What credits are
Credits are the single unit of usage on Echo AI. Instead of juggling separate meters for messages, voice minutes, crawled pages and image generations, everything your Echo does is converted into credits and deducted from one balance. Every deduction is logged with its action type, so you can always trace where a credit went.
You can see your balance and every line of usage in Settings > Credits, where the history lists each action (chat message, voice call, knowledge sync, website import, image generation, and so on) with the exact amount charged.
How charges work
Every action your Echo performs has a usage based credit cost. You are never billed a flat fee per message: a short reply costs a fraction of a long, context heavy one, and you only pay for what is actually used.
Text is billed per token, voice per minute or per character, knowledge per page, and media per generation.
Minimum charge: any billable action costs at least 0.01 credits, so very small operations never round down to zero.
Every deduction appears in your credit history with its action type and exact amount, so nothing is hidden.
If an action fails on our side, the credits are automatically refunded.
The figures below are the real amounts deducted from your balance, so you can plan your usage with confidence.
Where credits come from
Monthly plan allowance - Free includes 30 credits per month. Pro starts at 400 credits from $25 per month, Business starts at 400 credits from $50 per month, and both scale up to 10,000 credits per month. Enterprise is custom.
Credit packs - one-off top-ups you can buy at any time from the Subscription page.
Rollover - unused credits roll over on paid plans. Free plan credits reset each month.
Rewards - referral rewards, achievements and promotional free credits are added to the same balance and appear in your history as positive entries.
Refunds - when an action fails on our side, the credits are returned and logged as a refund.
Text conversations
Chat is billed per token, not per message, so short answers cost far less than long ones. A token is roughly four characters of English text. Both the input (your Echo instructions, knowledge, conversation history) and the output (the reply) are counted.
<table><tbody><tr><th>Model
</th><th>Input (per token)
</th><th>Output (per token)
</th></tr><tr><td>GPT-4o mini
</td><td>0.000024 credits
</td><td>0.000096 credits
</td></tr><tr><td>GPT-5 nano
</td><td>0.000008 credits
</td><td>0.000064 credits
</td></tr><tr><td>GPT-5 mini
</td><td>0.00004 credits
</td><td>0.00032 credits
</td></tr></tbody></table>Practical examples on GPT-4o mini:
A short exchange (about 1,000 input tokens, 150 output tokens): roughly 0.04 credits.
A rich answer using catalog and knowledge context (about 4,000 input tokens, 400 output tokens): roughly 0.13 credits.
A long troubleshooting conversation (about 10,000 input tokens, 1,000 output tokens): roughly 0.34 credits.
Tool calls: when your Echo performs an action (checking availability, creating a basket, taking a booking, opening a ticket) each tool call adds a flat 0.016 credits on top of the token cost.
Welcome message generation is billed with the same token rates and typically costs around 0.01 to 0.02 credits.
Messaging channels and social integrations
WhatsApp, Instagram, Messenger, Telegram, Discord, Slack, Reddit and X replies use exactly the same token rates as web chat, so a WhatsApp answer costs the same as the same answer on your Echo page. Platform message delivery itself is not marked up.
X (Twitter) is the exception, because the X API is a paid subscription with strict quotas. Each API operation is charged separately from the reply itself:
Read (fetching mentions, threads, profiles): 0.13 credits per request.
Write (posting a reply or a DM): 2.14 credits per request.
Voice
Speech to text (voice notes)
Transcription runs on Whisper at 0.96 credits per minute of audio. A 15 second voice note costs about 0.24 credits.
Text to speech (OpenAI voices)
2.4 credits per 1,000 characters. A typical 400 character reply read aloud costs about 0.96 credits.
Text to speech (ElevenLabs voices)
Multilingual v2 / v3 quality: 16 credits per 1,000 characters.
Flash / Turbo: 8 credits per 1,000 characters.
ElevenLabs voices are premium and cost noticeably more than OpenAI voices, which is why the voice you pick matters for your monthly consumption.
Voice design and cloning
Voice design previews: each request synthesises your sample text three times, billed at 0.016 credits per character generated (so a 200 character sample costs about 9.6 credits).
Saving a designed voice: 16 credits, covering the persistent voice slot.
Instant voice cloning: 80 credits per cloned voice.
Live voice conversations (OpenAI realtime)
Realtime voice is billed per audio and text token, which is why cost tracks how much is actually spoken rather than how long the line stays open.
<table><tbody><tr><th>Token type
</th><th>Input
</th><th>Output
</th></tr><tr><td>Audio
</td><td>0.00512 credits
</td><td>0.01024 credits
</td></tr><tr><td>Text
</td><td>0.00064 credits
</td><td>0.00256 credits
</td></tr></tbody></table>In practice, a conversational minute with normal back and forth speech lands in the region of 15 to 30 credits, and silence costs almost nothing.
Live voice conversations (ElevenLabs agents)
480 credits per minute, billed on session duration with a 0.05 credit minimum. This is the most expensive action on the platform, because ElevenLabs charges a premium per minute for conversational agents. Use it when voice quality is the product; use OpenAI realtime when volume matters.
Knowledge, crawling and imports
Website import (Echo creation and AI enhance)
Cost is proportional to what was actually fetched and understood: 1.6 credits per page scraped plus the token cost of the extraction, rounded to one decimal, with a 1 credit minimum. A single page import costs 1 credit; a deep crawl of a 20 page site typically lands between 30 and 40 credits.
Live knowledge sources (always on website sync)
0.4 credits per page fetched, including the initial sitemap/discovery request.
Syncs are skipped when nothing changed (ETag and content hash checks), so an unchanged site costs almost nothing beyond the discovery call.
A sync needs a balance of at least 5 credits. If your balance is too low, the source is paused and you get a notification instead of a silent failure.
Each source has a page cap and a cooldown, so a large site cannot drain your balance in one pass.
Document imports
Document import (building an Echo from a PDF, DOCX or text file) and document extraction (vision based reading of scans and images) are billed on the tokens used to read and structure the file. A short brochure is usually well under 1 credit; a long scanned PDF read page by page with vision can reach several credits.
Catalog import from a document is billed the same way, on the tokens needed to extract the products.
Media generation
Images (gpt-image-1-mini): 0.00032 credits per text input token and 0.00128 credits per image output token. A standard 1024x1024 medium quality image costs about 1.76 credits, which is also the flat fallback when token usage is not reported.
Video: 16 credits per generation.
AI Echo Builder
The builder is billed like chat: the same token rates for the conversation, plus 0.016 credits per tool call for each change it applies to your Echo. A simple tweak is usually around 0.03 credits, a multi step rebuild rarely exceeds 0.25 credits.
Conversation Intelligence
Sentiment, topic and resolution analysis runs on GPT-4o mini with the standard token rates and is logged as conversation analysis. Each analysed conversation costs a fraction of a credit, and analysis runs once per conversation rather than on every message.
Reading your credit history
In Settings > Credits, each entry shows the action type, the amount, and whether it was a deduction (minus) or a top up (plus). Action types map directly to what you did: chat message, voice call, voice note, voice transcription, voice TTS, voice design, voice clone, AI builder, website import, knowledge sync, document import, document extraction, catalog import, image generation, conversation analysis, welcome message, plus channel specific entries such as WhatsApp, Instagram, Telegram, Discord, Slack, Reddit and X. Positive entries include subscription renewal, credit top up, referral reward, achievement reward, free credits and refunds.
Running low on credits
You get in-app and email alerts as your balance gets low.
Live knowledge sources pause themselves rather than failing repeatedly.
Voice and media actions are blocked before they start if the balance cannot cover them, so you never end up with a partial charge.
You can top up at any time with a credit pack, or move to a higher credit tier on Pro or Business.
Frequently asked
Do my visitors spend my credits? Yes. Your Echo runs on your balance, whoever is talking to it.
Does a longer system prompt cost more? Yes, because instructions and knowledge are part of the input tokens on every message. Tight, well organised instructions are cheaper and usually produce better answers.
Do failed actions cost credits? No. Charges happen after a successful operation, and anything charged in error is refunded and logged.
Why is one chat message more expensive than another? Because of context. Catalog data, knowledge chunks, attachments and long conversation history all add input tokens.