Open Dungeon 0.3.1: connect a server + custom openings

Connect a server, and write your own openings.

• One provider for any backend. Text Model now has just two choices: Local (Ollama) and Connect a server. The latter talks to any OpenAI-compatible server, with the URL, model, and an optional key entered right in the app, plus one-tap presets for OpenRouter, LM Studio, llama.cpp, and Ollama. OpenRouter is now reached this way (the preset fills the URL); its attribution headers and OPENROUTER_* env fallbacks still apply.
• Write the story's opening yourself. The New Story dialog lets you either let the narrator open the scene (with an optional hint) or type the first passage yourself, used verbatim with no generation.
• Fix: custom-backend chats could not be saved in 0.3.0 (the chat API rejected the new provider). They persist correctly now. If you tried a custom server on 0.3.0, this release fixes it.

Install: open the DMG, drag Open Dungeon to Applications, and open it (right-click then Open the first time, since it is unsigned). Apple Silicon. The local default needs Ollama.

Full changelog: v0.3.0...v0.3.1

Open Dungeon 0.3.0: bring your own backend

Bring your own backend. You no longer have to use Ollama. Switch a chat's provider to Custom and point Open Dungeon at any OpenAI-compatible server you already run.

Set it up right in the app, no config files:

• Backend URL (a bare host, a /v1 base, or a full /chat/completions endpoint all work)
• Model name
• API key (optional, most local servers need none)

Works with llama.cpp, LM Studio, vLLM, TabbyAPI, KoboldCpp, or a remote Ollama. Everything stays on your machine. If a server lacks function-tool support, turns retry without the image tool so the story keeps flowing.

The built-in local path (Ollama + Gemma 4) and the optional OpenRouter provider are unchanged.

Install: open the DMG, drag Open Dungeon to Applications, and open it (right-click then Open the first time, since it is unsigned). Apple Silicon. For the default local model you will need Ollama installed.

Full changelog: v0.2.0...v0.3.0

Open Dungeon 0.2.0 — play controls

Adds the essential play controls and a native Apple Silicon launcher.

New: full play controls

• Do / Say / Story input modes — actions, dialogue, or write narration directly
• Continue — let the narrator advance the scene with no action from you
• Retry — discard the latest passage and regenerate it
• Erase — remove the most recent exchange
• Edit — hover any passage and rewrite it in place (saved locally)

Also

• The app is now a native arm64 (Apple Silicon) bundle — no more "Intel app needs Rosetta" notice.

Install (Mac, Apple Silicon)

1. Install Ollama and open it once.
2. Download Open-Dungeon-0.2.0.dmg, drag Open Dungeon to Applications.
3. Right-click → Open the first time (the app is unsigned).

First launch unpacks a bundled Node.js runtime, downloads the default narrator model (Gemma 4 12B QAT, 7.2 GB, one time), builds, and opens the app. Verified end-to-end from a clean machine state.

SHA-256: `375c21bbf3c3384fe7fbd9a50eba51c9cc8f14d4ba93525d1de865d3e30204ee`

Open Dungeon 0.1.0

First release of Open Dungeon — the easy-to-use, fully local AI roleplay app. The story and the inline scene images are generated entirely on your machine.

Install (Mac, Apple Silicon)

1. Install Ollama and open it once.
2. Download Open-Dungeon-0.1.0.dmg below, drag Open Dungeon to Applications.
3. Right-click → Open the first time (the app is unsigned).

First launch is self-guided: it unpacks a bundled Node.js runtime (no dev tools needed), downloads the default narrator model (Gemma 4 12B QAT, 7.2 GB, one time), builds, and opens the app in your browser. Later launches go straight to your stories.

Running from a clone instead? Double-click Launch.command.

What's inside

• Narrator: Gemma 4 QAT (Q4_0) models via Ollama, selectable per chat with measured RAM shown for each
• Full native context windows (128K–256K) with cache-friendly history packing and automatic "story so far" compaction — the narrator doesn't forget chapter one
• Quick starts: pick a setting, say who you are, get a custom opening scene
• Optional fully local inline scene images (FLUX.2-klein, Apple Silicon)
• Everything stored in a local SQLite file; nothing leaves your machine

SHA-256: f927ec399db74c58ead8c3fded2c2543ed22fc1b4533b77b0087318355721bf8