Data processing with a disposable AI agent
An agent profiles a CSV, finds duplicates and bad values, or groups survey answers into themes with quotes you can check. Ephemeral puts your files into a fresh microVM as read-only inputs, lets the agent run real commands there, keeps the network off unless you allow a domain, reserves the budget before the run and destroys the machine when it finishes.
What the run is allowed
Everything a run can touch is listed in its request. This is the request of the sample Clean and profile a messy CSV with an AI agent.
| Files | orders.csv (read-only) |
|---|---|
| Connections | None |
| Web access | Off |
| Shell | On, in the sealed microVM |
| Budget | $0.05 per run, reserved before it starts |
| Timeout | 120 seconds |
How to run it
Start a run with the API or the CLI. The result arrives by signed callback or when you read the run.
POST /v1/runs
{
"workflow": "csv-quality-report",
"input": {},
"max_cost_micros": 50000
}Samples
Each sample is a complete request you can read, replay and copy.
- Clean and profile a messy CSV with an AI agentAn agent profiles a CSV with real commands in a sandbox and reports duplicates, missing values, bad formats and outliers.
- Group customer feedback into themes with quotes, using AIAn agent reads survey responses and returns the main themes with counts and word-for-word quotes.
- Extract invoice data to JSON with an LLM, with zero data retentionAn agent turns invoice text into structured JSON and checks that the totals add up. The run keeps nothing afterwards.
Related terms
- MicroVM: The small virtual machine each run gets fresh and that is destroyed when the run ends.
- Zero data retention: A run mode in which providers keep nothing and the run's content is not stored.
- Run budget: The most a run may spend, reserved up front and enforced while it runs.
Questions
Does the data leave the sandbox?
The CSV sample has no web access, so there is nothing to send it to. The files are read-only inputs and the machine is destroyed when the run ends.
Does the agent count by eye?
No. The CSV sample asks it to count problems with the shell, with tools like Python and awk, and to list row numbers for each issue.
Can I check where a theme or count came from?
Yes. The feedback sample returns a count and a word-for-word quote for each theme, so each claim can be checked against the source file.
Prices are on the pricing page.