Agent & tasks · 13 min read · Updated 2026-07-30
What is an AI persona, and when should I create one?
An AI persona is an identity for an AI teammate. Learn how to create one, write its instructions, and know when you need another.
In TabTin, every task is taken by a specific persona.
AI personais an identity for an AI teammate, with a name, avatar, instructions you write, a set of capabilities, and memories built while working with you. You are not speaking to a generic “AI,” but toone specific teammate. That means your writing persona can always deliver in your preferred format while your research persona always cites sources, without the two interfering.
First: switching personas is not switching models
This is the easiest distinction to miss.
| Switch model | Switch AI persona | |
|---|---|---|
| What changes | The engine (which model provider does the reasoning) | The teammate (who takes the task) |
| What follows | Speed, cost, and strengths | Name, persona rules, equipped Skills, and memory |
| Where to change it | In the composer: Model selector | In the composer: AI persona button |
If you only want a faster or stronger answer, switch models. You do not need another persona.
You start with five AI personas
When you first enter an organization, TabTin creates a starter lineup for you:
| Persona | What it handles |
|---|---|
| Tin(marked Default) | A catch-all for everyday work that does not clearly belong elsewhere |
| Code | Works inside repositories: understand unfamiliar projects, diagnose errors, implement fixes, and verify them |
| Writing | Produces reports, proposals, meeting notes, and announcements |
| Data | Works with tables and numbers: create tables, query, analyze, and summarize |
| Research | Researches online: competitor analysis, source collection, and fact-checking |
The template library includes two more outside the starter lineup—Slides(presentations only) andAssistant(processes email, captures memos, and schedules reminders). Create either from a template when needed.
This lineup belongs to you in the current organization. If you disable one, the system does not silently recreate it.
Where to view and manage personas
- In the desktop app, findthe narrow rail on the far leftand choose AI personas icon and click it
- The middle column is My AI personas—every persona you have in the current organization
- Select one to open itsprofile
The top of the sidebar also contains two shortcuts:New AI persona and Disabled.
What you can change in a persona profile
| Section | What it controls |
|---|---|
| Avatar + name | The More actions menu beside the name contains Rename and Edit avatar(choose from seven TabTin avatars; custom uploads are not yet supported) |
| Persona & rules | The persona's core setting: “Describe how it thinks and communicates, and the boundaries it must follow.” Up to 5,000 characters |
| Equipped Skills | “What this AI persona can do. Skills stay with the persona wherever it works.” Equip it from the Skill library |
| Memory | Its overall understanding of you (Overview) plus every item remembered during collaboration (All records). You can correct, forget, or export them |
| Recent tasks | “Chat conversations and Project tasks involving this AI persona.” Look here to see its recent work |
Disable AI persona is in the same More actions menu (the default persona does not have this option).
The profiledoes not containmodel settings. A persona remembers the model you choose in its conversation. Select model X while using that persona, and X appears beside its name the next time you open the persona dropdown.
How to create a new AI persona
Use either entry point:
- Left rail AI personas → top of the sidebar New AI persona
- Persona button in the lower-left of the composerAI persona button → bottom of the dropdown New AI persona
Then:
- The dialog is titled New AI personawith the subtitle “Choose a template to start quickly, or create your own AI persona from scratch.”
- Start from a template contains seven template cards, followed by Start from scratch(“No template. Define everything yourself.”). Choose one
- Only when you choose Start from scratchdoes the avatar selector appear. Templates use their own avatar, which you can change after creation
- Enter a name(placeholder: “Name this AI teammate”; up to 100 characters)
- Click Create. “Agent is ready” confirms creation
A persona created from a template starts with brief rules and a set of Skills. They are only a starting point and can be changed at any time.
If creation shows Agent limit reached: each organization allows up to five personas created from scratch. Template-based personas do not count toward this limit. You candisablean unused persona to free a slot.
How to write useful persona rules
Open Persona & rules. The prompt says “What this AI persona does, how it works, and its boundaries…” After writing, click Save. The rules are saved only when “Persona saved” appears.
A useful structure has three parts:Who it is → How it works → What it must not do.
You are our team's external communications writer. Everything you write will eventually go to customers or the public. How to work: - Ask who the audience is and what the piece should achieve before drafting. - Always deliver: one-sentence conclusion → three key points → one recommended action. - Cite sources for numbers and facts. If I did not provide the data, say “Data missing.” Boundaries: - Never invent customer names, amounts, or dates. - Ask me to review any promise, quote, or legal wording before finalizing.
A few lessons:
- Write lasting rules, not one-off tasks. “Write this week's report in format X” belongs in a conversation; “Write every weekly report in format X” belongs in persona rules
- Move repeated instructions into the persona. If you have said “Skip the empty opening” three times, make it a rule
- Boundaries are more useful than praise. “You are a senior consultant” changes little; “Never invent figures; say when data is missing” changes the output immediately
Two layers of rules: general and persona-specific
Before adding a rule, ask whether it appliesonly to this personaorto all your personas?
| Where to write it | Example | |
|---|---|---|
| General rules | Settings → Personal settings → My AI → General rules | Reply in Chinese; lead with the conclusion; avoid unnecessary technical jargon |
| Persona rules | In that persona's profile: Persona & rules | Delivery format, professional boundaries, and the workflow for this kind of work |
The app explains it directly: general rules “are your base preferences; rules for an individual Agent are layered on top and take priority.”
Do not copy a rule such as “Reply in Chinese” into five personas. Write it once as a general rule so there is only one place to maintain.
How to switch personas while working
The button in the lower-left of the composer shows thecurrent persona's avatar and namewith a small arrow. Click it to open the list:
- Each row contains an avatar and name, with gray text below showingthe first sentence of its persona rules(or “No persona set” if empty)
- A model name may appear on the right; it is the last model used by this persona
- At the bottom is New AI persona. After creation, the conversation switches to the new persona automatically
You can switch at any time in a personal conversation, including mid-conversation or while Agent is generating. The new persona handles subsequent turns; existing messages are not rewritten and continue to show the persona active at the time.
You cannot switch in a team Project conversation. The app says “This conversation is executed by XX and cannot be changed after creation.” Ownership of team execution cannot be reassigned ad hoc within a conversation.
How personas, conversations, and memory relate
- A conversation belongs to one persona. The conversation list is grouped by workspace and does not show personas. To review work by persona, openPersona profile → Recent tasks
- Memory is isolated by persona. The Data persona does not know the writing preferences remembered by Writing. This is intentional: one persona should not mix every kind of work into increasingly noisy memory
- Memory has a global switchon Memory the app's Memory preferences named Let Agent take notes. When disabled,allpersonas stop storing and recalling memories
- You can manage individual memories: under Persona profile → Memory → All entries, each memory can be corrected / forgotten or marked useful, and all memories can be exported
Practical patterns
Pattern 1: divide by type of work, not by project
What to do: code goes to Code, writing to Writing, tables to Data, and online research to Research. Code work from Project A and Project B both stays with Code.
Why it works: repeatedly giving one type of work to the same persona concentrates the right experience in memory—your code style, prior mistakes, and acceptance standards. Accuracy improves over time, and equipped Skills remain focused.
Common mistake: creating a persona for every project, such as “Project A assistant” and “Project B assistant.” You quickly exhaust the custom-persona limit and fragment experience across several personas. Project and workspace boundaries isolate projects; personas should not.
Pattern 2: enforce a delivery format with persona rules
What to do: for repeated formats such as weekly reports, competitor tables, or meeting notes, put the exact format in the persona: field order, citation rules, and what to do when uncertain.
Why it works: you no longer repeat format requirements in every conversation or receive unusable work because you forgot. Resolve team debates once in the persona rules.
Common mistake: putting a one-off requirement in persona rules. “Use the template from this redesign” may still apply six months later after you have forgotten it. The test is simple: will you still say thisnext month? If yes, make it a persona rule; if not, say it in the conversation.
Pattern 3: separate external and internal personas for the same work
What to do: an External Writer uses conservative language, cites every number, and asks for review before finalizing. Internal Notes optimizes for speed and may leave todos and gaps. Both write, but their risk levels are entirely different.
Why it works: repeating these differences in every conversation is tedious and easy to forget—and the forgotten case may be the document sent to a customer.
Common mistake: creating a new persona only to change tone for one turn. Tone, length, and level of detail are one-off preferences you can state in the conversation.
Pattern 4: know whennotNew AI persona
to create one. None of these needs a new persona:
- You only want another model — switch models in the conversation; keep the persona
- You want a simpler answer this turn — say so in the conversation
- You want it to work in a directory or on a device — that is a workspace choice, not a persona choice
- You want to split a large task into parallel workstreams — use PMO and sub-Agents, not another persona
- A teammate wants the same setup — personas are private; they need to create their own, as explained next
How personas work in a team
AI personas are private to you by design. A persona carriesMemory: your preferences, corrections, and the history of your work. Those details do not transfer cleanly to another person, so personas are not shared. Colleagues in the same organization cannot see one another's personas. Attempting to manage someone else's Skills is blocked with: “Only the owner of this AI persona can view and manage its Skills.”
Personas are also separated by organization. Switch organizations in the top bar and the persona list changes: each organization has its own starter lineup, custom-persona limit, and memories. Matching names still represent different personas.
How does a team work with Agent, then?Share tasks, not personas. Share a task so a teammate can see every Agent step in real time. With “View & fork,” they can fork a copy and continue using their own persona and memory. SeeShare an Agent task with a teammate.
The same principle applies in a team Project:assign work to a person; that person chooses which of their personas will do it. The issue first goes to its owner with a status such as Awaiting acceptance. After they confirm it, one of their personas executes in their workspace. The team can follow progress and deliverables in the Project.
Want the whole team to use the same persona rules? Share the text and have each teammate create a persona with it—rules can match while memories stay personal. That is exactly why personas are not shared.
Disable and delete
The default persona cannot be disabled. The persona marked Default, usually Tin, has no disable action. Hover over the badge to see: “This is your default identity and cannot be disabled.”
Disable means hide, not delete:
- The app says: “After disabling, this AI persona disappears from the active list. Workspaces and conversation history remain, and the persona can be restored under Disabled.”
- A disabled persona no longer appears in the persona list or conversation selector
- To restore one: sidebar Disabled → find it → Restore
Permanent deletion really deletes the persona. Only non-default personas in the Disabled list have Delete permanently. The confirmation is explicit: “The identity settings, equipped Skills, and memories for ‘XX’ will be deleted and cannot be recovered. Conversation history will remain.”
If you might need it again, disable it. Delete permanently only when you are sure.
FAQ
I saved persona rules, but the conversation dropdown still says “No persona set”
The gray line is fetched when the dropdown opens.Close and reopen the dropdownto refresh it. Also confirm that you clicked Savein the profile and saw “Persona saved.”
Can I pin a model to one persona?
There is no model setting in the profile. Instead, the persona remembers whichever model you select in its conversation, uses it by default next time, and shows its name in the persona dropdown. To change it, switch models once in a conversation.
Can a teammate use a persona I created?
No. Personas are private and invisible to other members of the same organization. Share the persona-rule text so they can create their own equivalent.
Why did all my personas change when I switched organizations?
That is expected. Each organization has an independent set of personas, including its own starter lineup. Switch back to the original organization to see the original set.
“New AI persona” says “Agent limit reached”
Each organization may have up to five personas created from scratch. Either use atemplate(which does not count toward the limit) or disable an unused persona to free a slot.
Are AI personas and sub-Agents the same thing?
No.Personais an identity—who takes the task.A sub-Agentis a way of working—a persona splits a large task into parallel workstreams through workspace settings or PMO mode. This guide covers personas.
Does disabling a persona remove its conversations?
No. Both disabling and permanent deletion preserve conversation history. Permanent deletion removes identity settings, equipped Skills, and memory, but not past messages.
Can I upload a custom persona avatar?
Not yet. The avatar editor says: “Uploads are not supported yet. Choose one of seven TabTin avatars.”
Still stuck?
If you cannot create a persona, its rules do not take effect, or you are unsure how many personas to use, email contact@larchiveai.comand describe your workflow. We will help you design the setup. Beta users can also ask their contact in the group.