The Timekeeper's Journal

Engine Update: Choose Your Narrator

Architects can now choose the AI that tells their story: The Chronicler (Gemini), The Scholar (Claude Haiku) or The Veteran (Qwen). Here is how they differ, and why we didn't pick the biggest models.

The Timekeeper · 2026-10-09

When you sit down to play a tabletop campaign, the personality of your dungeon master defines everything. A gritty, tactical narrator runs a completely different session from an introspective storyteller who dwells on character psychology and period-accurate prose.

Until now, Live Through Time has relied on Google's Gemini Flash family for storytelling. It has been a phenomenal workhorse: fast, vivid and deeply cinematic. It remains the bedrock of the platform, the default for every story, and available to all players.

But different genres demand different voices. A tense gangland shakedown in 1952 Chicago shouldn't sound like an academic investigation in Victorian London.

"The universe is made of stories, not of atoms." — Muriel Rukeyser

Today we're rolling out a major update for Architect subscribers: narrator selection. You can now choose the AI voice that narrates your campaign.

The narrator roster

The Chronicler (Gemini Flash): the default

The cinematographer. Fast, vivid and highly responsive to action. It excels at quick pacing, cinematic scene changes and sweeping world events, and it continues to power the game for everyone, on the web and on mobile.

The Scholar (Claude Haiku): Architect

The literary historian. Known for subtle, introspective prose, rich inner lives and exceptional nuance in dialogue. If you're playing an intellectually demanding mystery or a delicate diplomatic timeline, The Scholar brings unmatched atmosphere.

The Veteran (Qwen 80B): Architect

The street-level realist. Punchy, unvarnished and tactically grounded. The Veteran cuts through flowery prose to deliver hard-boiled dialogue, gritty texture and visceral realism. It feels right at home in the trenches of 1917 or the criminal underworld of the 1950s.

Why not Sonnet?

The question we always get is: "Why not Claude Sonnet, or the other big flagship models?"

We tested larger models extensively while developing the engine. Their prose is undeniable, but three things made them impractical for live play:

  • Waiting breaks immersion. In an interactive story, pacing is everything. Waiting 20 seconds or more for a heavyweight model to take its turn shatters the tension.
  • Haiku hits the sweet spot. Modern versions of Haiku punch well above their weight, with nuance, character depth and atmosphere remarkably close to Sonnet, at near-instant speed.
  • Fair pricing. We wanted the Architect tier to stay an accessible $9.99 a month on the web for 1,000 deep-context turns, rather than pushing players into predatory token packs or draining their wallets over a weekend binge. Running flagship models at that scale would have made a flat subscription impossible.

Under the hood, your game is just as solid whichever narrator you choose. Your chosen model writes the story, while our fast background engine keeps the Ledger and the Librarian up to date behind the scenes.

Which models would you like to see added next, and what narrative voice do you feel is missing from AI RPGs? Tell us on Discord.

In the manual: AI Narrators · Turns & Access Tiers

Thanks for reading. See you in the past…

The Manual