A local-first chronicle for 5e campaigns. Record sessions, auto-transcribe with AI, generate bardic summaries, track characters, and build your campaign archive — all on your machine, no cloud required.
Built with NativePHP, Laravel 12, React, and Inertia.js.
Tested on Apple Silicon. Runs on any sufficiently powerful machine — Macs tend to excel at local AI workloads.
| Feature | Description | |
|---|---|---|
| 📚 | Campaigns | Manage multiple campaigns with party portraits and full session history |
| 🧙 | Characters | Full 5e character sheets — HP, spell slots, inventory, conditions, short/long rests |
| 📥 | Beyond Import | Import characters directly from D&D Beyond |
| 🎙️ | Session Recording | Record audio at the table and transcribe locally with WhisperX |
| 📖 | Bardic Summaries | AI-generated epic prose recaps of your sessions |
| ⚔️ | Encounter Tracker | Auto-detect encounters from transcripts with round-by-round breakdowns |
| 🧝 | NPC Roster | Track campaign NPCs with attitudes, locations, and AI-generated portraits |
| 🔮 | Oracle | AI chat assistant with full access to your campaign and character data |
| 🎨 | Art Generation | Generate character portraits and scene art via ComfyUI, z.ai, or DALL-E |
| 📄 | PDF Export | Export campaign or session data to PDF |
| 🔍 | Detail Extraction | LLM-powered extraction of stat changes and NPC appearances from transcripts |
- Download the latest
.dmgfrom the Releases page - Open the DMG and drag Lorefire.app to
/Applications - Because the app is not notarized, macOS Gatekeeper will block it on first launch. Remove the quarantine flag:
xattr -rd com.apple.quarantine /Applications/Lorefire.app- Open the app normally — the onboarding wizard will guide you through the rest.
# Clone the repo
git clone https://github.com/justinwoodring/lorefire.git
cd lorefire/lorefire-desktop
# Install PHP dependencies
composer install
# Install Node dependencies
npm install
# Copy and configure environment
cp .env.example .env
php artisan key:generate
# Run database migrations
DB_DATABASE=$(pwd)/database/nativephp.sqlite php artisan migrate
# Build frontend assets
npm run build
# Start the app
php artisan native:serveNote: Make sure no other Vite process is already running on port 5173 before starting
native:serve. You can check withlsof -i :5173.
All settings are configured inside the app via Settings (after onboarding). Nothing requires editing .env manually.
Lorefire bundles WhisperX in a local Python virtual environment. Audio is never sent off your device.
On first launch, the app automatically installs the Python venv in the background. You can monitor progress in the onboarding wizard or in Settings.
| Setting | Description |
|---|---|
| Model size | tiny → large-v3. base is recommended for most Macs. Larger = more accurate but slower. |
| Language | Default English. Auto-detect and 9 other languages supported. |
| HuggingFace Token | Required for speaker diarization (identifying who said what). Free at huggingface.co/settings/tokens. You must also accept both pyannote model licenses while logged into the same account: pyannote/speaker-diarization-3.1 and pyannote/segmentation-3.0. |
Large-v3 note: Needs 8 GB+ RAM and will be noticeably slow on CPU. On Apple Silicon with enough unified memory it works well.
Pick one provider in Settings. API keys are stored locally in SQLite and never leave your machine except to call the chosen provider.
| Provider | Notes |
|---|---|
| z.ai (recommended) | GLM models via the z.ai Coding Plan. Supports glm-4.6, glm-4.7, glm-4-flash, and others. Get a key at z.ai. |
| OpenAI | Uses gpt-4o-mini by default. Requires an OpenAI API key. |
| Anthropic | Uses claude-3-haiku. Requires an Anthropic API key. |
| Ollama | Any locally running model (e.g. llama3, mistral). Requires Ollama running on localhost:11434. No API key needed. |
| None | Falls back to template-based summaries. All other features still work. |
Configure in Settings → Image Generation.
| Provider | Notes |
|---|---|
| ComfyUI (local) | Connects to your local ComfyUI instance. Uses whatever checkpoint is currently loaded. Default URL: http://localhost:8188. See ComfyUI setup below. |
| z.ai | Uses your z.ai API key (configured above). Recommended model: cogview-4-flash. |
| OpenAI DALL-E | Uses your OpenAI API key. Recommended model: dall-e-3. |
| None | Disables image generation entirely. |
Two art styles are available: Lifelike (realistic fantasy painting) and Comic (graphic novel illustration). The default can be set globally in Settings and overridden per campaign.
ComfyUI is optional — it lets you generate portraits and scene art locally using Stable Diffusion / FLUX models.
Follow the official install guide: https://github.com/comfyanonymous/ComfyUI
git clone https://github.com/comfyanonymous/ComfyUI.git
cd ComfyUI
pip install -r requirements.txtLorefire's workflows use custom nodes. ComfyUI-Manager lets you install them in one click.
cd ComfyUI/custom_nodes
git clone https://github.com/ltdrdata/ComfyUI-Manager.gitRestart ComfyUI, then open the Manager (button in the top-right of the UI) and click Install Missing Custom Nodes.
Install these via ComfyUI-Manager → Install Custom Nodes:
| Node | Purpose |
|---|---|
| ComfyUI-Manager | Node manager (already installed above) |
| rgthree-comfy | Power Lora Loader used in FLUX workflows |
| ComfyUI_IPAdapter_plus | IP-Adapter for portrait consistency (optional but recommended) |
After installing, restart ComfyUI and use Manager → Install Missing Custom Nodes to resolve any remaining dependencies.
Place checkpoint files in ComfyUI/models/checkpoints/. Recommended models:
| Model | Notes |
|---|---|
| FLUX.1-dev | Best quality; requires HuggingFace login. Download flux1-dev.safetensors. |
| FLUX.1-schnell | Faster, fewer steps needed. |
| Any SDXL checkpoint | e.g. RealVisXL. Place in checkpoints/. |
For FLUX models you also need the text encoders and VAE — place them in:
ComfyUI/models/clip/—clip_l.safetensors,t5xxl_fp16.safetensorsComfyUI/models/vae/—ae.safetensors
These can all be downloaded from the FLUX.1-dev HuggingFace repo.
cd ComfyUI
python main.py
# Runs on http://localhost:8188 by default
# For GPU-poor machines: python main.py --lowvramIn Settings → Image Generation, set the provider to ComfyUI and confirm the base URL (http://localhost:8188).
Lorefire submits generation jobs via the ComfyUI API. The active checkpoint in ComfyUI is used automatically — swap the model in ComfyUI's checkpoint loader to change style.
| Layer | Technology |
|---|---|
| Framework | Laravel 12 |
| Desktop runtime | NativePHP / Electron 1.3 |
| Frontend | React 19 + TypeScript |
| Routing / SSR bridge | Inertia.js v2 |
| Styling | Tailwind CSS v4 |
| Build tool | Vite 7 |
| Database | SQLite (local, via NativePHP) |
| Transcription | WhisperX (local Python venv) |
| Queue | Laravel database queue |
- All data is local. The SQLite database lives in
~/Library/Application Support/lorefire/(production) or~/Library/Application Support/lorefire-dev/(dev). - No account required. Nothing is synced to a server. Your campaigns stay on your machine.
- Other hardware: Tested on Apple Silicon. It may work on Intel Macs, Windows, or Linux, but this is untested. WhisperX in particular will be significantly slower without a dedicated neural engine or GPU.
- Transcription is CPU/ANE-bound. On Apple Silicon, WhisperX uses the
mpsbackend. Large audio files may take a few minutes even onbasemodel.
MIT — see LICENSE
© 2026 Justin Woodring