Receipts

Find the receipt of any agent run, read where the time went and what it cost, export receipts, and quote one to support.

For owners and adminsRun your workspace · page 13 of 35

Every turn an agent takes keeps a receipt: the proof of a run, in numbers. A receipt says when the run happened, how it ended, how long it took and whose time it was, which model served it, an estimated cost and the work done. Nothing on it reveals the contents of your work. This page shows you how to find, read and export receipts, and how to use one with support.

Before you start

Open Governance › Operations, then the Receipts tab.
Receipts answer "whose time was this, and what did it cost". To find who did what, use the Audit trail.

Find a run

Open the Receipts tab.
The list, Recent runs, shows the last 48 hours, newest first: one line per turn.
Read the columns: Run, When, Duration, Whose time, Est. cost and Outcome.
Whose time is a bar split between the model, tools outside Nexa and the Nexa platform. Outcome reads Completed, With a warning or Did not finish.

The receipts list of Harbor & Pine with the latest runs of Pixel, Silvestro, Lumen and Luna, one With a warning, and the Whose time bars highlighted. — 01One line per turn: what ran, how long it took and whose time it was.

To find a run, type in Search by reference or agent: the agent's name, or the first characters of a reference.
To go further back in the 48 hours, click Load older runs.

Read a receipt

Click a run.
The receipt opens. All receipts leads back to the list.
Read the header: the agent, the situation, such as During a conversation, During project work or During a scheduled run, the time and the outcome.
Read Where the time went: the split between the model, External tools such as web search, and Nexa platform, everything Nexa did around the model. A one-line verdict sums it up, for example The model answered at N tokens per second and held N% of this run.

A receipt of a run of Pixel: Completed, Where the time went, and The numbers with Nexa-hosted, the estimated cost and the tokens. — 02A receipt shows where the time went and what the run cost.

Read The numbers:
Inference source
Nexa-hosted, Your key, Your bridge or Mixed sources.
Estimated cost
from published prices, not a bill. A model without a price on file is named and left out of the figure.
Signals from your provider
retries, timeouts, model errors and tool refusals, or No retries, timeouts or refusals.
Tokens processed
in and out, and the share served from cache.
Work done
rounds and tool calls.
Context read by the model
what the model was given to read on this turn, in characters.

On your own key or bridge, the receipt reads Your model, with Your key or Your bridge.

A receipt of a run on the workspace's own bridge, reading Your model and Your bridge. — 03A run on your own key or bridge says so.

Open a receipt from elsewhere

In the Audit trail, open the details of an entry that belongs to a run: under Receipt, click Open the receipt.
In the Company Brain, a memory written during a turn can open that turn's receipt.

Pixel's audit trail entry opened, with Open the receipt highlighted under Receipt. — 04Audit trail entries lead to the receipt of their run.

Export receipts

One receipt:

On the receipt, click Export receipt.
Keep or change the document title, and choose PDF, DOCX, XLSX, HTML, MD or TXT.
Save. The receipt is saved as a document in your private Library space, ready to download, and you can share it afterwards.

The receipt footer with Copy support reference and Export receipt. — 05Export a receipt, or copy its reference for support.

A whole window of runs:

On the list, click Export the last 48h.
Choose CSV or JSON.
The file holds up to 5,000 runs and says if it was cut short.

Export the last 48h with the CSV and JSON choices. — 06Export every run of the last 48 hours in one file.

Ask support about a run

Each receipt carries a reference, REF. Quote it, and support opens exactly that run.

On the receipt, click Copy support reference.
It reads Reference copied.
Open Settings › Support and start a new ticket.
Paste the reference, with the time and what you were doing.

What you will see

No runs yet. — Every turn an agent takes writes a receipt here.

No run matches that.

That is every run of the last 48 hours. — the end of the list.

Nothing measurable was spent on this run.

Waiting for … on your bridge, … on your own key or …, hosted by Nexa — the run waited on a model.

Not measured — runs from before the measure carry no figure.

This run kept no receipt.

Receipt unavailable — quote the reference in a support request, and support can open exactly this run.

The export did not go through.

Looking for who approved what? under a receipt, with See what … did in the audit trail →.

Good to know

Whose time tells you whether a slow answer came from the model, from an outside tool such as web search, or from the platform.
A receipt shows plain numbers only. It never shows prompts, tool names, the content of documents or the platform's internal steps.
Receipts do not quote a comparison figure until there are enough measured runs of the same kind to quote one honestly.
Times are measured on every turn.

Receipts

Find the receipt of any agent run, read where the time went and what it cost, export receipts, and quote one to support.

For owners and adminsRun your workspace · page 13 of 35

Before you start

Open Governance › Operations, then the Receipts tab.
Receipts answer "whose time was this, and what did it cost". To find who did what, use the Audit trail.

Find a run

Open the Receipts tab.
The list, Recent runs, shows the last 48 hours, newest first: one line per turn.
Read the columns: Run, When, Duration, Whose time, Est. cost and Outcome.
Whose time is a bar split between the model, tools outside Nexa and the Nexa platform. Outcome reads Completed, With a warning or Did not finish.

To find a run, type in Search by reference or agent: the agent's name, or the first characters of a reference.
To go further back in the 48 hours, click Load older runs.

Read a receipt

Click a run.
The receipt opens. All receipts leads back to the list.
Read the header: the agent, the situation, such as During a conversation, During project work or During a scheduled run, the time and the outcome.
Read Where the time went: the split between the model, External tools such as web search, and Nexa platform, everything Nexa did around the model. A one-line verdict sums it up, for example The model answered at N tokens per second and held N% of this run.

Read The numbers:
Inference source
Nexa-hosted, Your key, Your bridge or Mixed sources.
Estimated cost
from published prices, not a bill. A model without a price on file is named and left out of the figure.
Signals from your provider
retries, timeouts, model errors and tool refusals, or No retries, timeouts or refusals.
Tokens processed
in and out, and the share served from cache.
Work done
rounds and tool calls.
Context read by the model
what the model was given to read on this turn, in characters.

On your own key or bridge, the receipt reads Your model, with Your key or Your bridge.

Open a receipt from elsewhere

In the Audit trail, open the details of an entry that belongs to a run: under Receipt, click Open the receipt.
In the Company Brain, a memory written during a turn can open that turn's receipt.

Export receipts

One receipt:

On the receipt, click Export receipt.
Keep or change the document title, and choose PDF, DOCX, XLSX, HTML, MD or TXT.
Save. The receipt is saved as a document in your private Library space, ready to download, and you can share it afterwards.

A whole window of runs:

On the list, click Export the last 48h.
Choose CSV or JSON.
The file holds up to 5,000 runs and says if it was cut short.

Ask support about a run

Each receipt carries a reference, REF. Quote it, and support opens exactly that run.

On the receipt, click Copy support reference.
It reads Reference copied.
Open Settings › Support and start a new ticket.
Paste the reference, with the time and what you were doing.

What you will see

No runs yet. — Every turn an agent takes writes a receipt here.

No run matches that.

That is every run of the last 48 hours. — the end of the list.

Nothing measurable was spent on this run.

Waiting for … on your bridge, … on your own key or …, hosted by Nexa — the run waited on a model.

Not measured — runs from before the measure carry no figure.

This run kept no receipt.

Receipt unavailable — quote the reference in a support request, and support can open exactly this run.

The export did not go through.

Looking for who approved what? under a receipt, with See what … did in the audit trail →.

Good to know

Whose time tells you whether a slow answer came from the model, from an outside tool such as web search, or from the platform.
A receipt shows plain numbers only. It never shows prompts, tool names, the content of documents or the platform's internal steps.
Receipts do not quote a comparison figure until there are enough measured runs of the same kind to quote one honestly.
Times are measured on every turn.