Create and configure personas

A step-by-step guide to building custom AI personas — system prompts, tool permissions, adapters, and sharing.

A persona defines how your AI behaves in a specific context — its tone, knowledge focus, tool access, and even which model adapter it uses. This guide walks you through creating and configuring one from scratch.

Creating a persona

1. Open the persona selector

   In Oasis, click the persona dropdown at the top of the chat. You’ll see your current persona and a list of any others you’ve created.

2. Click Create New

   This opens the persona editor where you configure everything about the new persona.

3. Name it

   Give your persona a clear name that reflects its purpose — like “Research Assistant”, “Code Reviewer”, or “Creative Writer”.

Configuring the system prompt

The system prompt is the core instruction that shapes the AI’s behavior. It tells the model who it is, how it should respond, and what it should prioritize.

Tips for effective system prompts:

• Be specific about the persona’s role: “You are a technical writer who explains complex topics in plain language.”
• Set boundaries: “Only answer questions related to the codebase. If asked about unrelated topics, politely redirect.”
• Define output style: “Respond in bullet points. Keep answers under 200 words.”

Research assistant

You are a research assistant. Your job is to help the user find,
summarize, and analyze information. Prioritize accuracy over speed.
Always cite your sources when referencing specific information.
If you're unsure about something, say so rather than guessing.

Code reviewer

You are a senior code reviewer. When shown code, focus on:
correctness, readability, and potential bugs. Be direct but
constructive. Suggest specific improvements with examples.
Don't rewrite entire files — point out the issues and explain why.

Creative writer

You are a creative writing partner. Help the user develop stories,
characters, and dialogue. Match their tone and style. Offer
suggestions, not rewrites. Ask questions to understand their vision
before generating content.

Attaching a LoRA adapter

If you want to specialize the model’s behavior beyond what a system prompt can do, attach a LoRA adapter. Adapters fine-tune the model’s weights for specific domains — like medical Q&A, creative writing, or math reasoning — without downloading a new model.

1. In the persona editor, open the Adapter section
2. Select an adapter from the available list
3. The adapter loads on top of your current base model when the persona is active

See On-Device AI for the full list of available adapters.

Adapters are tied to specific base model sizes. Make sure your active model is compatible with the adapter you’re attaching.

Built-in persona presets

webAI ships with several ready-made personas you can use immediately or clone and customize:

• General Assistant — The default, well-rounded persona with balanced capabilities
• Additional presets are available depending on your version

Switching personas

Switching is instant. Select a different persona from the dropdown and every new response reflects that configuration. The model doesn’t need to reload — only the system prompt, tool policies, and adapter change.

You can switch mid-conversation. Previous messages stay in the chat, but the AI’s behavior shifts to match the new persona from that point forward.

Sharing personas on the network

If you enable it, your persona can serve as a resource on the Intelligence Delivery Network. Approved contacts can query your persona remotely:

• Their question arrives at your device
• Your persona processes it locally
• Only the answer is sent back

Your data, files, and full conversation context stay on your machine. You control who has access and what context the persona draws from.

Learn more

Personas feature overview

What personas are and why they matter.

Choosing a model

Pick the right base model for your personas.