Admin

Ask NephAI

Runs when you ask

Answers your questions, gathers context, and prepares guarded actions.

You see the result of this in Ask NephAI

Which model does this

Ask NephAI

Answers interactive questions across your mail, calendar, Tasks, prepared coordination, people, and knowledge. Optimized for fast, grounded conversation.

Customized

Which model does this

Every model that can actually run this is listed. The one at the top of each group is usually the safe choice.

How hard it should think

Available choices work with both the primary and fallback models.

Thinking harder gives better answers but takes longer and costs more.

How long an answer to allow

16,000 tokens · about 12,000 words of room

Longer answers cost more. “Normal” suits almost everything; only raise it if answers are being cut short.

Ordered fallbacks

These choices apply to the next request handled by this service.

Ask NephAI

Primary control

Ask NephAI enabledControls the assistant surface without changing other AI features.

Conversational voice

New sessions

Voice conversationsAllows live voice when its secure connection is ready.

Provider

Save your choice, then start a new voice conversation to hear it. Each provider keeps its own voice preference.

Uses the reviewed OpenAI voice connection.

Voice model

The voice model that handles live speech in and out.

OpenAI voice

Marin and Cedar are OpenAI's recommended choices for quality.

When to treat you as finished speaking

Semantic waits for a complete thought; Silence-based reacts to any pause.

Natural interruptionStop spoken playback immediately when you begin talking.Speak responses in voice modeWhen off, voice input remains available but answers stay on screen.

Dictation

OpenAI Realtime

Speaking an instruction while writing an email continues to use OpenAI Realtime in this release. Changing conversational voice does not change dictation.

Reading and memory budgets

1,000 tokens ≈ 750 words

Tokens are the word-pieces AI models read and write. These budgets decide how much NephAI reads, remembers, and gathers each time you ask something. They must all fit inside the model memory planned for below — saving checks the math for you.

Model memory to plan for

The total working memory NephAI assumes the model has. Every budget below must fit inside it.

Chat history

How much of the current conversation NephAI re-reads on every turn.

Evidence per question

The total email, calendar, task, and Knowledge text NephAI may gather to answer one question.

Room for the answer

Memory held back so a long answer always has space to be written.

One lookup's reading limit

The most text a single lookup — one email, one document, one search — may bring back at once.

Lookups at once

How many lookups NephAI runs at the same time.

Lookups per question

The most lookups one question may trigger before NephAI must answer.

Hand off to background after

Seconds before a long-running question keeps working in the background.

Spending cap per question

A hard stop on model cost, in US dollars, for a single question.

Search prior conversationsOlder conversations may be retrieved when they are relevant to the current question.Evidence inspectorKeep source references available without printing a noisy source list below every answer.Productivity follow-upsEnd answers with a specific next question and clickable options.

Conversations, files, and memory

Keep quiet conversations for

Days an untouched conversation stays before cleanup.

Trash recovery window

Days a deleted conversation can still be brought back.

Largest document (MB)

The biggest file a conversation will accept.

Largest audio file (MB)

The biggest recording a conversation will accept.

Explicit memoryAllow direct requests such as “remember this.”Conversation attachmentsAllow documents, images, spreadsheets, text, and audio to become conversation context.Add to KnowledgeFiles remain conversation-only until you explicitly promote them.

Actions

Outward actions require a click

Email actionsPrepare inline email send cards. Ask NephAI confirmation remains mandatory.Task actionsCreate one canonical Task for the outcome and prevent duplicates.DelegationPrepare a delegation email and your follow-up task as one confirmed action.Calendar actionsPrepare meetings and invitations for explicit review.RSVP actionsPrepare RSVP changes for explicit review.

Approval before outward actionsActions started in Ask NephAI require a visible confirmation click.Locked on

What NephAI remembers about you

Reviewing what NephAI remembers about you is a personal matter rather than an administrative one, so it lives with your own settings.

Review 0 memories Connection used for searching

How you have told it to approach this

Using NephAI's own wording

Coordination & assistant

Assistant chatused by Ask NephAI

How the assistant behaves in conversation. Your identity, the rules block, and today's date are always added around it.

Default

Voice sessionused by Ask NephAI

The spoken persona in Ask NephAI voice: role, tone, when it announces a lookup, how it handles unclear audio. Your identity and the rules block are always added above it; every real question still delegates to the assistant brain.

Default

Voice dictationused by Ask NephAI

The composer's dictation channel: transcribe your spoken drafting instructions and nothing else. It never speaks or authors content.

Default

How much this has been used

Last 28 days

Times it ran

Text handled

51,444

What it cost

$0.1342

See this alongside everything else

Other settings that affect this

Assistance Rules Automations