A new life for an original Portfolio

The Portfolio AI Interface connects a working 1989 Atari Portfolio to modern AI services while preserving its original keyboard, LCD and DOS environment. This is an external expansion, not a replacement computer hidden inside a vintage shell.

The Terminator 2 connection was impossible to resist: the assistant addresses the user as Operator and has a SKYNET-inspired personality. The voice is my own interpretation for this independent fan project, not an official film voice.

Working demonstration

Watch the complete build and demonstration on YouTube.

Architecture

Atari Portfolio running AI.COM ↔ original Atari Serial Interface ↔ MAX3232/SP3232 level converter ↔ Waveshare ESP32-S3-CAM-OV5640 V1.1 ↔ Wi-Fi and online services.

The Portfolio runs a small DOS client from an apfRAM memory card and supplies keyboard input, terminal display and command controls. The ESP32 handles network requests, camera capture, microphone recording, speech output and radio streaming. The language model and speech services are accessed online through the OpenAI API; they do not run locally on the Atari or ESP32. Internet access and your own API key are required, and API usage costs apply.

Serial and audio hardware

The serial link uses 9600 baud, 8 data bits, no parity and one stop bit. ESP32 UART transmit is GPIO43 and receive is GPIO44. The MAX3232/SP3232 converts between the Portfolio RS-232 signals and ESP32 logic levels, with TX/RX routed to the corresponding receive/transmit lines and a common ground. Do not connect RS-232 directly to ESP32 GPIO.

The tested board has two onboard microphones, an ES7210 microphone ADC and an ES8311 output codec. A separate speaker connects to the board. USB is for firmware upload and diagnostics; a PC is not required for normal operation. The ESP32 microSD slot must stay empty because GPIO43/44 are used by the serial link.

What it can do

  • Typed chat with text and spoken replies.
  • Microphone conversations with automatic listening.
  • Capture a camera photograph and ask the AI to describe it.
  • Display a monochrome camera preview, including a coarse mode.
  • Play synthwave, darksynth, chillsynth and vaporwave radio.
  • Pause background music for a spoken reply, then resume it.
  • Use a shared 0–100 voice/radio volume control and Shift + Up/Down shortcuts.
  • Configure Wi-Fi, credentials, location and saved audio options through a card text file or the terminal.

Design and development

I initially considered a Raspberry Pi, but the ESP32 camera board combined the camera, microphones and audio hardware in a compact module. The original computer still provides the interface.

The enclosure took many revisions to get right. Test prints exposed connector interference, awkward screw access and the need for better USB clearance. The finished design follows the appearance of the original expansion modules; even the speaker grille takes its shape from the Portfolio grille.

I tried a multicolour printed badge, but the fine lettering did not give the finish I wanted. The final module uses a photo-paper label attached to a thin printed backing. The Bambu Studio file has two plates: the dark case, lid, ESP32 support and backing on one, and white BOOT/PWR buttons on the other.

Software iteration focused on practical responsiveness: reducing the gap between text and speech, keeping radio controls responsive and coordinating music with spoken replies.

Build and setup

  1. Follow the repository build instructions for the exact supported board and connect through the RS-232 level converter.
  2. Copy AI.COM to a compatible Portfolio card.
  3. Copy the example configuration as AISETUP.TXT and enter your private network and API settings.
  4. Disconnect the card from USB before inserting it in the Portfolio. Never connect apfRAM USB while the card is inserted.
  5. Run AI, use /test for the serial link and /status for device status, then type a question. /help lists the available commands.

For battery operation, use a...

Read more »