Usage Rates
- How AutoTalk's pure-usage pricing works
- Every resource that's metered and what it measures
- How spending controls and the low-balance block protect you
AutoTalk has no plans or tiers — you pay only for what you use, in credits (100 credits = US$1, so 1 credit = 1¢), funded from your prepaid wallet. Purchased credits never expire.
An eligible new company can receive US$3 in signup credits (300 credits), with a signup-access period of 30 days from the grant. Anti-abuse eligibility can reduce the grant, including to zero. Check the access cutoff displayed in the wallet; paid/admin conversion can remove the signup-only restriction.
The exact per-unit rates are shown in the app; this page explains what each meter measures.
What's metered
Usage meters
| Meter | What it measures |
|---|---|
Compute time (compute_ms) | Server processing time consumed by your requests, in milliseconds |
Normalized input tokens (tokens_in) | Model-weighted input tokens sent to AI agents — see Normalized Tokens |
Normalized output tokens (tokens_out) | Model-weighted output tokens generated by AI agents — see Normalized Tokens |
Speech-to-text (stt_ms) | Audio transcribed, in milliseconds, normalized by a per-model weight — so the metered value is not necessarily the raw duration (the same mechanism as tokens above). See Normalized Tokens |
Text-to-speech (tts_ms) | Speech synthesized, estimated from text length and normalized by a per-model weight. Only an actual eligible platform cache hit avoids a new synthesis charge; identical text within ten minutes alone does not guarantee a hit |
Transcoding (transcoding_ms) | Time spent transcoding media files, in milliseconds |
File storage (storage_byte_day) | Bytes of files you keep, billed per byte-day |
Database storage (database_byte_day) | Bytes your company's records occupy, billed per byte-day |
Speech that runs on your own provider key — bring-your-own-key transcription or bring-your-own-key voice — is recorded on an internal audio_ms_byok counter for your reference and is not billed by AutoTalk; your provider bills you directly.
The synthesis cache is bounded and process-local. Its key includes provider, model, voice, speed, effective instructions, format and text. Entries can be evicted before their ten-minute TTL. BYO synthesis bypasses this platform cache; other request processing can still be billed.
Hosted channel uptime
Some channels keep a live connection open that AutoTalk hosts for you, so they are billed per connected-minute while online:
| Meter | What it measures |
|---|---|
Hosted WhatsApp uptime (channel_minutes_lancer) | Minutes an internal (AutoTalk-hosted) WhatsApp-Web channel is connected |
Discord uptime (channel_minutes_discord) | Minutes a Discord bot connection is live |
IRC uptime (channel_minutes_irc) | Minutes a relay-hosted IRC connection is live (including Twitch chat connected via the IRC Twitch preset) |
Email uptime (channel_minutes_email) | Minutes a relay-hosted email mailbox connection (IMAP) is live |
Channels that don't hold a hosted connection — the official WhatsApp Business API, Telegram, Meta/Instagram, the native Twitch channel, web widgets, and a bring-your-own Evolution server — do not accrue uptime charges. They are billed only for the usage they generate (compute, tokens, and so on). Note that Twitch chat connected through the IRC Twitch preset is an IRC channel and is billed as channel_minutes_irc.
Safety caps
A few resources keep hard caps purely to prevent abuse. These are not billed tiers and rarely bite:
| Limit | What it caps |
|---|---|
| Messaging channels | Configured, nondeleted customer channels, including disabled/disconnected records. Disconnecting alone does not free a slot; deleting does. AutoTalk's internal channels (the in-app agent and agent test chats) don't count; they have a separate cap |
| WhatsApp QR-code instances | Maximum hosted WhatsApp-Web instances |
| API tokens | Maximum long-lived API tokens |
| Team members | Maximum admin and regular employees |
Spending controls
You are always in control of how much you spend:
- Low-balance block — new-spend checks can block requests when available credit is exhausted. Hosted connections follow a separate grace and pause/disconnect lifecycle and may continue accruing uptime until disconnected. A top-up removes the credit obstacle when eligible; other budget or connection requirements can remain.
- Monthly budget (optional) — set a monthly credits limit on the Spending controls page. Choose to either block usage or just warn when you reach it, and get alerts at thresholds (for example 50% / 80% / 100%). You can also set per-category sub-budgets (AI, channels, audio, storage).
Next steps
- Wallet & Top-ups — Add credits and see how they are applied
- Normalized Tokens — How
tokens_inandtokens_outare computed across AI models - Usage history — Review your consumption trends