Skip to content

Voice & Style

Settings → Voice & Style controls defaults for Caret Assist and Page Draft. On desktop web and macOS, it also provides project-specific Voice conditioning. The native iOS app includes the workflow defaults; use desktop to manage a voice-conditioning sample.

Available models are loaded from Feldar for your account. The model list may therefore differ by plan or change over time.

Both workflows provide these base styles:

  • Default — balanced and adaptive
  • Literary — layered and evocative
  • Commercial — fast and punchy
  • Lyrical — poetic and sensory
  • Minimalist — spare and precise

The base style is a default direction, not a guarantee that every result will use identical phrasing.

Caret Assist provides:

  • A model selector
  • A base style
  • Short, Default, or Long insertion length
  • Custom instructions for behavior, style, and tone

Length is a preference for intents that support variable expansion. The selected action and surrounding draft still affect the result.

Page Draft provides:

  • A model selector
  • A base style
  • Five characteristics: Show vs. Tell, Pacing, Dialogue, Interiority, and Description
  • More, Default, or Less for each characteristic
  • Custom instructions for voice, world rules, point of view, tense, and tone

Use a sample of your own writing to help Feldar follow a project’s voice:

  1. In Voice conditioning, choose a Project.
  2. Paste 600–2,000 words into Writing sample, preserving paragraphs and punctuation. Around 1,000 words gives a stronger sample than the minimum.
  3. Select Save sample and wait for confirmation.

Feldar uses the sample until the manuscript grows longer, then follows the manuscript’s voice. A saved sample can take a few minutes to prepare. If an update is delayed, the status message explains that Feldar will retry.

Choose Remove sample to stop using the saved sample. Save or discard edits before choosing another project. If the sample changed elsewhere, copy any draft text you want to keep, then choose Discard draft and reload before editing the saved version again.

Voice conditioning provides context for generated prose; it does not change existing writing or guarantee an identical style in every result.

Model and style selectors save when changed. Custom instructions save when you leave the text field. The voice-conditioning sample has its own Save sample button. Wait for any in-progress update before closing Settings or changing devices.