> ## Documentation Index
> Fetch the complete documentation index at: https://docs.trydear.app/llms.txt
> Use this file to discover all available pages before exploring further.

# Inline completions: AI suggestions as you type

> Press Tab to accept AI-generated continuations in Write mode. Control suggestion length, response delay, and how closely completions match your voice.

Inline completions appear as ghost text at your cursor as you type in Write mode. When a suggestion looks right, press **Tab** to accept it. The suggestion is added to your document and attributed to the AI — it does not count toward your human-written word total.

## Enabling and disabling completions

Inline completions are controlled separately from other AI settings. You can toggle them from **AI preferences** in the user menu, or from the **Inline suggestions** section of the AI settings modal (open via your user menu → **AI preferences**, or use the keyboard shortcut).

When completions are off, the rest of the AI settings — voice learning, guardrails, voice blending — continue to work for other AI features in the app.

## Accepting and refreshing suggestions

| Key                         | Action                        |
| --------------------------- | ----------------------------- |
| **Tab**                     | Accept the current suggestion |
| **Shift + Tab**             | Request a different variation |
| **Escape** (or keep typing) | Dismiss the suggestion        |

<Tip>
  If the ghost text isn't quite right, pressing **Shift + Tab** asks the model for an alternative suggestion rather than accepting or discarding the one on screen.
</Tip>

## Suggestion length

Three modes control how much text Dear drafts ahead of your cursor.

<AccordionGroup>
  <Accordion title="Word — up to 2 words">
    The shortest option. Useful when you want a nudge on the current word or the next, without the model jumping ahead into full sentences.
  </Accordion>

  <Accordion title="Phrase — 1 to 20 words (default: 5)">
    The default mode. Dear suggests a short continuation — by default **5 words** — and you can tune the exact count in AI preferences using the phrase length slider or the quick-select badges (2, 3–4, 5, 8, or 12 words). Range: 1–20 words.
  </Accordion>

  <Accordion title="Sentence — up to several sentences">
    Dear drafts as far as the **max lookahead** setting allows (default: **2 sentences**, range: 1–5). Use this when you want full thought completions rather than fragment-level nudges.
  </Accordion>
</AccordionGroup>

You can switch between modes at any time from the **Inline suggestions** section of AI preferences.

### Phrase length quick-select

When **Phrase** mode is active, five preset badges appear in AI preferences:

| Badge                 | Word count |
| --------------------- | ---------- |
| 2 words               | 2          |
| 3–4 words             | 3–4        |
| **5 words** (default) | 5          |
| 8 words               | 8          |
| 12 words              | 12         |

You can also drag the slider or use the +/− buttons to set any value from 1 to 20.

## Suggestion delay

The delay controls how long Dear waits after you pause typing before a suggestion appears. A shorter delay keeps you in flow; a longer delay avoids distracting suggestions when you're mid-thought.

| Delay                | Index |
| -------------------- | ----- |
| 0 ms (instant)       | 0     |
| 200 ms               | 1     |
| **400 ms** (default) | 2     |
| 800 ms               | 3     |
| 1500 ms              | 4     |

Drag the **Instant → Patient** slider in AI preferences, or click any badge to jump directly to that value.

## Max lookahead (sentences mode)

When **Sentences** mode is active, **max lookahead** sets a hard cap on how far a single suggestion can draft. Default is **2 sentences**; the range is 1–5. This setting appears in the inline suggestions panel and applies only when suggestion length is set to **Sentences**.

## Match my voice

The **Match my voice** toggle (on by default) tells Dear to bias completions toward your voice profile — the same persona and vocabulary as the text you've already written in the current document. Turning it off shifts completions closer to generic model output.

This setting works alongside voice learning: the more human-written words Dear has observed, the more precisely it can mirror your style.

<Note>
  **Match my voice** is separate from voice blending. Voice blending sets a mix of named style references (My voice, Essay — Didion, etc.). Match my voice is a document-level instruction to stay consistent with the draft in front of you.
</Note>

## How completions interact with your voice settings

When Dear generates a suggestion, it assembles a hint from your active AI settings:

* **Voice learning on** — Dear notes your voice maturity percentage and asks the model to mirror your rhythm and diction.
* **Match my voice on** — Dear instructs the model to stay in the same persona and vocabulary as the current draft.
* **Language bank on** — Dear asks the model to prefer vocabulary and connective phrasing from your saved language bank.
* **Custom voices on** — Dear passes your current voice blend (e.g. "user baseline 40%, essay (Didion-like) 39%") as a style instruction.

All of these settings are combined into a single hint sent with each completion request. You can turn any of them off independently without affecting the others.

## Related

* [Voice learning](/ai/voice-learning) — how Dear measures and learns your writing style
* [Voice blending](/ai/voice-blending) — mix named style references in AI suggestions
* [AI guardrails](/ai/guardrails) — block unwanted syntax and framing patterns