You can create scoped assistants inside AVA — narrower, purpose-built AI personas that your team members or external users can interact with.
What a custom assistant is
A custom assistant is an AVA instance with:
- A specific name and persona
- Access to only the data you specify
- Only the actions you allow
- A custom system prompt that defines its scope and behavior
Example use cases:
- A client-facing assistant that can answer questions about a client's project status but cannot see any other data
- An accounts-payable assistant that has read access to QuickBooks but cannot modify any records
- A field team assistant that can look up job records and log updates but cannot see financial data
Step 1: Tell AVA what you want to build
"I want to build a custom assistant for our clients at Acme Corp. It should be able to answer questions about their project but not see anything else."
AVA will ask about the data scope, the allowed actions, and the persona details.
Step 2: Define the scope
AVA will ask:
- Which data sources should this assistant have access to?
- What actions can it take? (query only, or also update, generate documents, etc.)
- Who will use it — specific team members, or a public-facing link?
- What should this assistant know about itself and its role?
Answer in plain language.
Step 3: Review and test
AVA will show you the assistant's configuration in plain language and offer to open a test session so you can try it before deploying. The test session runs as that assistant, with its restrictions active.
Step 4: Deploy
Custom assistants can be:
- Shared as a private link for specific users
- Embedded as a chat widget on a page
- Restricted to specific team members by email domain
Limits
Custom assistants share your organization's token budget. Heavy use by a custom assistant counts against your organization's $30/million token billing.
Related: Introducing AVA · How to Add a Team Member and Set Their Access