Personal AI settings

Personal AI settings change Centrus for your account only. They do not change what your colleagues see, and they do not change the files Centrus can search. Those stay with your tags. An admin sets the starting point for everyone on AI Model. Anything you set here is used instead, for you.

Six choices sit together under Personal AI Settings on the Account tab. Chat history search sits just above that block. It is separate: it decides whether Centrus may look through your other chats, and it does not change the model or the voice.

• Personal AI Model is the model that answers in every chat you own, including WhatsApp and Microsoft Teams.
• Personal Temperature is how much the wording varies, and only some models accept it.
• Personal Live voice is the voice Centrus uses in Voice mode.
• Personal image model and Personal image quality control pictures you ask for in web chat.
• Personal AI Context is background Centrus reads on your answers, added to your organisation's context.
• Chat history search lets Centrus look up your own earlier chats when you ask it to.

Comparing models, credits and reasoning is covered in Choose the AI model. Voices are covered in Talk to Centrus. Picture models, quality and credits are covered in Web search and images. This guide summarises those three and then stays with Personal AI Context and Chat history search, which are not written up anywhere else.

Everyone can change their own settings, whatever their role. Admins see a heading Account at the top of the tab. Anyone else sees the same form without that heading. The rest of the tab, your name, phone, theme, password and two-factor authentication, is in Profile and sign-in.

1. 1Click Settings in the menu on the left. The Account tab opens. The page heading is Settings, and the subtitle is Update your account details.
2. 2Scroll past Theme. Chat history search is next, then the floating help button, then Personal AI Settings.
3. 3Change what you need. Chat history search saves as soon as you click On or Off. The fields under Personal AI Settings wait until you click Save changes at the foot of the form.
4. 4Click Save changes. A notification reading Updated confirms the form. If Personal AI Context changed, a second message reads Your AI context has been updated successfully.

Discard changes puts the form back to the last saved names, model, voice, images, temperature and context. It does not undo Chat history search, Theme or the help button, because those three save on their own the moment you click them. Your next message uses the saved model and context. Settings does not reload the page.

Admins see the same personal choices again on Settings, then AI Model, under Show advanced. The context box there is labelled Personal Context, not Personal AI Context, but it is the same text. Changing it in either place changes it for your account. The organisation's own context, model, voice and images are on that tab too. See AI Model.

You can also change Personal AI Model from a chat you own: open the gear menu and click AI Model. That window says the model is for this chat, but it saves to your account and reloads the page. It is the same setting as the list on this tab. See Change your model from a chat.

Personal AI Model is the model that writes your answers. Leave it on Use organisation AI model to follow the model an admin has set for everyone. Choose a named model to use that one in every chat you own, from the next message. Earlier answers stay as they were.

The information icon beside the label opens What does Personal AI Model do? It says: if set, this will override any default organisation AI model for your account only, allowing you to customise your own AI experience.

The list has 15 models and continues below the ones in the figure, through to GPT-4.1. Every model in the list can search your files and the web, create images, and read the PDFs and photos you attach. GPT-4.1 is the one with a red cross for reasoning. The information icon on a row opens that model's details, including how many message credits one answer uses: 1 for most models, 2 for GPT-5.6 Terra and GPT-5.2 Pro, 4 for GPT-5.6 Sol, and 10 for GPT-6 Astra. Building a file adds 2 more, and a picture adds its own credits on top.

For most day to day work, GPT-5.6 Terra is the model Centrus recommends. GPT-5.6 Luna costs half as much and suits short questions. GPT-5.6 Sol and GPT-6 Astra suit the occasional heavy job. The full comparison, who a change affects, and what happens to WhatsApp, Microsoft Teams and shared chats, is in Choose the AI model. Credits for the whole organisation are in Credits and browser minutes.

Use organisation AI model does not tell you which model that is. If you need the name, ask an admin, or look at Organisation AI Model on the AI Model tab if you are one. A shared chat keeps its owner's model. People the chat is shared with can read it, and they cannot change the model from that chat. See Share a chat and see team history.

Temperature is how much the wording of answers varies. A low number stays closer to the same phrasing each time. A high number varies more. It does not change which documents Centrus searches, and it does not change the credit cost.

The slider only appears when Personal AI Model is a model that accepts a temperature: GPT-5.2, GPT-5.1 or GPT-4.1. It is labelled Personal Temperature: followed by the number to two decimal places, from 0.00 to 1.00, in steps of 0.01. It starts at 0.50. The information icon opens What does Temperature do? Lower values, such as 0.1 to 0.3, keep answers more focused and predictable. Higher values, such as 0.7 to 1.0, vary more. The window recommends a moderate setting, around 0.5, for most professional work.

For every other model, and whenever the list is still on Use organisation AI model, the slider is hidden and the page reads Temperature is not supported for the selected reasoning model. That line also shows when you have not chosen a personal model at all, so it is what most people see.

GPT-4.1 uses the temperature on every answer. GPT-5.2 and GPT-5.1 only use it when reasoning is set to None. At any other reasoning level, Centrus leaves the temperature out of the answer, even if the slider shows a number. Reasoning is the brain button on the message bar. It is not saved, and it returns to the model's starting level when the page reloads. The levels for each model are in Reasoning effort and temperature.

There is no Use organisation temperature choice. The number on the slider is stored with your account when you click Save changes. Centrus then uses your number, instead of the organisation's, whenever the model that answers you accepts a temperature. That includes answers on the organisation's model, if that model accepts one. If your account has no temperature stored yet, Centrus uses the organisation's number, or 0.50 if the organisation has not set one. WhatsApp and Microsoft Teams have no reasoning button, so they use the model's starting reasoning level, and the temperature only applies when that level allows it.

Personal Live voice is the voice Centrus speaks with in Voice mode. The line under the label reads: Overrides the organisation Live voice for your account only. It does not change dictation, and it does not change the play button on written answers, which always uses the same standard voice.

The closed list shows the voice and its accent, such as Willow · Irish. The open list adds a short style beside the accent. Vesper is marked RECOMMENDED. The full set is Vesper (British), Marin (English), Quartz and Ripple (Australian), Willow and Stone (Irish), Gleam and Meridian (North American), Delta and Cinder (Southern US), Bossa and Tempo (Brazilian Portuguese) and Beacon (Filipino). Centrus speaks British English in all of them. The voice changes how it sounds, not the language.

Click the play button to hear the voice that will be used: your choice, or the organisation's voice when the list says Use organisation Live voice. If the sample cannot play, the message is Could not play this voice preview. You can still save the voice. A new choice applies from the next Voice mode conversation you start, after you click Save changes. How to start Voice mode, the 15 minute limit, and what it costs are in Talk to Centrus.

Admins set the voice everyone else follows on Settings, AI Model, under Organisation Live voice. If neither you nor the organisation has a valid voice stored, Centrus uses Vesper.

Pictures in web chat use two personal choices, both under Personal AI Settings. They apply when the picture button on the message bar is on. They do not apply on WhatsApp or in Microsoft Teams, which do not generate pictures this way.

• Personal image model is Flare or Sunburst. The line under the label describes Flare as fast everyday generation and Sunburst as slower, with the highest quality and the most precise edits. It tells you to leave the list on the organisation default unless you want your own. Flare is marked RECOMMENDED in the list. The closed list names the organisation's model in brackets, for example Use organisation image model (Flare).
• Personal image quality sets how much extra detail a picture is given, and how many extra credits it costs. The closed list names the organisation's quality the same way, for example Use organisation image quality (Auto).

The line under Personal image quality reads: Image generation uses extra credits when an image is produced: 1 (low) to 10 (max). Auto and medium are 2. The chat model's usual credits still apply. Saying high quality or low quality in a prompt still uses that quality for that message. The same happens for extra high, max, medium and draft wording in the message. Picture credits are only added when a picture is actually made. On GPT-5.6 Terra, a message is 2 credits, and the same message with one High quality picture is 6.

If you leave both lists on the organisation choice, you follow the admin's settings, or Flare and Auto when the organisation has not chosen. Turning the picture button on, revising a picture, and the full credit table are in Web search and images.

Personal AI Context is text Centrus reads before it answers you. Use it for things that are true in every chat of yours: your role, the sites or contracts you look after, how you want answers written, and what you do not want it to do. It is not a place to store a fact that should be checked against a document. Centrus still searches your files. The context tells it how to use what it finds.

The information icon opens What does Personal AI Context do? The window says Personal AI Context is combined with any context set by your organisation, but personal context is only applied to your own conversations. Use it to tailor responses to your needs, preferences or role, without affecting other users.

The box has no character counter and the page does not show a limit. It is a fixed height and does not grow: longer text scrolls inside it. Centrus stores whatever you save, and it sends the whole text with your answers, so keep it to a short paragraph. A page of instructions crowds out the organisation context and the documents.

A useful context names you, says what you are responsible for, and gives one or two rules about the shape of the answer. Sarah Price, contracts manager at Meridian FM, uses this:

Write it as instructions to Centrus, in plain sentences. These work well:

• Your role and the sites, contracts or trades you cover, so answers name the right ones.
• How you want the answer shaped: short enough for an email, British English, pounds, or a list rather than a long paragraph.
• A standing limit, such as not proposing a change to a contracted response time, or not drafting a client email unless you ask.

These work badly, and are better put somewhere else:

• A price, a part number or a clause from a contract. That belongs in a file on Train, so Centrus can quote it and you can update the file. See Upload files.
• A correction to one wrong answer, such as the battery size on a panel. That is a lesson, and it can apply to you, to a tag, or to the whole organisation. See Teach Centrus from a chat.
• Anything about a colleague. The context is only read on your chats, so it will not change their answers, and it is the wrong place for a rule the whole team should follow. That belongs in the organisation context, or in a tag's context.
• A request to ignore the organisation's context or your tags. Centrus is told to use all of them together. It is not told that yours replaces theirs.

On each answer, Centrus is given the pieces in this order: the organisation context, organisation lessons, the context and lessons for the tags on that chat, your name and position from the Account tab, then your Personal AI Context, then your personal lessons. Lessons are there to build on what the documents say. Your context does not switch the others off. Tag context is set on the tag, up to 1,000 characters, and is covered in Tag context. The organisation context is covered in AI Model.

The same text is labelled Personal Context on the AI Model tab, with the line: This context applies only to your conversations and is combined with the organisation context. The placeholder on both boxes is Add personal context to customise AI responses for you... Saving from either tab writes the same field. It applies on the web, on WhatsApp and in Microsoft Teams, because those conversations are yours. It does not apply to an agent that has its own instructions in the Agent Builder.

Nothing is saved until you click Save changes, and only if the text is different from what is already stored. You then see Your AI context has been updated successfully. If that save fails, the message is Failed to update your AI context. Please try again. The rest of the form can still show Updated. Empty the box and save to stop sending any personal context. Your colleagues' answers do not change either way.

Chat history search lets Centrus look through your own earlier chats when you ask it to remember one. It is on until you turn it off. The line under the heading reads: Let Centrus look up your other chats when you ask it to. This only affects your account.

Clicking On or Off saves immediately. You do not need Save changes, though that button sends the same choice again if you save the rest of the form. Discard changes will not put the switch back. Turning it off does not delete any chat. It only stops Centrus looking through them. Your chats stay in the list, and you can still open them yourself.

With the switch on, Centrus looks something up when you ask if it remembers another conversation, refer to something you said in a different chat, or name a date you remember. If you refer to something earlier in the current chat that is no longer in view, it searches this chat instead. It is told not to invent a memory. If nothing matches, it should say so. When it uses what it found, it should cite the chat's title and the date.

On the web, the steps appear in the work panel at the top of the answer: Searching your other chats, Searching this chat, and Opening a past chat. Afterwards the panel folds to Worked, with the number of steps. On WhatsApp there is no work panel, so Centrus is told to summarise what it found in the reply itself, and not to tell you to look at a panel. The same search covers chats you had in Microsoft Teams. Looking through chats does not add message credits of its own. The answer still costs whatever your model costs.

What it can look up is narrower than "my chats" sounds:

• Only chats you own. It does not see a colleague's chats, and it does not see a chat someone else has shared with you. You can still open a shared chat from the Shared tab. See Share a chat and see team history.
• Chats on the web, on WhatsApp and in Microsoft Teams. Each result says which of those it came from.
• The title, the last message, and the text of your messages and Centrus's replies. It matches words, not meaning. A word has to be at least two letters, so "boiler fault" finds those words and will miss a chat that only says "heating unit".
• Your most recent chats. It reads up to 80 of your latest chats that are not archived, skips any that are agent runs, scores up to 40 of what remains, and returns up to 8 matches.
• A short extract, up to 280 characters. If that is not enough, it can open the chat and read the last 30 plain-text messages, each cut off at 800 characters.
• A date window, when you say when the chat happened. The date on a result is the day the chat was last updated, written as DD/MM/YYYY.

What it cannot look up:

• Archived chats, and chats that are agent runs. Archive a chat and this search stops seeing it. The chat is still in your archived list.
• The current chat, when it is searching your other chats. The current chat has its own search, used when you mean something said earlier in this conversation.
• Photos, files you attached in a chat, or documents in Train. Documents are the separate file search, and that one does follow the tags on the chat. See Control who sees what with tags.
• Tags. The search does not read the tags on a past chat, and it does not check the tags you hold now. A chat you started on Engineers can still be found from a chat on Office, and it can still be found if Engineers is later taken off your user. What comes back is the text already stored in that chat, not a fresh look at the files.
• A search phrase longer than 400 characters. Anything past that is ignored.

Chat history search is not a memory you can rely on for a standing rule. It only sees recent chats, it matches words, and a later answer can miss the chat entirely. A fact you want on every future answer belongs in a lesson, or in Personal AI Context if it is about how to answer you. Lessons are in Teach Centrus from a chat.

The date is worked out in universal time. A chat updated just after midnight in the UK can show the previous day's date. Treat the title as the surest way to recognise the chat, and open it yourself if the extract is not enough.

Centrus decides the model, voice and picture settings again on every answer. Your choice wins when you have made one. Otherwise it uses the organisation's choice. If that is missing too, it uses its own default. There is no third choice stored on a single chat. The AI Model item in a chat's gear menu writes Personal AI Model, the same field as on this page.

• Model. Your Personal AI Model, then the organisation's model, then GPT-5.6 Terra. Shared chats use the owner's model. An agent can be given a model of its own in the Agent Builder, and that model is used for the agent instead of these settings.
• Voice. Your Personal Live voice, then the organisation's voice, then Vesper. It applies the next time you start Voice mode.
• Pictures. Your image model, then the organisation's, then Flare. Your quality, then the organisation's, then Auto. Words in the message that name a quality, such as high quality or draft, replace the saved quality for that message only.
• Temperature. Your stored number when the model that answers accepts temperature, otherwise the organisation's number, otherwise 0.50. GPT-5.2 and GPT-5.1 accept it only at reasoning None. Most models do not accept it, so the number is ignored.
• Context. Organisation context, then the context for the tags on the chat, then your Personal AI Context. Yours is added. It does not replace the others. Empty means that layer is skipped.
• Reasoning. The brain button on the message bar, for this browser tab only. It is not one of these settings, and it is not saved. WhatsApp and Microsoft Teams always use the model's starting level.

A change to your model, voice, images or context applies from the next message after it is saved. It does not rewrite old answers. Colleagues who have left their own lists on Use organisation follow the admin's settings, and colleagues who have chosen their own keep those. Who is affected by an organisation change is set out in Who a change affects.

To follow the organisation again, set each list back to the organisation choice, clear Personal AI Context, and click Save changes. There is no single reset button on the Account tab.

1. 1Set Personal AI Model to Use organisation AI model.
2. 2Set Personal Live voice to Use organisation Live voice.
3. 3Set Personal image model to Use organisation image model, and Personal image quality to Use organisation image quality. The brackets still show the organisation's current Flare, Sunburst, Auto or other choice, so you can see what you are returning to.
4. 4Delete the text in Personal AI Context.
5. 5Click Save changes, and wait for Updated. If the context box changed, you also see Your AI context has been updated successfully.

Temperature cannot be pointed back at the organisation from this page. If the slider is visible, it still holds a number, and saving the form stores that number. For models that do not accept a temperature, the number is simply not used.

Admins also have Reset to default on the AI Model tab. That button clears your personal context and puts your personal choices back to the organisation's, and it also replaces the organisation's own context and model. Nothing happens until Save changes. If you press it by mistake, click Discard changes before you save. What it puts back is described in Set your organisation's model.

Chat history search is not part of this reset. Click Off or On on its own switch. That saves immediately.

The Account tab is the same page on a phone. The menu is the bar at the top, the Settings tabs scroll sideways, and the form stacks in one column. Chat history search and Personal AI Settings are still below Theme. The model list, when you open it, uses most of the screen width and shows the model name and the four marks on one row.

09:41

Centrus

Settings

Update your account details

Chat history search

Let Centrus look up your other chats when you ask it to. This only affects your account.

OnOff

1

Personal AI Settings

Personal AI ModelUse organisation AI model▼

2

Temperature is not supported for the selected reasoning model.

Personal Live voiceUse organisation Live voice

1. 1Chat history search. On and Off are large enough to tap, and the choice saves straight away.
2. 2Personal AI Model. Open it to pick a model, then scroll down for images, context and Save changes.

Saving works as on a computer: Updated for the form, and Your AI context has been updated successfully. when the context changed. Voice mode on a phone uses the voice you saved here. The sample button is the same play control beside the list. Changing the model from a chat on a phone is covered in Change the model on a phone.

These are the messages on the Account tab for personal AI settings and Chat history search. Messages from the model window inside a chat are in Choose the AI model.

Anyone with an account can set their own personal AI settings and turn Chat history search on or off. Nobody can change someone else's from this page.

People without Train access still have these settings. Train access is about files, tags and skills, not about the model on your account. To change who is an admin, see Edit users and roles.