Skip to main content

AI Credits

AI usage on SendSeven is measured in AI credits: messages written by AI bots and Conversational Agents, the tools they use, and AI features for your team. One AI credit costs EUR 0.01. AI credits are billed separately from messages, but included messages your plan has left over in a month also cover AI credits (see Invoice and included pool).

SendSeven Cowork, the in-app assistant, and voice dictation are on every plan and use AI credits. Conversational Agents, AI bots, Create a flow with AI, the Knowledge Base and the other AI features are available on Professional, Scale and Enterprise. There is no AI add-on and there are no per-user fees.

How AI pricing works​

  • AI is charged by the AI work it does: a short task costs a fraction of a credit, a long and complex one costs more.
  • Knowledge Base search and translations on Fast are free. Translations on Advanced are charged by usage.
  • Building with AI (Create a flow with AI, the AI agent builder) is charged by usage, typically 5 to 26 credits per build.
  • Advanced uses a stronger model that thinks more before it answers. A task on Advanced costs about 3.3 times as much as the same task on Fast.
  • Failed or cancelled AI runs are not charged.

Prices​

Typical credits per use. The real cost depends on how much work the AI does for a task, so these are typical values, not fixed prices:

UsageFastAdvanced
A simple question to an agent (no web search or files)under 1about 3 times Fast
A typical Conversational Agent reply (also a draft in Assist mode)about 1 to 2about 3 to 8
A reply suggestion (Reply with AI, Write with AI, Improve, Team Chat /ai)under 1about 3 to 5
A question to Coworkabout 2about 7 to 10
Translating a message, email, flow, bot or WhatsApp template text (the translation level in Settings > AI)freeabout 0.6 per text, up to 2.5 for a long email
A conversation summary (when a conversation is closed, or on request)about 0.3 to 0.8about 1.3 to 3
Reading a customer's photo / PDFabout 0.5 / 0.8about 0.8 / 1.4
Transcribing a voice message (the Transcribe button, the transcript API, or an agent reading a voice note)about 0.25 for a 30-second note, 0.5 per minuteabout 0.9 for a 30-second note, 1.6 per minute
Create a flow with AI (always Advanced)about 15 to 26
AI agent builder: one build (always Advanced)about 12 to 18
Generate a WhatsApp Flow with AI (always Advanced)about 24 to 31
AI file summary in a conversation (POST /api/v1/attachments/{attachment_id}/summary, the Attachments in chat level in Settings > AI)about 0.5, up to 2 for a large file, once per fileabout 2, up to 5 for a large file, once per file
Voice dictation in the app (always Fast)about 0.1 for a short dictation, about 0.4 per minute of audio
Web search or live page read by an agent3 each3 each
Adding knowledge (crawl, upload, re-import)1 per 10 pages1 per 10 pages

An agent opens a customer's photo or PDF only when the conversation needs it. A file it never opens costs nothing. For adding knowledge, a PDF counts its real pages. Other files and web pages count one page per 5,000 characters (at least one). When a website is crawled again, only new or changed pages are charged.

Credits are counted in fractions per use, for example 0.4 credits. There is no minimum per request. Only the monthly total on the invoice is rounded. The same list is in the app under Settings > Billing > AI credit prices.

Free:

  • Knowledge Base search (/kb, the search API, bot and agent lookups). The answer the AI writes from the results is charged.
  • Translations on Fast (messages, emails, flows, bots and WhatsApp templates), and language detection.
  • Failed or cancelled AI runs: an agent reply, a Cowork answer or plan step, an agent build or a WhatsApp Flow generation that fails, times out or is cancelled. Messages sent after a failure on our side (apology, fallback or hand-over notice) are free too.
  • Keyword bots and Flow send nodes. They pay the normal message charge only.
  • Bot test chats on paid plans, and the agent test console on paid plans up to 500 test turns per day per billing account (beyond that a test turn is charged like a live turn).
  • File summaries that could not be made, and showing an existing summary (including one an agent already wrote when it read the file).

Fast and Advanced​

LevelGood for
FastQuick and economical: common questions, short replies, reply suggestions.
AdvancedA stronger model for harder tasks: long or tricky conversations, multi-step requests, building flows and agents. Costs about 3.3 times as much per task as Fast.
  • Per task: Settings > AI in the app. Reply suggestions, translation, conversation summaries and Attachments in chat (AI summaries of files in a conversation) default to Fast. Cowork defaults to Auto. Voice message transcription defaults to Advanced. The voice message transcription level applies to every transcript in the workspace, including voice notes a Conversational Agent reads, whatever the agent's own level. The translation level applies to every translation in the workspace: messages and emails (the translate button, the composer and automatic translation of outgoing messages), and also flow texts, bot messages and WhatsApp template texts.
  • Cowork on Auto: by default Cowork runs on Auto: it picks Fast or Advanced for each question, depending on how hard the question is. You can set it to Fast or Advanced instead.
  • Plan limits: the translation level runs on Advanced only while translation is available to the workspace (Professional, Scale and Enterprise, with AI features and Chat Translation switched on). Transcription, summaries and Attachments in chat run on Advanced only while AI features are available and switched on. Otherwise they run on Fast.
  • Per agent: the agent's model_tier. standard is Fast and the default. advanced is Advanced.
  • Always Advanced: creating an agent with AI, Create a flow with AI and Generate a WhatsApp Flow with AI. Always Fast: voice dictation, language detection, tagging and similar background tasks.

AI usage API​

GET /api/v1/billing/ai-usage returns the AI usage of a billing period per feature group: how many AI requests were made and the AI credits they cost. It takes the same scope and period parameters and the same access rules as the usage breakdown API.

{
"scope": "account",
"period": 1,
"period_start": "2026-11-01T00:00:00+00:00",
"period_end": "2026-11-30T23:59:59.999999+00:00",
"totals": {
"calls": 1820,
"credits": "1240.37",
"credit_split": {
"fast": {"input": 610.2, "cached_input": 48.115, "output": 402.9, "total": 1061.215},
"advanced": {"input": 92.4, "cached_input": 6.155, "output": 80.6, "total": 179.155}
}
},
"groups": [
{
"key": "agents", "calls": 950, "credits": "980.12", "credit_split": {"...": "..."}, "free": false,
"items": [
{"key": "agent:6f1c...", "action": null, "agent_id": "6f1c...", "name": "Support agent",
"calls": 950, "credits": "980.12", "credit_split": {"...": "..."}, "free": false}
]
}
]
}
  • groups always lists all seven groups in this order: assistant, agents (one item per agent), drafts, translation, knowledge_base, builders, other. Items of the other groups are per action.
  • credits is a decimal string with 2 places. credit_split gives the credits per level (fast, advanced) and by input, cached_input and output, with 3 places. A level without credits is absent. credit_split is null when nothing in the row was charged by usage.
  • free marks rows whose action is free.
  • No euro amounts are returned. Multiply credits by EUR 0.01 for the gross amount before the included pool.

Usage breakdown API​

GET /api/v1/billing/tenant-usage returns the usage of each workspace on your billing account, including an ai_credits breakdown. The full response, with cost per workspace and per channel, is described in Billing for Multiple Workspaces.

ParameterValues
scopeaccount (every workspace on the billing account) or tenant (the current workspace).
period0 = current month to date, 1 = last closed month, 2 = the month before.

Requires the billing:read scope. You must also be the billing account owner or have the account:billing_read permission. Otherwise the API returns 403. An API token must itself include the account:billing_read scope. OAuth apps cannot call this endpoint.

ai_credits appears once per workspace (tenants[*].ai_credits) and once at the top level for all listed workspaces:

"ai_credits": {
"quantity": 1240,
"amount_eur": 12.4,
"unit_price_eur": 0.01,
"bot_messages": {"quantity": 0, "amount_eur": 0.0, "by_charge": {}},
"tool_usage": {
"quantity": 60, "amount_eur": 0.6,
"by_charge": {"web_search": {"quantity": 60, "amount_eur": 0.6}}
},
"other": {
"quantity": 1180, "amount_eur": 11.8,
"by_charge": {"ai_credits": {"quantity": 1180, "amount_eur": 11.8}}
}
}
Bucketby_charge keys
bot_messagesbot_message
tool_usagekb_lookup, web_search, image_analysis, document_analysis, kb_live_refresh, tool
otherai_credits (AI credits charged by usage: agent replies, Cowork, builders, suggestions, summaries, adding knowledge), agent_assist, whatsapp_flow_ai, agent_builder, unlabeled (older usage without a label)
  • Only keys with usage are present.
  • quantity is in AI credits, not messages. AI credits charged by usage are listed under other.by_charge.ai_credits. tool_usage keeps the fixed-price web searches. The other keys only carry usage from before the switch to usage-based AI credits.
  • bot_messages + tool_usage + other equals quantity and amount_eur.
  • The totals match the AI credits category elsewhere in the same response: counts.kb_query and gross_by_type.kb_query. The category key is kb_query for compatibility. It means AI credits.
  • Amounts are gross, before the included pool. The pool is applied once for the whole billing account, so it is not split across these buckets.
  • ai_credits is null if the breakdown could not be computed. The totals in counts stay correct.

Invoice and included pool​

  • One invoice line, "AI Credits". There is no separate line per level or feature. The detail is in the APIs above and in Settings > Usage.
  • Billed in arrears, together with the rest of your usage: November usage is on the invoice issued on 1 December.
  • Unused included messages cover AI credits too. The included pool of the billing account (all workspaces together) is applied to the usage with the highest unit rate first, and one included message covers one AI credit. Conversation messages and campaign sends cost the same as or more than an AI credit, so they are covered first. Whatever is left of the pool then covers AI credits, before email campaign sends, which cost the same as or less than an AI credit. Only AI credits the pool does not cover are charged, at EUR 0.01 per credit.
  • Knowledge Base search is free and never uses the pool.

Free trial limits​

A free trial has its own AI credit budget: 50 AI credits, plus +100 when you publish your first Conversational Agent and +100 when you add your first knowledge (up to 250). Trial AI credits are never invoiced. Your first 20 successful AI builds (the AI agent builder, Generate a WhatsApp Flow with AI, Create a flow with AI) and your first 200 questions to SendSeven Cowork do not use trial AI credits. Paid plans have none of these limits.

When the budget is used up, AI endpoints refuse the request before any AI work starts, so nothing is charged:

Statusdetail.codeWhen
403trial_ai_credits_exhaustedThe trial AI credits are used up. Returned by AI endpoints such as the agent test console, POST /api/v1/knowledge-base/search and POST /api/v1/attachments/{attachment_id}/summary.
503trial_ai_credits_unavailableThe trial budget could not be checked right now. Retry shortly.
402trial_ai_generation_limit_reachedAn AI build was refused: the free builds are used up and no trial AI credits are left. Returned by the AI agent builder (POST /api/v1/automation/agent-builder/generate), the WhatsApp Flows AI builder (POST /api/v1/whatsapp-flows/ai/generate and /refine) and the AI Flow Builder confirm step.
503ai_unavailableThe build allowance could not be checked. Retry after the Retry-After seconds (30).
{
"detail": {
"code": "trial_ai_credits_exhausted",
"message": "You have used the AI credits included in your free trial. Pick a plan to keep using AI features.",
"limit": 250,
"used": 250
}
}
{
"detail": {
"code": "trial_ai_generation_limit_reached",
"error": "trial_ai_generation_limit_reached",
"message": "You have used the free AI builds included in your trial, and your trial AI credits do not cover another build. Upgrade to keep building with AI.",
"kind": "agent",
"limit": 20,
"used": 20,
"credits_required": 1,
"credits_remaining": 0,
"credits_limit": 250,
"free_generations_limit": 20,
"free_generations_used": 20,
"upgrade_url": "/settings/billing"
}
}
  • kind is agent (AI agent builder and AI Flow Builder) or whatsapp_flow.
  • limit and used count builds of that kind: used is the number of successful builds, limit is used plus the builds that still fit.
  • credits_required is what one build reserves before it starts: 1. A build is charged by its usage, so it only needs some trial AI credits left. credits_remaining is what is left of the trial budget minus builds that are still running. A failed or cancelled build costs nothing and frees its reservation.
  • Pick a plan to lift every trial limit.

Details​

  • Output costs more than input. Writing is more work for the AI than reading.
  • Cached input is cheaper. Text the AI read a moment ago, such as the same instructions and conversation on the next reply, costs much less the second time.
  • Everything the AI reads counts: the customer's message, the conversation so far, the agent's instructions and skills, and the knowledge base articles it found. That is why a one-sentence answer can still use a noticeable share of a credit.
  • Advanced uses about 3.3x as many credits as Fast for the same request. Actual usage depends on context, instructions, skills and conversation length.

Tips to use fewer AI credits​

Everything the AI reads on a reply counts, so short and precise instructions save credits on every single reply.

  • One short foundation skill, everything else on demand. Make one skill always_on and keep only the rules that apply to every conversation in it: who you are, tone, what the agent must never do. Make all other skills model_selected. They are loaded only when a conversation needs them, so a simple question does not pay for your return policy or your lead qualification steps. See Skills.
  • Write precise skill descriptions. The agent decides which skills to load from their descriptions. Say when the skill applies, for example "The customer wants to return or exchange an order". A vague description loads the wrong skill or too many. The foundation skill does not need to explain when to use the other skills, because their descriptions already do that.
  • Keep the persona short. Two or three sentences are enough: the agent's name, its tone and how it addresses customers. A long backstory is read on every reply and rarely makes answers better.
  • Write rules, not essays. Use short bullet points with one rule each. Say each rule once, in one place. Do not repeat it in the persona, the instructions and a skill.
  • Put facts in the Knowledge Base, not in skills. Price lists, product details, opening hours and policies belong in the Knowledge Base. Searching it is free and the agent reads only the articles that match. Text pasted into a skill is read every time the skill is loaded, whether it is needed or not.
  • Use only the tools you need. Each web search or live page read costs 3 credits. If the Knowledge Base usually has the answer, set web search to kb_fallback, so the web is searched only when the Knowledge Base finds nothing. Leave image and PDF reading off unless customers send files the agent has to understand.
  • Start with Fast. Switch an agent to Advanced only when its answers on Fast are not good enough.
  • Check in the test console. It shows which skills were loaded and what each turn cost. If a simple question loads several skills, sharpen their descriptions. See Test console.

FAQ​

Am I charged for failed AI requests? No. Failed or cancelled AI runs are not charged.

Is translation free? On Fast, yes: translations (messages, emails, flows, bots and WhatsApp templates) and language detection are free. With the translation level set to Advanced in Settings > AI, each of these translations is charged by usage.

Is searching the knowledge base free? Yes. When the AI writes an answer from the articles it found, that answer is charged.

Next steps​