GM Profiles

The AI Game Master narrates your journey but never changes the mechanics. All three profiles play the same game — only the storytelling voice differs.

How the narration arrives

The GM writes token-by-token. Each beat is composed live, word laid after word as the model generates it, threaded through the interface so the page fills as the story is told rather than appearing all at once after a pause. The first narrated turn loads the model and can take ten to thirty seconds; that is the model warming, not a hang. After that, the words come as fast as the model can write them.

The GM never bricks a run. It is an optional voice on top of a game that stands without it. If Ollama is unreachable, slow, or refuses, the trail tells its own deterministic story from plain consequences — you keep playing, you just read fallback text in place of the narration. When the GM has fallen back, the interface says so rather than leaving you to guess.

Profiles

Chronicler

Tone: Grounded, practical, spare.

The Chronicler reports what happened. Minimal folklore, no embellishment. Events are described plainly. Consequences are stated as facts. If you want the game to feel like a logbook, this is your narrator.

Best for: Players who want facts. Repeat players who already know the world.

Fireside

Tone: Serious campfire narrator. Subtle uncanny moments.

Fireside is the default profile. It tells the story like someone recounting their journey around a fire — mostly grounded, with occasional moments that feel slightly off. Not horror, not fantasy. Just the uneasy feeling that the valley has a memory.

Best for: First playthrough. Players who want atmosphere without spectacle.

Lantern-Bearer

Tone: Uncanny and liminal, but grounded in consequences.

The Lantern-Bearer narrates the same events but frames them through a stranger lens. The valley feels alive. Coincidences feel deliberate. Once the uncanny is unlocked (weirdness >= 2, see below), it leans into it harder than Fireside — but a Lantern-Bearer run at weirdness 0-1 is still plain grounded survival. The profile sets the voice; the weirdness gate decides whether the uncanny is on the table at all. Still grounded either way — no magic, no supernatural mechanics — but the tone is distinctly eerie when the gate is open.

Best for: Experienced players who want a different feel. Players who enjoy atmospheric tension.

Switching profiles

Set the GM profile at launch:

# Use the default (Fireside)
trail tui --seed 42

# Switch to Chronicler
trail tui --seed 42 --gm-profile chronicler

# Switch to Lantern-Bearer
trail tui --seed 42 --gm-profile lantern

Weirdness level

The --weirdness flag (0-3) sets the band the GM is allowed to play in. The default is 2. The uncanny is gated — it is not a smooth gradient that turns on a little at every level:

Weirdness	Band	What the GM may do
0-1	Grounded survival	Folklore stays rumor, omen, or coincidence. No uncanny-token spend — the run is plain survival.
2	Uncanny enabled	The GM may inject an uncanny hint (default). Spends an uncanny token.
3	Uncanny strong	The GM may inject a strong uncanny detail.

Each run starts with 2 uncanny tokens. A token is only ever spent at weirdness 2 or higher and while tokens remain; below the level-2 gate, no token is spent no matter the profile. So a Lantern-Bearer run at weirdness 1 is still fully grounded — the profile changes the voice, but the level-2 gate decides whether the uncanny is on the table at all.

# Maximum uncanny (Lantern-Bearer recommended)
trail tui --seed 42 --gm-profile lantern --weirdness 3

# Grounded survival, no uncanny-token spend
trail tui --seed 42 --weirdness 0

Choosing a model

The --model flag selects which Ollama model generates narration. The default is llama3.2. Larger models produce richer text but are slower:

trail tui --seed 42 --model mistral

Voice narration

Enable spoken narration with --voice. Each GM profile maps to a different voice and pacing style. Control playback speed with --voice-pace:

trail tui --seed 42 --voice --voice-pace slow

Voice pacing options: fast, normal (default), slow. Spoken narration requires the voice extra, which pulls in the voice-soundboard package:

pip install "escape-the-valley[voice]"

Audio plays in the background and can be interrupted with any key press.

Disabling the GM

Run without AI narration entirely:

trail tui --seed 42 --gm-off

The game plays identically — you just see fallback text instead of GM narration. Useful for speed runs or systems without Ollama.

Requirements

The GM requires Ollama running locally. No API keys, no cloud calls, no cost. The game auto-detects whether Ollama is available and falls back gracefully if it isn’t. The default 30-second timeout allows for first-load model warm-up on slower hardware.