What do image and voice messages add to an OpenClaw agent bill?

A message is not a billing unit

An image or voice note can take a different route from plain text before an OpenClaw agent answers. A text-token projection will not necessarily include media preprocessing or the model's treatment of a native attachment. Nor should you add a transcription fee to every voice note by default. Find the route that actually ran, then compare usage against the statements for the accounts that served it.

Trace one attachment before pricing a month

OpenClaw's media-understanding guide says inbound images, audio, and video may be processed by a local tool or a configured provider before the reply. In the managed reply path, an image that was successfully described is represented as text instead of being attached again as a native image. An image that was not described can still be delivered to a vision-capable reply model. Provider support and the configured route matter more than the file extension.

Its API-usage guide lists media understanding and the core model response as separate possible sources of provider usage. The token-use guide also includes attachments and transcripts among things that can occupy context. That does not mean every attachment incurs two separate charges: local processing, text-only transcripts, native model input, and provider media operations have different billing paths. A later turn can reuse context, but what is billed depends on the serving provider and its usage record.

Make a route-and-bill worksheet

Choose one billing window and a few representative messages: one with an image, one with a voice note, and a comparable text-only message. Record the serving model and account for each reply without copying credentials. Inspect your OpenClaw media configuration and the relevant provider's usage detail. Keep an unknown row when you cannot establish the route.

Operation	Evidence to record	Billing check
Local preprocessing	Observed local tool path and resulting text	No assumed remote media fee; check for fallback
Provider media understanding	Serving provider, media type, usage unit, and request time	Match to that provider's statement for the same window
Reply model	Native attachment or text transcript, input/output and any cache usage	Match model usage to its account statement
Unresolved route	Message time and redacted request identifier	Leave cost unallocated until matched

For a monthly estimate, count provider-billed media operations by the provider's actual unit, then price them using the applicable account terms. Reconcile the model's billable input, output, and cache categories separately. If a provider already includes native image input in its model-usage line, do not invent an extra per-image fee on top. If a voice note was processed locally, do not price it as a remote transcription just because it was audio. Avoid turning file count or minutes recorded into billable units without that provider's rule and usage data.

Where the Clawback calculator stops

The live Clawback calculator lets you vary models, messages, heartbeats, child agents, token assumptions, and channels for a 30-day scenario. It has no media-processing route, attachment-size, or transcription-usage field and does not read your invoices. Use it for a text-token planning scenario, not a claimed total for image and voice traffic. Add only the separately verified media charges and any model units missing from that scenario. The resulting worksheet is a budget hypothesis until it reconciles with the actual provider records.