversion, and that some of those electrolytics are dried out and malfunctioning. So my next step is to disassemble the machine, remove this board, inspect it and/or replace the capacitors. Hopefully that will solve both of my current issues, but with vintage computers, you never know!

5 February 2026

So, I disassembled the machine once more, removed the board assembly and disconnected all of the daughterboards, and took some high-resolution photos of all the boards with the 50MP camera on my phone: Side 1 and Side 2. (The foam block in the first photo was a paid actor.)

I gave both sides of the board a good visual inspection and noticed some suspicious looking items:

1. Some brownish looking areas where the DE-9 connector pins are soldered to the board
2. The power board has nine aluminum electrolytic capacitors on it, all of which have some suspicious looking whitish residue on the top of the cap. They also don't appear to be vented like others that I've seen before - there are no lines in the metal where the cap will bulge when it fails. Photos I've seen of these capacitors in known good conditions do not have the white stuff on top.

References

1. Multiple sources:
   • Earliest mention I can find of the PT-9000's code name "Bullet".
   • comp.sys.pen message stating that the PT-9000 would be out in April.
   • comp.os.geos message about the GEOS 2.0 SDK, released on May 23, 1994, including support for the PT-9000.
   • comp.sys.pen message from May 23, 1994 stating that a few PT-9000 units are "out and about" and that the target marked is industrial/verticals.
   • Announcement of a meeting of the Bay Area GEOS Developers' Group meeting in June 1994, mentioning that the PT-9000 will be demonstrated.
2. comp.os.geos.misc message from Steve Main, former employee of GeoWorks, mentioning that Sharp only shipped "a few dozen" PT-9000s
3. Specifications for the PT-9000 as gleaned from a Sharp brochure that seems to have been available via fax in the 1990s. Also, this photo which is probably from the same source.