2026-06-03 • By Aradhya Tiwari

Raw, capped AI agent pricing

Why Embage avoids flat per-session chat fees and keeps AI agent pricing closer to raw model usage and completed actions.

PricingAI Agent BuilderCustomer Support

Most AI chat products make pricing look simple by charging per session. That can be expensive for businesses because a two-message FAQ chat gets treated like a long support investigation.

Embage takes a different approach. Chat is not billed as one flat session. Paid plans include a model-provider budget and capped monthly action quotas for the work an agent actually completes: knowledgebase searches, datastore writes, and integration calls.

The problem with per-session chat pricing

Per-session pricing hides the raw cost of the conversation. A visitor who asks "What are your hours?" may only need one short model response. Another visitor may need twenty messages, a knowledgebase search, a ticket write, and a Gmail follow-up.

If both are priced as the same session, the tiny chat is subsidizing the complex workflow.

How Embage keeps pricing closer to raw usage

Embage separates the pieces:

Model usage is governed by the model-provider budget included in your plan.

Knowledgebase searches count only when the agent searches trusted business content.

Datastore writes count only when the agent writes a structured record such as a lead, ticket, booking request, or feedback item.

Integration calls count when the agent uses connected tools such as Gmail, Shopify, HubSpot, or calendars.

This makes the bill easier to reason about. You are not buying vague sessions. You are buying a capped monthly operating envelope for the agent.

A concrete example

Imagine two website chats.

The first chat is tiny:

Visitor: "Are you open on Saturday?"

Agent: "Yes, support is available from 9 AM to 1 PM."

Under flat session pricing, that may still count as one full paid session. In Embage, it mostly touches the model budget. If no knowledgebase search, datastore write, or integration call is needed, no action quota is consumed.

The second chat is real support work:

The agent searches the knowledgebase.

The agent collects invoice IDs.

The agent writes a billing ticket.

The agent drafts a Gmail follow-up for the support team.

That should cost more than the tiny FAQ chat because more useful work happened. Embage exposes that work as capped actions instead of hiding it inside an averaged session price.

Why caps matter

Each plan has clear monthly limits. When a cap is reached, the specific tool stops until the next cycle or an upgrade. The agent can still continue with the tools that remain available.

This protects businesses from runaway costs. It also keeps the pricing honest: no surprise overage bill because one workflow suddenly got popular.

What about voice-to-voice agents?

Voice-to-voice usage is different from chat because it uses real-time speech, transcription, and audio output. Embage plans include voice-minute allotments for voice agents. If the voice agent searches knowledge, writes a datastore record, or calls an integration, those tool actions use the same action caps as chat.