Can I paste my own production logs into Ghost in the Log?

No. The public version only lets you pick from three fictional incidents. There is no free-text field, and only the selected incident ID and language are sent to the AI.

Which AI model does the public version use?

It uses the open-weights model Muse Spark Open, called through Baseten's hosted inference API. An earlier test ran Gemma on Modal, but per-call GPU startup costs were higher.

What limits does the app place on AI calls?

Before each call it takes a reservation in Neon, a shared database, and records it when done. If no reservation is available, it does not send, and it will not resend automatically when a result is unclear.

Open-Source AIZenn AI/ML46m agoOct 31, 2026, 06:00 JST

Ghost in the Log: an AI log game that takes only fictional 3件

Send on LINE

3 Key Points

What happened
The developer released Ghost in the Log, a Halloween-themed game where players investigate three fictional incidents and the AI returns candidate causes, supporting lines, and open questions.
Why it matters
Players practice separating evidence from plausible-sounding story, since the game gives only a fixed set of fictional incidents and the AI outputs candidate causes rather than confirmed diagnoses.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Get these free every morning

Context & Analysis

Ghost in the Log grew out of a concrete worry: log-reading tools tend to invite pasting real production logs, and those logs often carry secrets or personal data. The developer narrowed the entry point instead—only three fictional incidents are selectable, and no real logs can be entered—so the public version could still run AI behind the scenes while keeping inputs fixed. The build sequence shows the same caution. An AI-free demo went live on October 29, data moved to Neon and hosting to Vercel on October 30, and only then did the developer open access after a gated test confirmed the model returned causes, supporting lines, and open questions from fictional database logs. The model path changed too: an early test ran Gemma on Modal, but per-call GPU startup added time-based cost, so the public version moved to calling an open-weights model through an API. The developer is explicit that these measures guard against unexpected input and runaway calls, not against wrong answers.

FAQ

Can I paste my own production logs into Ghost in the Log?: No. The public version only lets you pick from three fictional incidents. There is no free-text field, and only the selected incident ID and language are sent to the AI.
Which AI model does the public version use?: It uses the open-weights model Muse Spark Open, called through Baseten's hosted inference API. An earlier test ran Gemma on Modal, but per-call GPU startup costs were higher.
What limits does the app place on AI calls?: Before each call it takes a reservation in Neon, a shared database, and records it when done. If no reservation is available, it does not send, and it will not resend automatically when a result is unclear.

Zenn AI/MLRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.