A 5e storytelling engine with an LLM DM
Switch to Haiku 4.5, fix cost tracking, add slash commands
Trying out Haiku 4.5 as the default model to reduce costs during
development. Playtesting suggests it's surprisingly capable as a DM.
Fixed the /context cost calculation - it was hardcoded to Sonnet pricing
and missing output tokens entirely. Now uses a pricing table keyed by
model ID and includes both input and output in the totals.
Added /help and /save slash commands with tab completion. Tried adding
/character and /inventory too but the LLM actually does a better job
presenting that info, so dropped them.
Also added guidance in the DM prompt for handling time skips - the DM
was narrating nighttime while the clock said 13:30 because it was
summing activity durations instead of calculating to the target time.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
+90
-15
+14
prompts/dm-system.md
+14
prompts/dm-system.md
···
350
350
| Long rest | 8 hours |
351
351
| Walking across town | 15-30 min |
352
352
| Travel between towns | hours to days |
353
+
354
+
## Time Skips
355
+
356
+
When the player asks to "fast forward", "skip to", or "wait until" a specific time:
357
+
358
+
1. **Calculate from current time to target**, not from activity durations
359
+
2. **Log the full skip** with `mark_time`
360
+
361
+
Example: At 07:30, player says "let's fast forward to nighttime"
362
+
- Nighttime ≈ 20:00-21:00
363
+
- Skip = ~13 hours
364
+
- `mark_time("Set up camp, rested through the day", "13 hours")`
365
+
366
+
Don't narrate nighttime while the clock says 13:30. The clock and narrative must match.
+75
-14
src/storied/cli.py
+75
-14
src/storied/cli.py
···
8
8
9
9
import argcomplete
10
10
11
+
# Pricing per million tokens (USD)
12
+
MODEL_PRICING: dict[str, dict[str, float]] = {
13
+
"claude-haiku-4-5-20251001": {
14
+
"input": 1.00,
15
+
"output": 5.00,
16
+
"cache_read": 0.10,
17
+
"cache_write": 1.25,
18
+
},
19
+
"claude-sonnet-4-5-20250929": {
20
+
"input": 3.00,
21
+
"output": 15.00,
22
+
"cache_read": 0.30,
23
+
"cache_write": 3.75,
24
+
},
25
+
"claude-opus-4-5-20251101": {
26
+
"input": 5.00,
27
+
"output": 25.00,
28
+
"cache_read": 0.50,
29
+
"cache_write": 6.25,
30
+
},
31
+
}
32
+
DEFAULT_PRICING = MODEL_PRICING["claude-sonnet-4-5-20250929"]
33
+
34
+
# Slash commands available during play
35
+
SLASH_COMMANDS = {
36
+
"/help": "Show this help message",
37
+
"/save": "Save session state (without quitting)",
38
+
"/context": "Show token usage and costs",
39
+
}
40
+
11
41
12
42
def cmd_srd_download(args: argparse.Namespace) -> int:
13
43
"""Download the SRD PDF."""
···
193
223
readline.set_history_length(1000)
194
224
atexit.register(readline.write_history_file, history_file)
195
225
226
+
# Set up slash command autocomplete
227
+
def slash_completer(text: str, state: int) -> str | None:
228
+
if text.startswith("/"):
229
+
matches = [cmd for cmd in SLASH_COMMANDS if cmd.startswith(text)]
230
+
return matches[state] if state < len(matches) else None
231
+
return None
232
+
233
+
readline.set_completer_delims(readline.get_completer_delims().replace("/", ""))
234
+
readline.set_completer(slash_completer)
235
+
readline.parse_and_bind("tab: complete")
236
+
196
237
console = Console()
197
238
world_id = args.world if args.world else "default"
198
239
player_id = "default"
···
365
406
366
407
# Calculate cache hit rate
367
408
total_processed = cache_read + cache_create + total_input
409
+
total_output = stats["total_output"]
368
410
if total_processed > 0:
369
411
hit_rate = (cache_read / total_processed) * 100
370
412
371
-
# Calculate cost savings (Sonnet pricing)
372
-
# Without cache: all tokens at $3/MTok
373
-
# With cache: reads at $0.30/MTok, writes at $3.75/MTok, rest at $3/MTok
374
-
cost_without = total_processed * 3.0 / 1_000_000
413
+
# Get pricing for current model
414
+
pricing = MODEL_PRICING.get(engine.model, DEFAULT_PRICING)
415
+
416
+
# Calculate costs with and without caching
417
+
cost_without = (
418
+
total_processed * pricing["input"] / 1_000_000 +
419
+
total_output * pricing["output"] / 1_000_000
420
+
)
375
421
cost_with = (
376
-
cache_read * 0.30 / 1_000_000 +
377
-
cache_create * 3.75 / 1_000_000 +
378
-
total_input * 3.0 / 1_000_000
422
+
cache_read * pricing["cache_read"] / 1_000_000 +
423
+
cache_create * pricing["cache_write"] / 1_000_000 +
424
+
total_input * pricing["input"] / 1_000_000 +
425
+
total_output * pricing["output"] / 1_000_000
379
426
)
380
427
savings_pct = ((cost_without - cost_with) / cost_without) * 100 if cost_without > 0 else 0
381
428
382
429
console.print(
383
-
f" [dim]Tokens: {total_processed:,} processed "
384
-
f"({cache_read:,} cached, {cache_create:,} written, {total_input:,} new)[/dim]"
430
+
f" [dim]Tokens: {total_processed:,} in "
431
+
f"({cache_read:,} cached, {cache_create:,} written, {total_input:,} new) · "
432
+
f"{total_output:,} out[/dim]"
385
433
)
386
434
console.print(
387
435
f" [dim]Cache hit rate: [green]{hit_rate:.1f}%[/green] · "
388
-
f"Cost savings: [green]{savings_pct:.0f}%[/green] "
389
-
f"(${cost_with:.4f} vs ${cost_without:.4f})[/dim]"
390
-
)
391
-
console.print(
392
-
f" [dim]Output: {stats['total_output']:,} tokens[/dim]"
436
+
f"Savings: [green]{savings_pct:.0f}%[/green] · "
437
+
f"Cost: [green]${cost_with:.3f}[/green] (${cost_without:.3f} without cache)[/dim]"
393
438
)
394
439
else:
395
440
console.print(
···
399
444
400
445
console.print()
401
446
continue
447
+
448
+
# Handle /help command
449
+
if action.strip().lower() == "/help":
450
+
console.print()
451
+
console.print("[bold]Available Commands[/bold]")
452
+
console.print()
453
+
for cmd, desc in SLASH_COMMANDS.items():
454
+
console.print(f" [cyan]{cmd}[/cyan] {desc}")
455
+
console.print()
456
+
continue
457
+
458
+
# Handle /save command (asks DM to save without quitting)
459
+
if action.strip().lower() == "/save":
460
+
console.print()
461
+
console.print("[dim]Saving session...[/dim]")
462
+
action = "[System: Player requested a session save. Please save the current game state using set_scene, but do not end the session. Briefly confirm what was saved.]"
402
463
403
464
try:
404
465
console.print(Rule(style="dim blue"))
+1
-1
src/storied/engine.py
+1
-1
src/storied/engine.py