Skip to content

Settings

The Artificer's Workshop

The Settings page lets you configure TaleKeeper's behavior, AI providers, and integrations.

Settings page showing Transcription section with Whisper model selector and batch size, HuggingFace token field, LLM Provider section, Image Generation section, Email SMTP configuration, Data directory selector, Reset to Defaults, and Save Settings buttons

Transcription

Setting Description Default
Whisper Model Speech recognition model distil-large-v3
Batch Size Parallel processing chunks (empty = auto-detected) Auto

Model guide:

Model Speed Accuracy Use Case
tiny ~30 sec / 10 min audio Lower Quick tests
base ~1 min / 10 min audio Fair Short sessions
small ~2 min / 10 min audio Good Casual use
medium ~3 min / 10 min audio Very Good Balanced option
distil-large-v3 ~2 min / 10 min audio Excellent Recommended
large-v3 ~5 min / 10 min audio Best Critical recordings, accented speech

Optimized for Apple Silicon

TaleKeeper's transcription is built specifically for Apple Silicon Macs, giving you fast, accurate results. It automatically filters out silences and background noise before processing your recording, so you get cleaner transcripts without any extra steps.

Batch Size

This controls how much of your recording TaleKeeper processes at once. Leave it empty and TaleKeeper will choose the best value for your Mac automatically. Only change this if you notice slowdowns or unresponsiveness during transcription.

Providers

HuggingFace

Field Description
Token HuggingFace access token for speaker diarization

Automatic speaker identification (figuring out who said what) requires a HuggingFace token. To obtain one:

  1. Create a free account at huggingface.co
  2. Accept the license at pyannote/speaker-diarization-3.1
  3. Generate an access token in your HuggingFace settings
  4. Paste it into the HuggingFace Token field

Required for Speaker Detection

Without a HuggingFace token, transcription still works but all speech will appear under a single speaker — TaleKeeper won't be able to tell voices apart.

LLM Provider

Configure the AI that powers summaries, session naming, and scene descriptions.

Field Description Default
Base URL Your AI service's address http://localhost:11434/v1
API Key Access key (not needed for Ollama) —
Model Which model to use llama3.1:8b

Click Test Connection to verify. A green checkmark means you're connected.

Provider Compatibility

TaleKeeper works with Ollama, OpenAI, LM Studio, and most other AI services.

Image Generation

TaleKeeper generates scene illustrations directly on your Mac — no internet connection or external service required.

Field Description Default
Model Which image model to use FLUX.2-Klein-4B-Distilled
Steps How many passes the model makes (more = more detail, slower) 4
Guidance Scale How closely the image follows the description (higher = stricter) 0

Click Check Availability to verify that mflux is installed.

Apple Silicon Required

Image generation runs natively on Apple Silicon (M1 and newer). No internet connection, external service, or special hardware is needed — everything happens directly on your Mac.

Tuning Image Quality

The defaults (4 steps, 0 guidance) are optimized for the Klein distilled model. Increasing steps produces more detailed images but takes longer. Increasing guidance scale makes the image follow the prompt more closely.

Email (SMTP)

Configure email for sharing summaries.

Field Description Example
SMTP Host Mail server smtp.gmail.com
SMTP Port Server port 587
Username Login you@gmail.com
Password App password ••••••••
Sender Address From address you@gmail.com

Password Security

Passwords are encrypted at rest in TaleKeeper's database. They're never stored in plaintext.

Data Directory

Where TaleKeeper stores all recordings, transcripts, images, and the database.

  • Click Browse to open your system's native folder picker
  • Or type a path directly
  • Default: data/ in the TaleKeeper directory

Backup This Folder

Your entire TaleKeeper archive lives in this directory. Back it up to preserve your campaign history.

Theme

TaleKeeper supports dark and light themes. Toggle between them using the theme button in the sidebar footer (☀️ / 🌙).

  • Dark mode is the default
  • On first launch, TaleKeeper matches your Mac's current dark or light mode setting
  • Your preference is remembered across sessions

Re-run Setup Wizard

Click Run Setup Wizard at the bottom to walk through initial configuration again.

Reset to Defaults

Click Reset to Defaults to restore all settings to their original values. Your API keys and tokens are preserved — only model, URL, and tuning preferences are reset.

Advanced: Environment Variables

If you're running TaleKeeper in a shared or server environment, these settings can be configured before launch. Most users won't need these — everything is configurable through the Settings page.

Variable Description Default
TALEKEEPER_CORS_ORIGINS Allowed CORS origins (comma-separated) http://localhost:5173
TALEKEEPER_SECRET Encryption key for stored passwords Built-in default
LLM_BASE_URL LLM provider URL http://localhost:11434/v1
LLM_API_KEY LLM API key —
LLM_MODEL LLM model name llama3.1:8b
HF_TOKEN HuggingFace token for diarization —
IMAGE_MODEL Image model name FLUX.2-Klein-4B-Distilled
IMAGE_STEPS Image inference steps 4
IMAGE_GUIDANCE_SCALE Image guidance scale 0

Priority

Settings saved in the UI take precedence over environment variables.

Next: Tips & Tricks →