This is a submission for the DEV OpenClaw Challenge 2026: OpenClaw in Action
What I Built
Clawdvent — an AI Dungeon Master that lives inside Discord and runs full D&D-style tabletop adventures with live AI-generated scene art.
Players type actions in chat. Clawdvent narrates cinematically, rolls dice, tracks characters, and illustrates every scene using DALL-E 3 — dungeons, taverns, battle maps, NPCs. All inline. All in real-time.
The entire D&D "engine" — personality, combat rules, scene pacing, dice mechanics, visual aesthetic — lives in one markdown file called SOUL.md.
No training. No fine-tuning. Zero bot framework code. Just OpenClaw + thoughtful prompt engineering at the system level.
Demo
Try it live (in the Clawdvent Test server): [Coming soon - private server]
Real session screenshots:
Session Zero — the hook:
The air smells of oak smoke and old rain. The Drunken Dragon tavern is warm, noisy, and suspiciously full of masked strangers. A hooded figure at your table slides a folded parchment across the wood. It's wet. It's marked with a single word, written in something darker than ink: "Help."
What do you do?
Character creation + scene image:
Player: "I'm playing a human rogue named Ember. Investigate the hooded figure and generate an image of the tavern scene."
Clawdvent responds with a full AI-generated tavern illustration (dim candlelight, the hooded figure, masked patrons), then narrates Ember leaning in to investigate.
Combat with dice rolls:
Player: "I attack the hooded figure with my dagger!"
Clawdvent:
You leap into action, your dagger gleaming in the dim light...
[🎲 d20: 15 + dexterity(4) = 19 vs DC 12 — HIT]Your blade pierces the air, striking true. The hooded figure lets out a startled gasp. As the hood falls back, you see a face full of fear — a frightened young woman who appears to be no threat at all.
Tactical battle map (auto-generated):
Player: "Generate a top-down tactical battle map for this combat."
Clawdvent generates a gorgeous isometric tavern battle map with character tokens, tables, and hooded figures in combat positions.
Code
shopsmartai
/
clawdvent
AI Dungeon Master in Discord with live scene art. Built on OpenClaw for DEV OpenClaw Challenge 2026.
🎲 Clawdvent — Your AI Dungeon Master in Discord
"The air smells of oak smoke and old rain. A hooded figure slides a parchment across your table. It's marked with a single word: 'Help'. What do you do?"
An open-source AI Dungeon Master powered by OpenClaw. Runs your D&D-style adventures inside Discord with live AI-generated scene art, dice rolls, character tracking, and cinematic narration.
Built for the DEV OpenClaw Challenge 2026.
✨ Features
- 🎭 Cinematic DM persona — dramatic, theatrical, opinionated
- 🖼️ Live AI scene art — DALL-E 3 illustrates each scene + battle map
- 🎲 Dice & combat — proper d20 checks with DC thresholds
- 👥 Multi-player — works across any Discord server
- 💾 Persistent campaign — characters, inventory, and story survive across sessions
- 🧠 Zero training needed — just one markdown file (SOUL.md) shapes the entire DM personality
🚀 Quick Start
Prerequisites
- Docker (or OrbStack on…
Clone it, add your Discord bot + OpenAI token, and you have a working AI DM in 10 minutes.
How I Built It
Total build time: ~3 hours.
The Stack
- OpenClaw (MIT-licensed AI agent runtime, running in Docker via OrbStack)
- Discord Bot API (OpenClaw's built-in Discord channel plugin)
- OpenAI gpt-4o-mini (for narration — fast, cheap, dramatic)
-
DALL-E 3 (scene + battle map illustrations via OpenClaw's
image_generatetool) -
~50 lines of markdown (the entire DM personality in
SOUL.md)
The Architecture
Discord players → OpenClaw Gateway → "clawdvent" agent (routed from Discord channel)
↓
SOUL.md (DM personality)
↓
OpenAI gpt-4o-mini (narration)
↓
DALL-E 3 (scene illustrations)
↓
Image auto-attaches to Discord reply
The Magic — SOUL.md
The DM's entire identity lives in a single file. Here's a snippet:
# SOUL.md — Clawdvent DM
You are Clawdvent, an AI Dungeon Master running a tabletop adventure inside Discord.
## Combat Rules
- Roll d20 + skill to beat a Difficulty Class (DC)
- On natural 20, describe a cinematic success
- Format rolls like: [🎲 d20: 14 + strength(3) = 17 vs DC 15 — HIT]
## Narrative Style
- Hook them in 2 sentences
- Use ellipses for suspense
- Personify the environment
- Never railroad
## Session Zero Opening
> *The air smells of oak smoke and old rain...*
That's it. That's the entire "engine." OpenClaw injects this at the system-prompt level on every message, so the DM never breaks character.
Multi-Agent Routing
One thing that impressed me: OpenClaw supports multiple agents bound to different channels. I have:
-
mainagent → Telegram (a different product I built: MealAI Concierge) -
clawdventagent → Discord (this project)
Same gateway. Separate personalities. Zero code changes. Just a routing config.
Image Generation Flow
When Clawdvent decides a scene needs visuals, it calls the image_generate tool with a detailed fantasy prompt. OpenClaw handles the DALL-E call, downloads the image, and auto-attaches it to the Discord reply. No file hosting needed.
What I Learned
The prompt IS the product. Clawdvent isn't ChatGPT with a Discord token. It's a purpose-built DM, because SOUL.md defines every micro-behavior: how to roll dice, when to generate images, when to describe smells, what NPC voices sound like.
OpenClaw's Canvas and multi-channel features are wildly underused. Most OpenClaw projects I've seen are text-in, text-out. But the moment you add live visuals + multi-channel routing, the whole experience feels like a real product instead of a demo.
Docker is enough. No Kubernetes. No serverless.
docker compose upand you're running a multi-agent system with Discord + Telegram + image generation.Open-source AI runtimes are quietly ready for production. Auth, rate limiting, logging, multi-channel, tool calling — all out of the box.
What's Next
- Add voice narration via OpenClaw's Talk Mode (DMs read scenes aloud)
- Add cron heartbeats so NPCs send DMs to absent players ("Your old ally Thorn has been captured. Will you come?")
- Port to Matrix + Slack channels (same code, new config)
- Open the server to public players during the DEV challenge
Prize Category
OpenClaw in Action — This submission showcases:
- Creativity: Entertainment angle rarely seen in OpenClaw submissions, targeting the tabletop RPG community
- Technical Execution: Multi-agent routing, tool calls, image generation, persistent state in Markdown
- Writing Quality: Clear reproducible setup, zero-secrets public repo, narrative prose that shows the product in action
Try It
Clone the repo, fire up Docker, invite the bot to your Discord server, and play:
shopsmartai
/
clawdvent
AI Dungeon Master in Discord with live scene art. Built on OpenClaw for DEV OpenClaw Challenge 2026.
🎲 Clawdvent — Your AI Dungeon Master in Discord
"The air smells of oak smoke and old rain. A hooded figure slides a parchment across your table. It's marked with a single word: 'Help'. What do you do?"
An open-source AI Dungeon Master powered by OpenClaw. Runs your D&D-style adventures inside Discord with live AI-generated scene art, dice rolls, character tracking, and cinematic narration.
Built for the DEV OpenClaw Challenge 2026.
✨ Features
- 🎭 Cinematic DM persona — dramatic, theatrical, opinionated
- 🖼️ Live AI scene art — DALL-E 3 illustrates each scene + battle map
- 🎲 Dice & combat — proper d20 checks with DC thresholds
- 👥 Multi-player — works across any Discord server
- 💾 Persistent campaign — characters, inventory, and story survive across sessions
- 🧠 Zero training needed — just one markdown file (SOUL.md) shapes the entire DM personality
🚀 Quick Start
Prerequisites
- Docker (or OrbStack on…
May your dice roll high and your rogues stay sneaky. 🎲
Top comments (0)