Dawn now applies per-user personalization to replies. A new Account Settings → Personalization tab in Dawn Web lets each user set a preferred name, a short bit of personal context, and a tone of voice — and Dawn picks them up automatically whenever it replies to that user, on any surface where their identity is linked.
The setting belongs to the account, not the workspace. That distinction is what makes it useful: a personal preference like “call me Sam, not Samuel” no longer has to be restated in every workspace, and tone preferences no longer need to be encoded into the workspace agent prompt where they would apply to everyone.
What You Can Configure
The new tab is Account Settings → Personalization in Dawn Web. Three fields, all optional except the tone selector.
The first is a preferred name. If your work account is [email protected] but everyone calls you Sam, this is where you say so. Dawn will use that form of address in replies and in any place it would otherwise default to your email or full name.
The second is what Dawn should know about you — a free-text field, capped at two thousand characters, intended for the kind of stable context you would otherwise repeat in prompt after prompt. “I lead engineering and prefer direct answers with concrete tradeoffs.” “I’m a product manager working mostly with the data team — I read SQL but do not write it.” “I’m in EU/Berlin, so dates default to DD/MM and currency to EUR.” A short paragraph is plenty. Anything you would say once at the start of a long working session and not want to repeat.
The third is tone of voice. Eight presets ship in the box — default, friendly, executive, technical, coach, minimal, pirate, and rubber duck — covering the styles teams ask for most. If none of them fit, the custom option lets you write a short tone prompt of your own (“Be direct, technical, concise, and explicit about tradeoffs”). The custom prompt is treated as preference, not policy: it cannot turn off safety, change tool permissions, or override workspace and agent configuration.
One Setting, Every Surface
Personalization is account-owned and resolved at the backend, not at the client. That means the preferences you set in Dawn Web apply automatically to every other surface where Dawn knows who you are.
A signed-in chat in Dawn Web, a thread in the VS Code extension, a session in the Dawn CLI, or a DM through a channel integration where you have linked your identity — all of them read from the same setting. There is no per-surface re-setup. If you change your tone preset in Dawn Web at lunch, the next message you send Dawn from Slack that afternoon already uses it.
The flip side of that decision matters too. Personalization only applies when Dawn can resolve a real signed-in identity. In a shared Slack channel where Dawn is replying to several people at once, or in an automation context with no human user behind the request, no personal preferences are injected. Dawn does not guess “this looks like Alice” from a display name or an email address in the message body — it requires an actual linked identity. The result is that personalization stays a personal setting, not something that leaks into shared spaces or gets misapplied to the wrong person.
How It Interacts With Workspace Rules
Personalization sits below everything that is policy. Dawn safety, workspace and agent configuration, the workspace AGENTS.md file, and any explicit instruction in the current request all take priority.
The most useful version of that rule is the per-message override. If your saved tone is technical but you ask “answer this in one line”, Dawn answers in one line. If your saved preferred name is “Sam” but the agent has been configured to address everyone formally in customer-facing replies, the agent setup wins. Personalization is the default for ambiguous cases, not a way to bend the rules of the workspace you are working in.
It also stays out of workspace memory. Your personalization is not visible to your teammates, not exported with workspace memory, and not readable by workspace admins through any workspace API. It is your account data, not the workspace’s.
What a Personalized Reply Actually Looks Like
The change is subtle on purpose. You will not see a tone banner or a “personalization applied” indicator. What you will notice is that Dawn stops asking “should I call you Samuel or Sam?”, stops re-discovering that you read SQL but do not write it, and stops drifting back to a neutral assistant voice halfway through a long thread.
Take a concrete example. A product manager has set their preferred name to “Sam”, their about-you to “I run product for the payments team — I care about user-facing behaviour, not implementation details”, and their tone to executive. They ask Dawn the same question in three places over the same afternoon: in a Dawn Web thread, in a Slack DM, and in a VS Code session while reviewing a pull request. Without personalization, three different agents in three different contexts produce three differently-shaped answers; with it, all three lead with the user-facing change, summarise tradeoffs in one or two lines, and address Sam by name. The agent setup, the connected sources, and the skills are still doing the heavy lifting on what the answer says — personalization shapes how it is delivered.
Empty by Default, Easy to Reset
If you do not touch the new tab, nothing changes. The default tone is the same direct, professional voice Dawn has used since launch, and no extra lines are added to the prompt for users who have not configured anything.
When you do save preferences, the form has an explicit Save personalization action and a Reset to defaults action next to it. Resetting clears all four fields and reverts to the same behaviour as a brand-new account. There is no notion of “soft” or partial personalization to reason about — you either have something configured or you do not.
A Note on What Goes In the About-You Field
The about-you field is a free-text channel into your prompt, so it is worth being deliberate about what it carries. Stable role context, time zone, language preferences, and the kinds of decisions you usually make are good candidates. Secrets, credentials, API keys, and sensitive personal data are not — the field is sent to model providers as part of the prompt whenever Dawn is acting on your behalf, and the helper text in the form says so explicitly.
A useful rule of thumb: write what you would put in the first paragraph of a working brief to a new colleague who is going to help you for the next six months. That is the level of context this field is sized for.
What This Replaces and What It Does Not
Before this release, the workarounds were familiar. Some teams put per-user tone instructions in the workspace agent prompt and accepted that this only worked for solo workspaces. Others built their own skill that reached out to a directory to look up names. Most just lived with restating preferences at the top of long threads. None of those scaled across surfaces, and none of them survived a teammate joining the workspace.
Personalization replaces all three by putting the setting where it belongs — on the user, not on the workspace — and resolving it once, on the backend, so every surface picks it up the same way. Workspace agents stay focused on what the agent does for the team. Personal preferences stay personal.
What it deliberately does not replace is workspace memory or memory pages. Memory is shared knowledge: the team’s product, the team’s conventions, the decisions the team has made. Personalization is who is asking and how they like to be answered. The two layers compose — the agent draws on memory for what the answer is and on personalization for who it is for — and keeping them separate is what makes both work without leaking into each other.
Getting Started
Open Dawn Web, head to Account Settings, and you will find the Personalization tab between Profile and Security. Set whichever fields are useful, save, and the next message you send Dawn from any surface where your identity is linked will pick them up.
If you have not linked your channel identities yet, the identity linking guide is the right place to start — once linked, your DMs in Slack, Teams, Telegram, and Discord all become eligible for personalized replies the same way Dawn Web and the VS Code extension already are.