Profile and response style

Save lasting collaboration preferences, set the AI's communication tendencies, and choose a style with the right scope. These controls manage background, expression, and adaptation separately; the current task can still give specific instructions.

Choose the right place

Open Settings > Personalization to view Collaboration profile, Style presets, and Adaptive response.

For example, “Give the conclusion first, then explain the evidence” suits a lasting reply preference. “Use a formal tone and four sections in this report” can go directly in the current task.

Maintain the collaboration profile

View recorded names, background, and reply preferences in Collaboration profile. To edit text, follow its entry to the conversation-profile area on the Memory page, edit and save, then return to check the result.

Start with a preference you can verify:

Begin with a short conclusion, then explain the evidence. State uncertainty when evidence is missing. Use tables when comparing options.

Put project-specific terminology, checks, and deliverable requirements in Project working rules. Manage past decisions and experiences as memory entries, recalled according to relevance.

How language preferences apply

The profile language is normally a fallback when input language cannot be determined. To use a chosen language regardless of your input language, enable “Always reply in this language.”

Change the interface language in Desktop settings. A menu-language change alone does not establish that profile language preferences failed.

Adjust the AI's personality axes

Personality axes set the AI's lasting communication tendencies. They are not a personality test for you or a capability score for the model.

There are five axes: introversion–extraversion, sensing–intuition, thinking–feeling, judging–perceiving, and assertive–turbulent. A value of 50 is neutral. Saving recalculates the MBTI letters from these values.

Keep the midpoint if you are unsure. To adjust:

1. Change one axis according to the communication tendency you want.
2. Save and confirm the saved acknowledgement. “Not saved yet” after moving a slider is not a completed change.
3. Compare replies to the same task.
4. Adjust and save again if needed. Use the revert action to discard unsaved changes.

These tendencies do not add model capabilities or tool permissions, and cannot guarantee factual accuracy or a fixed form for every response. State exact formatting requirements in the task.

Create and activate a style preset

Under Style presets, choose an existing preset or create one. A new preset has separate fields:

A name is required, along with at least one of naming style or reply tone. The name allows up to 30 characters; each instruction field allows up to 800.

After saving, activate the preset and choose its scopes. Saving a preset, activating it, and applying it to the running service are separate states to confirm. If the panel says it was saved but not applied, use Retry or View service status. Appearing in the preset list alone does not prove it is active.

The AI-generation entry can help draft a style. It takes you to a conversation for authoring. Review the result, remove unsuitable lasting instructions, and confirm saving and activation.

Choose application scopes

To make a report's body more formal, state that requirement in the task. Enabling Cowork deliverable copy alone will not do it. Enable Reply tone as well if you also want formal conversation replies.

Style can reach beyond the conversation you are viewing on desktop. Consider connected messaging channels before enabling Reply tone.

Choose “No preset” when you no longer want one. Some built-in styles require Reply tone and explain their reach before activation; deactivation restores saved scope settings. An existing conversation may still echo its historical tone, so use a new conversation to check deactivation.

Adaptive response and tool-risk modes

Adaptive response lets EvoX adjust replies to match your state. Its enable switch and tool-risk mode are separate controls.

When enabled, each reply consumes a little additional model quota. The extra analysis does not read conversation content or use tools; if it fails, the normal reply is unaffected. Decide based on whether you want this adaptation and accept the additional usage.

Analysis and risk-mode changes apply to new conversations. Use “Start new conversation” after saving rather than judging the saved configuration from an old conversation.

The panel explains when organization policy or environment configuration locks the switch. You cannot override it here. Risk mode does not replace existing Permissions and approvals or authorize an otherwise unauthorized action.

Verify with the same task

Use fixed input to reduce differences caused by task content:

Turn these notes into a project update: retrieval is complete; connection testing is unfinished; the release date is undecided. Lead with the conclusion and do not invent a date.

Check the conclusion and uncertainty, then inspect conversation naming and deliverable titles and summaries separately. Content changes and naming changes need separate checks.

Begin with one preference, one preset, and one short task. Add more requirements once you have verified the scopes.