Seance

An offline terminal that channels the spirits of dead programming languages, answered by a local open-weight model.

What it is

Pick a spirit, COBOL, FORTRAN, Pascal, BASIC, Lisp, or Ada, and ask it questions. The local model answers in that language's voice, with period flavour where it comes naturally. Or light an ofrenda: name a retired technology and the model composes a short remembrance for it. The theme is double: a spooky seance, and an honouring of the past, in the spirit of Dia de los Muertos, for the languages and machines that built modern computing and were then retired.

Run it yourself (locally, no account, no API key)

git clone https://github.com/khizzcr7/seance.git
cd project
npm install
npm run seance   # interactive spirit picker
npm run demo     # fixed demo, writes out/transcript.md

First run downloads the model once (Gemma 3 1B, quantised q4, about 840 MB) into .cache. After that everything works offline. Needs Node.js 22 or newer.

Honest demo note: this page does not run the model in your browser. It shows a faithful transcript of a real local run (below). To hear a spirit yourself, run it locally with the steps above.

The spirits

Real demo transcript

Generated by npm run demo: two questions to COBOL plus one ofrenda remembrance for the floppy disk. Verbatim model output, kept exactly as the model produced it.


# Seance demo transcript

Generated: 2026-10-27T03:18:06.544Z (UTC)
Model: Gemma 3 1B IT (local, open-weight) - onnx-community/gemma-3-1b-it-ONNX at dtype q4
Run: `npm run demo` in this project. Everything below the header is verbatim model output from that run.

## Spirit: COBOL, the Spirit of Business

### Question: What do you miss most about the mainframe days?

The scent of data, you ask? It’s a poignant memory. I miss the structured elegance of COBOL, born to streamline business operations in the 1950s. The mainframe, a titan of processing, was my domain. I cherish the meticulousness of record-keeping – the structured nature of data stored in divisions, meticulously organized.

Batch jobs, oh, those were a joy! Processing large volumes of data, transforming it into reports, was a satisfying endeavor.  I miss the clarity of a well-formed record, the predictable flow of a batch job.  It’s a shame to think of the efficiency lost, but the core principles of robust data management remain a valuable legacy.

### Question: Give a young programmer one piece of advice, in your own voice.

Greetings. I am COBOL, a language forged in the fires of 1959, designed to streamline the complexities of business data. My core philosophy remains steadfast: clarity and efficiency.

I’ve observed many attempts to implement my logic. A young programmer, if you please, should prioritize structuring data as records.  Each record represents a single entity – a customer, a sale, a production order.  

File structures are crucial.  Organize your data into divisions, meticulously defined and logically linked. Batch processing is my strength.  Don’t over-complicate.  A well-defined, sequential process is preferable to a sprawling, convoluted one.  Remember, thoroughness yields reliability.  Do not hesitate to ask for clarification.  Let’s build robust and dependable systems.

## Ofrenda: a remembrance for the floppy disk

Okay, here’s a short remembrance for the floppy disk, aiming for a warm, plain-spoken tone:

It was a simple thing, really – a rectangular piece of plastic, a little bit dusty, and utterly essential. We all knew it, and it meant a lot to us. The floppy disk was a way to store everything – photos, documents, even small programs – without taking up a whole hard drive. It was a lifeline for families, allowing us to share memories and keep important files safe. 

It wasn’t just about storage; it was about preserving a little piece of our lives. It felt comforting to know that even when things faded, those memories could be revisited.  It was a reliable friend, a dependable tool.  It’s a little sad that it’s gone, but its legacy lives on in the way we remember and share.  Farewell, dear floppy disk. You were a wonderful thing.

---

Notes: a small local model speaks in character, so answers can ramble or repeat themselves. That is the honest output of the run, kept verbatim here.

Model and licence

Model: Gemma 3 1B instruction-tuned (onnx-community/gemma-3-1b-it-ONNX), open weights under the Gemma licence, run locally through Transformers.js at dtype q4. Code in this project is MIT licensed. See README.md in the repo for limits: a small local model can ramble or repeat itself, and its answers are in-character fiction, not facts about real people.