Personal and custom identities, from zero to an identity that sounds like you

If you are brand new to Eden and just want to ship one good identity, start at the building your identity walkthrough. This article is the reference for everything underneath that, written so a beginner can read top to bottom and a returning user can jump straight to the section they need.

What an identity is, and why it matters

Most AI writing tools let you "set a tone" with a few sliders or a paragraph of instructions. That works for one-off prompts and falls apart the moment you want consistent output across sessions, formats, and topics. The output drifts because the model has no anchor: no point of view to argue from, no vocabulary to lean on, no audience to write to, no examples of how you actually phrase things.

An identity in Eden solves that by storing a deeper profile. Not "professional but friendly". The actual mission you write under, the worldview you write from, the people who shaped your thinking, the words you use verbatim, the structures you default to, the audience you write for, the moves you avoid. The model reads that profile on every drafting turn and produces text from it.

The practical result: when the chat hands you a draft, it sounds like you wrote it on a focused day. Not a generic AI version of you.

Personal identity vs custom identities

You always have exactly one personal identity. It is private, lives on your user, and follows you across every workspace you join. This is the identity for everything you publish under your own name (newsletters, posts, threads, talks, anything that goes out as you).

You can have any number of custom identities per workspace. Custom identities are workspace-scoped (everyone in the workspace can pick them in their composer) and are meant for things that are not "you", such as:

• A brand identity for the company you run.
• A separate identity for a podcast you co-host.
• A client's identity if you write for them.
• A character or persona you publish under that is intentionally distinct from your personal identity.

The two kinds use the same profile shape and the same chat surfaces. The only difference is scope and intent. If something speaks for a brand, a client, or a separate persona, it belongs on a custom identity. Everything else belongs on your personal identity.

The four ways to build an identity

Eden gives you four entry points so you can match the effort to what you actually have on hand right now. Each one produces the same kind of saved profile; they differ in how much input you provide upfront.

1. 01

   Build in chat (Build my identity)

   A guided AI conversation that asks specific, story-shaped questions and assembles your profile from your answers. Best when you do not have writing samples to lean on yet, or when you want the model to interview you. Walks you through mission, point of view, core ideas, vocabulary, personality, and audience in 10 to 20 minutes. See the building your identity walkthrough for the full flow.

2. 02

   Paste links of your content

   You paste links to anything that already represents you (YouTube videos, podcasts, X threads, newsletters, blog posts, Substack issues, Loom recordings, LinkedIn posts). Eden pulls the transcripts and text, sends them into a guided identity-building chat, and the model proposes a complete profile based on what it found. You confirm with one click. Best when you already publish and want to skip the interview.

3. 03

   Start from a template

   Pick one of six archetypes as a working draft and refine from there. Best when you want a real starting profile in a couple of clicks and plan to edit it by hand or refine it in chat afterward.

4. 04

   Edit the fields by hand

   Skip the chat entirely. Open the Identity tab and fill in the fields directly. Best when you already know exactly what you want each field to say.

For your personal identity, all four are available the first time you open the Identity tab. For custom identities, the entry points are the same (Build in chat, Paste links, Start from template, Edit by hand) and live in the Create custom identity modal on the same page.

What is on an identity profile

An identity is not just a tone description. It is a structured profile with fields that map to the different things a model needs to know to write like you. Some fields are required for a useful first draft; most are optional and get richer over time. Every field is editable by hand at any moment.

The essentials

These are the fields that carry the most weight on drafting turns. A profile with these filled in is enough to start producing real output.

• Mission. One sentence describing what you are here to help people do, become, or understand. The job-to-be-done line for your work, written in your own diction.
• Point of view. A paragraph capturing your worldview: the unique combination of influences and lived experience that produces your angle, what default mode you write against, what you treat as obvious that most people do not. This is where the signature lives.
• Core ideas. The perspectives (not topics) you want to be known for. One idea per line, phrased the way you would say it. Capture every distinct one. The field has no practical cap, so do not curate.
• Vocabulary. Phrases lifted verbatim from your own writing or speech. This is the single most powerful field for making generated drafts sound like you. The model reaches for these words first.
• Audience. One or two sentences naming exactly who you write to. Specific person, not "creators" or "founders". What they want, what they bristle at, what would make them close the tab.
• Personality. A paragraph naming WHO you are when you write. Temperament (calm, restless, mischievous, earnest), what charms you, what makes you angry, what you are allergic to, where you are warm and where you are sharp.

The deeper layers

These fields are what take an identity from "sounds vaguely like me" to "I would have written this on a good day". They are optional on day one and grow as you use it.

• Influences. The thinkers, writers, or creators whose lens shaped yours. The model uses these to understand the intellectual lineage of your point of view.
• Tone tags. Short descriptors of how your writing sounds at the sentence level. "Blunt but not cruel", "warm with strangers, sharp with peers", "drily funny".
• Rhythm. Punchy, flowing, or mixed. The default cadence the model reaches for unless told otherwise.
• Format habits. Observations about how you structure pieces. "Opens with a scene, never a thesis", "ends on a question, not a directive", "uses one-line paragraphs for emphasis".
• Prefer list. Patterns you actively want the AI to reach for. "Use second person", "lead with the stake", "name the cost before naming the upside".
• Avoid list. Patterns you want banned. "No em dashes", "no 'in today's world'", "no guru tone", "no rhetorical questions as section breaks". The model treats these as hard rules.
• Anchor stories. Short labeled vignettes from your lived experience. Each one is a slug-style label (month-14, the-call-from-naval, the-room-where-everyone-nodded) plus a paragraph of your actual telling. The model reaches for these as concrete evidence when writing as you, instead of fabricating generic moments.
• Format scaffolds. Per-format structural patterns mined from your actual content. One scaffold per format you ship (newsletter, X thread, long-form YouTube, LinkedIn post, podcast intro). Each carries a macro arc (how the piece opens, develops, and closes) and a list of micro patterns (hook archetypes, intro shapes, transitions, evidence patterns, closes). When you say "write me a newsletter", the model writes in the scaffold for that format, not a generic newsletter.
• Notes. A free-form scratchpad. Biographical context, season-specific shifts ("voice has gotten sharper this year after the divorce"), one-off don'ts that do not fit anywhere else, links you want the model to be aware of.
• Writing samples. One to twenty short paragraphs of your actual prose, copied verbatim from things you authored. The model uses these as ground truth for cadence. This is the field most other tools cannot replicate because it requires storing real text, not instructions about text.

Why this profile shape produces a more accurate identity

The fields above are not arbitrary. They map to the four things a model has to get right to sound like a specific person.

1. The intent layer. Mission, point of view, audience, core ideas. These answer "what is this person trying to do, for whom, from what angle". Without them, the model defaults to a generic helpful tone.
2. The identity layer. Personality, influences, anchor stories. These answer "WHO is talking". Without them, the model produces words a thousand people could have written.
3. The surface layer. Vocabulary, tone tags, rhythm, prefer, avoid. These answer "how does it sound at the sentence". Without them, the model uses its default phrasing, not yours.
4. The structural layer. Format habits, format scaffolds, writing samples. These answer "how do they organize a piece". Without them, a newsletter and a thread come out with the same shape.

Tools that only let you set a tone are working on layer 3 and ignoring the other three. The result is recognizable as AI even when the tone is right, because the intent is generic, the identity is missing, and every format comes out shaped the same way. Eden's profile covers all four layers because that is what it takes to produce a draft you would actually publish.

Drafts vs Persona mode

Every identity runs in one of two modes. The mode controls when your identity kicks in during a chat.

• Drafts mode. The AI stays itself in chat. It answers questions, brainstorms, and explains things in its own neutral voice. Your identity kicks in only when the model is producing prose you would publish (a post, a draft, a thread, a script). This is the default and the right pick for most chats. You want a sharp assistant that hands you finished writing in your voice, not an assistant pretending to be you for every reply.
• Persona mode. The AI replies AS the identity for every turn. Answers, brainstorms, comments, all in character. Best when you want to talk to the identity, not have it draft for you. Rubber-duck a problem against your own published worldview, see how your brand identity would respond to a customer email, or work through a thought in character.

Set the default mode on the identity in the Identity tab. Flip it per-chat from the chat composer's chip if you need the other mode for a single conversation. Changing it on the chip does not change the saved default.

Where your identity shows up in the app

Identities are a workspace primitive, so they appear in a handful of places:

• The chat composer. Every chat carries a "My identity" chip showing which identity is active and which mode it is in. Click it to switch identities, change the mode for this chat, or open it in the Identity tab.
• The "+" menu on the composer. Click the plus button on the chat input and you get a "Choose identity" row. The row carries a status chip showing the active identity's state at a glance.
• The Identity tab. The full editor for your personal identity, the list of custom identities in the workspace, and the Build my identity and Create custom identity entry points.
• The picker in chat. Open the chip in any chat and you see a list of every identity you have access to (personal + every custom identity in the workspace). Click one to make it active for that chat.
• Save confirmations. Every time an identity is saved (via Build my identity, Paste links, or any in-chat update), the toast confirming the save carries an Open in Settings action that deep-links you straight into that identity's editor.

Refining an identity over time

Identities are not one-shot artifacts. They get better the more you use them, because every chat surfaces new vocabulary, new angles, and new patterns the model can save back into the profile.

Edit the fields by hand for small tweaks

Open the Identity tab (or click any identity in the picker). Edit the fields directly and click Save changes. This is the fastest path for "drop that phrase from my vocabulary" or "update my mission to X".

Say "update my identity" in any chat

If you notice something while you are in a chat, just say it:

"Update my identity. Add 'non-negotiable decisions' to my core ideas and drop the 'systems thinker' tone tag."

The model produces a Save to my identity card with your existing profile merged with the requested change. Every untouched field is carried through unchanged. Accept the card and your identity updates. This works the same way for custom identities: name the one you mean ("update the Newsletter identity and add...") and the chat saves to the right one.

Refine in chat for a larger refresh

If a lot needs to change at once, open the identity in the Identity tab and click Refine in chat. This opens a fresh guided chat with the current identity loaded as context. The first turn asks what you want to change. If you do not know, the model offers three specific angles based on what is actually in your saved profile. You refine from there, and the model saves the merged profile when you are done.

Add more source material as you publish

The single best way to keep an identity accurate is to feed it more of your real writing as you produce it. Every quarter or so, paste links to your latest five or ten pieces in a chat and say "update my identity with what you find in these". The model mines the new material, lifts fresh vocabulary, refines the format scaffolds for any format you have shipped more of, and proposes a save card you can accept.

Versions, rollback, and clearing

Every save to an identity (whether by hand, by chat, or by guided build) snapshots the prior version into its history. Open the identity in the Identity tab and scroll to the Recent versions list. Click any entry to preview it, and click Restore to roll back. The current version is snapshotted before any restore, so rolling back is itself reversible.

To start completely over on an identity without losing the history, click Clear identity at the bottom of the editor. The current profile is snapshotted into history first, so you can restore it if you change your mind.

Privacy and scope

• Personal identity is per-user. It is never visible to other users, never shared across workspaces, and never leaves your account.
• Custom identities are workspace-scoped. Anyone with access to the workspace can pick them in their composer. Editing a custom identity is gated by workspace permissions the same way every other workspace artifact is.
• What we store. The profile fields you have filled in, the version history, and the source attribution (manual edit, guided chat, paste-links extraction, template). Nothing else.
• What we log. Counts only (how many users finish their identity, how many custom identities a workspace has). The contents of your profile are never logged or analyzed in aggregate.

Tips for an identity that produces drafts you actually ship

A few patterns from identities that produce the best output:

• Be specific about your audience. "Creators" is not an audience. "A 32-year-old solo operator running a $40k MRR newsletter from their kitchen" is. Specificity in the audience cascades into every other field.
• Make every core idea pass the book test. If you could not write 50,000 words defending it, it is a take, not a core idea. Set it aside.
• Lift vocabulary from your own writing, never from suggestions. The most powerful field for matching your voice is the vocabulary list, and the only useful entries are phrases you have actually used. A generated list of "phrases that sound like you" defeats the purpose.
• Use the Avoid list aggressively. This is where you ban the patterns that bother you. The model treats these as hard rules and they do more to shape the identity than any positive instruction.
• Add anchor stories as they come up. Every time you tell a real story in a chat or a draft, ask the model to save it as an anchor story. Over time, the model has a catalog of your actual material to reach for instead of fabricating generic examples.
• Mine your own content for format scaffolds at least once. Paste links to three or four of your strongest pieces in each format you ship and let the model reverse-engineer the macro arc and micro patterns. The drafts you get back will follow your structural defaults, not a generic template.

Common questions

Do I have to use the chat to build an identity? No. Every path ends in the same editable profile, and you can skip the chat entirely by editing fields by hand. The chat exists because most people get to a sharper profile faster by being interviewed than by filling out a form.

Can I have an identity that is on for some chats but off for others? Yes. Toggle the My identity chip on the composer to turn the active identity off for the current chat. Toggle it back on for the next chat. The identity itself stays saved and active globally.

What happens if I pick a custom identity in a chat? That chat now writes in the custom identity instead of your personal one. Switching back is one click on the chip. The custom identity's mode (Drafts or Persona) takes over for that chat.

Can two identities share a profile? No. Each identity has its own profile. If you have a personal identity and a custom identity for your newsletter, they each carry their own mission, vocabulary, scaffolds, and so on. The newsletter identity can be more constrained or more brand-y than your personal one, which is the point.

The AI's writing still does not sound like me. What now? Two common causes. First, your Vocabulary list is empty or thin. Add five to ten phrases you have actually used. Second, your Avoid list is empty. Add the patterns that bother you. Your voice is shaped at least as much by what you ban as by what you affirm. Past that, the highest-leverage move is to add writing samples (one to three short paragraphs of your real prose) so the model has ground truth to anchor to.

Where to go next

• For the step-by-step walkthrough of the guided chat that builds your first identity, read building your identity with the guided chat.
• For the empty-chat presets that let you remix a board's worth of saved posts in your voice, read chatting with a board.
• For the per-message boost actions (rewrite in my voice, headline variations, find blind spots), read using boosts.