Operating Flux

Usage and analytics

Flux measures four things: triggers, tokens, sessions and cost. The analytics dashboard breaks each one down by where it came from.

The four numbers

MetricCountsDriven by
TriggersTimes a workflow startedMessages received, API calls, interval firings, spreadsheet rows
TokensInput and output consumed by AI nodesPrompt length, memory depth, tool descriptions, answer length
SessionsDistinct runs, including multi-message conversationsHow many conversations, not how many messages
CostWhat the above adds up toProvider rates for each metric

The dashboard shows Today and This Month.

Triggers, tokens, cost and sessions, with distribution charts under each.

The five views

TabShows
OverviewTrigger distribution, cost distribution and token usage together
TriggersTrigger types, comparison between them, and where volume comes from
AI ModelsToken distribution, usage against cost, and the cost breakdown per model
MessagingUsage per channel — WhatsApp, Viber, Telegram, Facebook, Instagram
SpeechSpeech-to-text and text-to-speech consumption

Where usage comes from

Every provider is counted separately, with its own unit:

ProviderMeasured in
Flux UI, Flux API, Flux IntervalTriggers
WhatsApp, Viber, Telegram, Facebook, InstagramTriggers
Cognify inbound chatTriggers
GPT, GeminiTokens
Speech-to-textSeconds
Text-to-speechCharacters

What actually drives cost

It is rarely the number of nodes. In practice:

  • Conversation length. With memory attached, the whole history is sent on every message — so a twenty-message conversation costs far more than twenty separate one-message ones.
  • Tool count. Every tool's description is sent on every call. Six tools cost more per message than two.
  • Interval frequency. A one-minute interval is 1,440 triggers a day, plus whatever each run consumes downstream.
  • Prompt length. A long system prompt is paid for on every single message.
If cost is higher than expected, check the AI Models tab for usage against cost, then the Triggers tab for what is firing. An interval you set to test and forgot is the most common surprise.

Reducing it

  1. Lengthen intervals. Ask whether every fifteen minutes is really needed.
  2. Trim the system prompt. It is paid for on every message.
  3. Remove unused tools. Each one is described on every call.
  4. Filter earlier. A Decision node before the AI node stops you paying for messages that never needed a model.
  5. Compare models. Run the same prompt on GPT and Gemini and read the cost side by side.

Usage and billing

Usage on Flux's built-in keys draws down your project's credits wallet. Usage on your own credentials is billed by the provider directly and does not touch the wallet. See Billing and credits.