AI Assistant (Beta)

The AI Assistant is an agentic chat over your whole AutoPersonas account. Open Assistant in the sidebar, ask in plain language, and it operates the platform for you: creating and editing influencers, generating images and video, managing the content queue, scheduling, digging through analytics, and more. It's available on every plan, including Free, billed pay-per-use by tokens.

What it can do

The assistant has access to roughly 190 platform tools, the same surface the REST API and MCP server expose. In practice that covers:

• Creating and editing characters and profiles.
• Generating images and videos, and filling the content queue.
• Reviewing content: approving, rejecting, editing captions, rescheduling.
• Scheduling posts and picking publish times.
• Managing brand goals and engagement settings.
• Answering questions from your analytics and billing usage.

Example prompts (the same ones suggested on the empty thread):

• "Generate 3 post ideas for Ava"
• "What's waiting in my review queue?"
• "How did my recent posts perform?"
• "Schedule my approved posts for the best times"
• "Which of my profiles got the most engagement?"
• "Fill the content queue for next week"

You don't need any profiles yet to use it. On a fresh account the assistant switches to onboarding mode: it can walk you through how AutoPersonas works, help you pick a niche, and create your first influencer with you.

Rich results in the thread

Results render inline as real UI, not text dumps: generated images and videos appear as media, queue lookups render as content cards, generation batches show live progress, and analytics queries come back as charts and stat summaries. Profiles and schedules get their own cards too.

Confirmations & safety

The assistant reads freely, but three classes of action always pause the turn and ask you first, with an in-thread approval card:

• Destructive, deletes and disconnects that can't be undone.
• Publish, anything that posts publicly on a connected platform.
• Spend, generation that bills your account. Spend cards show a cost estimate before you approve.

Nothing in those classes executes until you approve, and the server executes exactly the call you approved, the card's arguments are not editable after the fact. When the assistant is missing information, it asks a clarifying question with one-tap preset answers plus a free-text box.

Conversations

Conversations persist across sessions. You can switch between them, rename them, and delete them from the conversation list. Each conversation shows its running token usage.

Pricing

The assistant is available on every plan, including Free. There is no subscription gate, but you need a payment method on file because usage is billed pay-per-use by tokens:

A typical back-and-forth costs a fraction of a cent. Actions the assistant takes on your behalf (image generation, video generation, publishing) bill at the same per-action rates as doing them yourself, see Billing & Plans. Chat usage counts toward your optional monthly budget cap like any other metered action.

Limits & troubleshooting

• Daily spend cap: chat spend is capped at $10 per day (UTC) as runaway protection. The cap resets automatically at the start of the next day.
• Rate limits: up to 10 turns per minute, and each turn is bounded in how many tool rounds it can run before it reports back.
• "Add a payment method" prompt: the assistant needs a card on file (or an active subscription) before the first message. The setup flow is Stripe-hosted from the assistant page itself.
• Account suspended: if billing is suspended the thread opens read-only until it's resolved.
• The assistant is in Beta. It confirms anything risky before acting, but review what it queues the same way you would review any generated content.

Related

• Content & Scheduling, what the queue, review, and scheduling flows do.
• Custom Flows, build your own generation pipelines (the assistant operates the standard pipeline).
• Billing & Plans, usage rates and budget caps.
• MCP, drive the same tool surface from an external AI agent instead.