> This page is for Voice Agents.

> Sarvam Voice Agents (Samvaad) is a voice AI agent platform for Indian languages: build, test and deploy AI calling agents on your own phone numbers, from ₹3 a minute. Start here: https://docs.sarvam.ai/voice-agents/why-voice-agents.
> For clean Markdown of any page, append `.md` to the page URL.
> For a complete documentation index, see https://docs.sarvam.ai/llms.txt. For a section index, append `/llms.txt` to any section URL (for example https://docs.sarvam.ai/voice-agents/llms.txt).
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.sarvam.ai/_mcp/server.

# Connect: export voice agent call data to SFTP and Amazon S3

> Use Connect to send voice agent call reports and recordings to your own SFTP server or Amazon S3 bucket, on a daily, weekly or monthly schedule or once for a past date range.

Connect pushes your voice agents' call data to storage you own on a schedule. You add a **connection** once (an SFTP server or an Amazon S3 bucket), then create **export jobs** that send data to it. You don't need to poll Sarvam for files.

Open **Monitor → Connect** in the sidebar. **How it works** gives a short in-product walkthrough.

> **Note**
>
> Supported today: **SFTP** and **Amazon S3** (including S3-compatible stores such as MinIO). Azure Blob and Google Cloud Storage are coming soon.

## Before you start

#### SFTP

* A dedicated SFTP user that can create folders and write files under the export folder.
* The server is reachable from Sarvam. If it's firewalled, ask your Sarvam account team for the egress IPs to allowlist.
* Enough disk space for your largest export. Recordings are much larger than reports.

#### Amazon S3

* A bucket for the exports.
* An AWS access key ID and secret access key with write access to that bucket. IAM roles and instance profiles aren't supported.

## Add a connection

#### Open the connection form

Click **New connection**. If you already have connections, click **Manage connections** and add one from there.

#### Enter the details

Pick **SFTP** or **Amazon S3** under **Type**, add a **Display name**, and fill in the fields below.

#### Test and continue

Sarvam checks it can reach and sign in to your storage before saving. If the test fails, you see the reason and nothing is saved. Check the host, port, credentials, firewall rules and folder permissions, then try again.

#### Pick a folder (optional)

Browse to the folder jobs should write into and click **Use this folder**, or click **Skip** to keep the path you entered. On SFTP you can create a **New folder** here.

#### SFTP fields

| Field                       | Notes                                          |
| --------------------------- | ---------------------------------------------- |
| **Host**                    | Server name only, without `sftp://` or a path. |
| **Port**                    | Defaults to `22`.                              |
| **Username** / **Password** | The dedicated export user.                     |
| **Base path**               | Top folder jobs write under. Defaults to `/`.  |

#### Amazon S3 fields

| Field                                     | Notes                                                                                     |
| ----------------------------------------- | ----------------------------------------------------------------------------------------- |
| **Bucket**                                | Bucket files are written to.                                                              |
| **Access key ID** / **Secret access key** | Static AWS keys with write access.                                                        |
| **Region**                                | Optional. Detected during the test.                                                       |
| **Endpoint URL**                          | Optional. Only for S3-compatible stores such as MinIO.                                    |
| **Session token**                         | Optional. Only for temporary credentials.                                                 |
| **Prefix**                                | Optional. Files land under this key prefix, which appears once the first file is written. |

Credentials are encrypted at rest and never shown again after saving.

## Create an export job

Click **New job**.

#### Destination

Choose a **Connection** (only tested connections are listed, or pick **Add a connection**) and a **Frequency**: **Daily**, **Weekly**, **Monthly** or **One time**. All times are IST.

#### Data

Choose **All agents** or specific agents, then tick what to **Include**. **All agents** is resolved at each run, so new agents are picked up automatically.

#### Review

Name the job, check the summary, and click **Create job**. A one-time job shows **Start export** instead and runs immediately.

### What you can export

| Data               | Contents                                                                                                          | Format           |
| ------------------ | ----------------------------------------------------------------------------------------------------------------- | ---------------- |
| **Unified report** | One row per call attempt, connected or not, with agent variables and transcripts (`transcript`, `transcript_en`). | CSV in a `.tar`  |
| **Interactions**   | Connected calls only, with the same fields.                                                                       | CSV in a `.tar`  |
| **Recordings**     | One WAV per call. Skipped for a window with no audio.                                                             | WAVs in a `.tar` |

Debug calls are excluded. Archives are plain tar, not gzip.

### What each run covers

| Frequency    | Window                                                                                      | Example                                                       |
| ------------ | ------------------------------------------------------------------------------------------- | ------------------------------------------------------------- |
| **Daily**    | The 24 hours ending at the scheduled time.                                                  | Daily at 02:00 on Nov 5 covers Nov 4 02:00 → Nov 5 02:00 IST. |
| **Weekly**   | The 7 days ending at the scheduled time.                                                    | Monday 02:00 covers the previous Monday 02:00 onward.         |
| **Monthly**  | The month ending at the scheduled time (day 1–28).                                          | 1st at 02:00 on Dec 1 covers Nov 1 02:00 → Dec 1 02:00 IST.   |
| **One time** | A past date range, start and end days included, up to 92 days. Must end before today (IST). | Oct 1 → Oct 7 covers Oct 1 00:00 → Oct 7 23:59:59 IST.        |

* Jobs start within 15 minutes of the scheduled time.
* Missed windows aren't backfilled automatically. Use a one-time job.
* A job's schedule can't be edited. Create a new job and pause the old one.

## Delivered files

Each run writes a dated folder per agent:

```txt
{base path}/{job name}/{organization}/{workspace}/{agent}/{date}/
  {agent}_control-{date}.json
  {agent}_unified_report-{date}.tar
  {agent}_interactions-{date}.tar
  {agent}_recordings-{date}.tar
```

`{date}` is the IST date the window ends. Files are uploaded atomically, so you never see a partly written file.

Each folder includes a **control file**. Only process an agent's files when `status` is `"success"`. A data type missing from the control file wasn't produced for that window.

```json
{
  "date": "2026-11-05",
  "status": "success",
  "total_bytes": 123456,
  "unified_report": { "records": 1000, "bytes": 120000 },
  "interactions": { "records": 800, "bytes": 3000 },
  "recordings": { "records": 50, "bytes": 456 }
}
```

Sarvam never deletes files from your storage.

## Monitor and manage

* **Jobs table:** filter by **All jobs**, **Active** or **Paused**. The cards above it show active jobs, runs today and failed runs in the last 7 days.
* **Job details:** click a job for its next run, last run and run history, with each run's date range, size and status. Open a failed run to see the error.
* **Pause / Resume:** resuming schedules the next run from now, so a long-paused job doesn't fire straight away. One-time jobs can't be paused.
* **Manage connections:** **Test**, **Edit** or **Delete** a connection. Untested connections show **Unverified** and can't be used in jobs. To change credentials, re-enter the full set. Deleting a connection stops every job that uses it.

## Troubleshooting

| Problem                                 | What to check                                                                              |
| --------------------------------------- | ------------------------------------------------------------------------------------------ |
| **Connect** is missing from the sidebar | Ask your workspace admin for access.                                                       |
| Connection test fails                   | Host, port, credentials, firewall allowlist, and write permission on the folder.           |
| Job didn't run at the exact time        | Jobs start within 15 minutes of the scheduled time. Check **Next run** in the job details. |
| No recordings file                      | No call audio exists for that window. The run still succeeds.                              |
| Files look incomplete                   | Wait for the run to pass, then check the control file's `status`.                          |