Gleap's bill has two parts, and they behave differently. The plan is a fixed subscription for the product. AI is usage-based, paid from credits.
Four tiers, each including everything below it.
Starter — one seat, one project. The messenger, Kai, live chat, in-app bug reporting, the roadmap and the knowledge base.
Team — unlimited seats and projects, every support channel, custom domains, integrations and outbound messaging.
Pro — adds the deeper AI: custom agents, a choice of models rather than automatic selection, Kai PM, Kai Code and Kai Resolve, plus a discount on AI usage.
Enterprise — the compliance and control tier: SSO, HIPAA BAA, DORA support, bring-your-own-key, payment by invoice, a dedicated contact and an SLA.
Current prices are on Gleap's pricing page. Annual billing costs less than monthly, and new accounts start with a free trial.
AI isn't included in the subscription; it's paid from a credit balance and billed on what's actually used — the volume of text processed, and which model handled it.
The billing page shows the balance, what's been consumed, and a breakdown you can export as CSV. That breakdown is the thing to look at when a month is more expensive than expected: it shows which AI services the spend came from.
Two consequences worth internalising:
A quiet month costs less than a busy one. AI spend follows conversation volume.
The model matters. A larger model on every conversation costs more than one chosen for the job, which is what the automatic model selection is for.
You can add credits at any time. Better, turn on auto-recharge: set a threshold and an amount, and Gleap buys more when the balance drops below it.
There's also a usage limit, which caps what can be spent. It exists so a misconfigured automation or a sudden spike can't produce a surprise. If you're hitting it legitimately, raise it from the same page.
Limit and balance are different things: the balance is what you've bought, the limit is what you allow.
Most overspend comes from the same places: an AI agent answering things a help center article would have handled, a model heavier than the task needs, or automations doing AI work on tickets nobody reads.
The usage breakdown tells you which of those it is faster than any amount of guessing.
