How AI Requests Work (Business)

How AI Requests Work (Business Plans)

hivi's AI runs on hivi's own server. Business accounts use the same AI features as regular users (image extraction, CV parsing, Q&A about your data) but with different limits and priority.

Priority order

When many people are using AI at once, requests are processed in this order, from highest to lowest:

1. $5 regular user
2. $99 business ← top business tier
3. $59 business
4. Free regular user
5. $39 business

Higher rows start sooner and finish sooner under load. Anti-starvation rules guarantee even the lowest tier still makes progress every day — they will not be permanently blocked when the system is busy.

The $5 regular plan being on top is a deliberate product choice. Business plans get strong access to AI capacity through reserved slots and looser per-account limits.

Quick lane vs heavy lane

Industry research shows mixing quick Q&A with large document jobs in one queue hurts everyone. hivi uses two separate lanes:

A pile of CV uploads cannot block a quick Q&A — they sit in different lines.

Limits per account (fair use)

To prevent any one account from dominating, every plan has:

• A cap on running AI tasks at a time.
• A cap on waiting AI tasks at a time.
• A maximum wait before a queued request times out.

Caps are looser on higher tiers. $99 business gets the most generous caps among business plans. We do not publish exact numbers because we tune them based on real load; if you have bulk needs, contact support.

Statuses and reliability

• Every job has a clear status: queued → in_progress → completed | failed | cancelled.
• Duplicate submits (double-click, network retry) collapse into one job.
• Temporary failures retry automatically with backoff.
• Permanent failures (bad file, unsupported format) fail once with a specific reason.
• Cancel is supported while queued or in progress.

"Try again later" — when and why

If the queue is completely full (a system-wide cap) or your account hits a fair-use limit, new requests get a short, plain-language message:

High demand right now — please try again in a few minutes.

This is intentional: silently growing the queue degrades the experience for everyone. Reserved slots ensure $99 business users almost never see this message during normal load.

What to expect under load

We are honest about it: no AI vendor — third-party or self-hosted — can guarantee instant responses during peak usage. What we do commit to:

• Transparent status on every request.
• Plan priority applied fairly within each lane.
• Anti-starvation so lower tiers still progress.
• A plain message when the system is too busy, not a vague error.

Monitoring

Business admins can ask support for queue health data (line length, average wait, failures) for their account. We watch these signals internally and tune capacity before users feel issues.