# Test mode

**Availability: Test mode**

Test mode lets you exercise the whole API without dialling a real phone or spending balance. Calls progress through realistic states on a compressed timeline and produce simulated transcripts, recordings and analysis, and webhooks fire just as they do in live mode. Every request made with a `rai_test_` key is in test mode.

## What test mode does and does not do

Test mode **never** contacts a phone network, never places a real call, never charges your workspace and never reads or changes your live data.

What is the same as live mode:

- The same operations, request validation, error codes, scopes, idempotency and rate limits.
- The same compliance checks, against your **test-mode** contacts and do-not-call list: a number on the test do-not-call list is refused with [`dnc_listed`](/errors/dnc_listed), a number that matches a contact without consent (when the workspace requires it) with [`consent_required`](/errors/consent_required) (a number with no contact record is not checked for consent), and calling hours and the per-number daily limit apply. As in live mode, campaigns are screened when they start and resume, not again while they run.
- Webhook events, sent to test-mode endpoints with `"livemode": false`.

What does not apply in test mode:

- Balance, monthly spend limit and account status checks.
- The daily limit on calls placed through the API.
- Charges: `cost.amount` is always `0`.

## Test phone numbers

The destination number chooses the simulated result. Use these fictional numbers:

| Destination | Simulated result | `outcome` |
| --- | --- | --- |
| `+12025550100` | Answered and completed, with transcript, recording and analysis | `completed` |
| `+12025550101` | No answer | `no_answer` |
| `+12025550102` | Busy | `busy` |
| `+12025550103` | Failed: the call could not be connected | `failed` |
| `+12025550104` | Answered by voicemail; nobody spoke | `no_response` |
| `+12025550105` | Completed after a longer conversation | `completed` |
| `+12025550106` | The person hangs up early; short duration | `completed` |
| Any other valid number | Completed | `completed` |

Calls that are not answered (`+12025550101` to `+12025550103`) have no transcript, recording or analysis: their `artifacts` report `absent`. The voicemail case (`+12025550104`) has a recording but no transcript or analysis. Prefer the numbers above to any real number, even though test mode never dials it.

## Sample agent and number

Every workspace has a **Sample agent** in test mode, id `agent_1ZltiWzxrXmMprGD6q63`, marked `"sample": true`, always switched on, published and ready. It calls from the sample number `+12025550150` (id `pn_6Vhm3sFrB5UpmXAnsvmDu`, also `"sample": true`). Both appear in [`GET /agents`](/api/agents#list-agents) and [`GET /phone-numbers`](/api/phone-numbers#list-phone-numbers) for test keys only; with a live key they answer [`resource_not_found`](/errors/resource_not_found).

You can also test with your own agents. In test mode an agent must be switched on and saved for calling in the console; it does not need to be published or to have a phone number.

## Timing

A test call reaches `status: ended` within about 20 seconds and `settled: true` within about 30 seconds of being accepted. The progress is computed from timestamps, so reading the call later gives the same result. `duration_seconds` is simulated too and does not match the wall-clock time the test call took (for example 42 seconds for `+12025550100`).

Test-mode campaigns use test-mode contacts only. When you start one, each recipient gets a simulated call whose result is chosen by the recipient's number (the table above), and campaign events fire with `"livemode": false`. Pause and stop take effect immediately.

## Simulated results

- **Transcripts** and **analysis** are fixtures marked `"simulated": true`.
- **Recordings** are short synthetic audio files (`audio/wav`, about one second), served through the same short-lived link as live recordings, also with `"simulated": true`.
- Every simulated object has `"livemode": false`.

Do not use simulated content to test your own language processing: it is fixed sample content, not a real conversation.

## Test data

Calls, campaigns, contacts, do-not-call entries and webhook endpoints that you create with a test key stay in test mode. They are invisible to live keys and to the console's normal screens. Use them freely: nothing in test mode affects real people or your balance.

## Moving to live mode

Switch the key, not the code: deploy a `rai_live_` key, check `GET /v1/workspace` says `"livemode": true`, and make sure your live agents are published and ready. See [Going live](/guides/going-live).