Wyrdlyre

How it works

Three steps between a thought and a story

No typing. No menus. Just a held button, your voice, and a DM who actually listens.

You speak

Hold the button, say what your character does. The app runs on any laptop, phone, or tablet in the room.

The DM listens

Your words are transcribed locally, the scene is understood in context, and a response is written in real time — in the mood you chose for the campaign.

The story speaks back

Narration in one voice, each NPC in their own — warm innkeeper, gruff watchman, cautious ranger — all at the table with you.

Built for your table

Your friends, their phones, one story

Wyrdlyre runs on the laptop in the middle of the table, and every other player joins from their own phone with a shared URL. Each player has their own push-to-talk, their own voice, and the DM knows exactly who said what.

No shared mic, no sign-ups, no accounts for the guests — the session is yours, the code is yours, the door closes when you close the lid.

Friends around a table with phones, warm candlelight, D&D character sheets

When there's no one else around

Or bring your own party

Add AI companions to your party and the DM voices them for you — each one with their own personality, their own voice, their own reasons to speak up when the moment calls.

The stoic ranger you've traveled with for three sessions will sound the same in session four. She'll argue with the weary monk. She'll warn you when she thinks you're being stupid. She might even be right.

Lone player at a table with translucent glowing AI companion figures

Every campaign, its own voice

Five moods, baked into the DM's delivery

Pick the tone of your campaign once. The DM's narration, the pacing, the pauses, even the quality of the silence — all adjust to match.

Moonlit graveyard with ruined cathedral, fog, raven on broken arch

Gothic Horror Hushed, deliberate, undercurrent of unease

Sunlit hilltop at dawn, medieval castle with banners, winding road

Heroic Fantasy Warm, earnest, measured reverence

Torchlit stone chamber with treasure, skeleton fragments, shafts of light

Pulp Adventure Energetic, theatrical, varied pacing

Plus Gritty Low-Fantasy and Saturday Morning Cartoon, for when the mood calls.

Dice you can trust

The DM never rolls for you

Every roll in Wyrdlyre comes from a deterministic dice engine — plain Python, operating-system entropy, fully audited per-die. When the DM asks for a Dexterity save, the number on your screen is the real one. Not a number Claude decided would be dramatic. Every result is inspectable, every roll is yours.

Three ornate antique keys in candlelight with glowing runes

Your keys, your control

No margin on inference. Just the software.

Wyrdlyre is software, not a subscription on top of AI calls. You bring your own Anthropic, OpenAI, and (optionally) ElevenLabs keys — the ones you already have, or the ones you sign up for in two minutes — and you pay those providers directly for what you use.

No per-minute meter. No surprise bills. Heavy use doesn't cost us more and it doesn't cost you more than what the providers charge. If you'd rather run local, plug in Ollama and pay nothing per session.

What you pay us

One flat license for Wyrdlyre. Set at launch.

What you pay providers

Your own Anthropic + OpenAI usage. Pennies per session.

Your real cost

Predictable. No meter. No one marking up the AI on you.

Shipping at launch

What you get in v1

D&D 5e rules via the Creative Commons SRD 5.1 — 12 classes, 9 races, ~300 spells, ~100 monsters bundled.
Party play across phones, tablets, laptops — live sync, per-player voice recognition.
Solo mode with AI companions who react, argue, and voice themselves in character.
Per-NPC voices — 11 OpenAI voices to choose from, plus optional voice cloning via ElevenLabs (premium).
Five mood presets — Heroic, Gothic, Pulp, Gritty, Cartoon — each shaping narration and delivery.
Real dice, audited per-die, never LLM-generated.
Session transcripts — export as Markdown or JSON whenever you want to share a highlight.
Self-host via Ollama for zero cloud cost.
Local-first — campaigns, keys, and audio never touch our servers.
BYO rulebook mode for non-SRD editions is planned for v2 (AD&D 1e/2e, 3.5, 4e via your own PDFs).

Where we are

Pre-alpha — feature complete for solo testing

The full voice loop is built. A player holds push-to-talk, speaks, and the DM speaks back with distinct voices for each NPC. Party-NPC solo play, multi-device WebSocket sync, per-NPC voice assignment, transcript export, and a real dice engine all work end-to-end.

We're currently running our own AI testing harness against it — an automated player plays sessions overnight, an audio judge listens to the output and scores it — before we recruit human testers. If you'd like to be one of those testers, drop your email below.

Questions we get asked

FAQ

What does a session actually cost me?

At current provider pricing (2026), expect roughly $0.30–0.80 per hour of play in AI calls across Anthropic and OpenAI. That's the whole bill — no markup from us. Lighter campaigns (text-first or local Ollama) are effectively free after setup.

Do I need to be technical to use this?

Not to play — once it's installed, it's a voice app. You do need to get two API keys (Anthropic and OpenAI) from their dashboards and paste them into the onboarding wizard. The app walks you through it with screenshots. "Test connection" buttons confirm each key works before you can proceed.

Which editions of D&D are supported?

D&D 5e at launch, via the Creative Commons SRD 5.1. For 1st Edition AD&D, 2e, 3.5, and 4e — since those rulebooks are still under copyright — Wyrdlyre's planned v2 "BYO rulebook" mode will let you upload your own PDFs and the DM will reference them during play.

Can I really play solo?

Yes. Add AI companions to your party during campaign setup; each gets a voice and a personality. When you enter a scene, they speak up — react to what's happening, suggest approaches, argue with each other. Many hours of single-player D&D have already been played this way during development.

Do you see my campaigns or my voice?

No. Wyrdlyre runs entirely on the host machine. Your API keys live in your operating system's keychain, your campaign database is a local SQLite file, and raw audio is transcribed and discarded — not persisted unless you explicitly turn on session recording. We don't have a server that receives your gameplay. We don't have a server at all.

What's the system requirement?

Any machine that can run Python 3.13+ and Node 20+. Windows, macOS, or Linux. No GPU required — the AI inference runs on the cloud providers (or on a separate Ollama box, if you want to self-host). The host machine's only job is orchestrating the session, which is cheap.

When will it launch?

When it's ready. We're running automated tester sessions against it right now and tuning the DM prompt based on the scorecards. Human playtesting is the next milestone. Early access signups below get first invites.