description: "How data flows into, through, and out of a conversation"
---
Dograh has a simple data model for passing information through a call. Understanding it is key to building agents that feel personalised and to extracting useful results after a call.
Data available to the agent before the call starts — the contact's name, account details, appointment information, anything the agent should know upfront. It can be set from several places:
- **[API trigger](/voice-agent/api-trigger)** — pass it in the request body when calling `POST /public/agent/{uuid}` or `POST /telephony/initiate-call`
- **[Campaign CSV](/core-concepts/campaigns)** — columns beyond `phone_number` automatically become `initial_context` fields for each contact's call
- **[Pre-call data fetch](/voice-agent/pre-call-data-fetch)** — enrich the context with data from your CRM or ERP via an HTTP call as the call starts, before the agent speaks
- **[Agent Settings](/voice-agent/template-variables#using-template-variables-for-testing)** — set template context variables on the agent for testing; they're included in test calls from the workflow editor and ignored on production calls
Values from `initial_context` are available in your agent's prompt using `{{double_brace}}` syntax.
```
You are calling {{customer_name}} about their {{plan}} plan,
which renews on {{renewal_date}}. Be friendly and confirm
whether they'd like to continue.
```
When the call starts, Dograh substitutes the values before sending the prompt to the LLM — so the agent speaks naturally as if it already knows the contact.
| `{{current_time}}` | Current time in UTC (or inferred timezone) | `2026-04-02 14:30:45 UTC` |
| `{{current_time_<TIMEZONE>}}` | Current time in the specified timezone | `2026-04-02 20:00:45 IST` |
| `{{current_weekday}}` | Current weekday name in UTC (or inferred timezone) | `Thursday` |
| `{{current_weekday_<TIMEZONE>}}` | Current weekday name in the specified timezone | `Thursday` |
Replace `<TIMEZONE>` with an [IANA timezone name](https://en.wikipedia.org/wiki/List_of_tz_database_time_zones) such as `Asia/Kolkata`, `America/New_York`, or `Europe/London`.
```
Today is {{current_weekday}} and the current time is {{current_time_America/New_York}}.
```
<Note>
When you use a timezone suffix on **either** `current_time` or `current_weekday`, the other variable without a suffix will automatically use the same timezone instead of UTC. For example, if your prompt contains both `{{current_time_Asia/Kolkata}}` and `{{current_weekday}}`, the weekday will also be resolved in `Asia/Kolkata`.
For telephony calls (inbound and outbound), Dograh automatically adds these variables to `initial_context`:
| Variable | Description | Example |
|---|---|---|
| `{{caller_number}}` | The phone number that initiated the call | `+14155550100` |
| `{{called_number}}` | The phone number that received the call | `+18005550199` |
For **inbound** calls, `caller_number` is the customer's number and `called_number` is your Dograh number. For **outbound** calls, it's the reverse — `caller_number` is your Dograh number and `called_number` is the customer's number.
```
You are speaking with the caller at {{caller_number}}.
Data the agent extracts *during* the call — the opposite direction of `initial_context`. Use it to turn a conversation into structured data: what the customer wants, whether they confirmed something, a value they gave you out loud.
#### How it gets populated
Turn on **extraction** on an [Agent](/voice-agent/agent) or [End Call](/voice-agent/end-call) node and define one or more variables to extract. Each variable has:
| Field | Description |
|---|---|
| `name` | The key it will appear under in `gathered_context` |
| `type` | `string`, `number`, or `boolean` |
| `prompt` | A natural-language description of what to look for, e.g. *"Did the customer confirm the appointment?"* |
When the conversation reaches that node, the LLM reads the transcript so far and fills in each variable based on its `prompt`. If a value can't be determined from the conversation, the variable is left empty rather than guessed — leave the `prompt` specific enough that the LLM knows exactly what counts as a match.
You can add extraction to more than one node. Each node's extracted variables are merged into the same `gathered_context` object as the call progresses, keyed by `name` — reuse a `name` at a later node if you want to overwrite an earlier value.
#### How to reference it downstream
`gathered_context` is **not available in Agent prompts** — a prompt can only reference `initial_context` fields, because extraction typically happens after the conversation that would use it. To act on extracted data, send it out via a node instead:
| Where | Syntax | Notes |
|---|---|---|
| [Webhook node](/voice-agent/webhook) payload | `{{gathered_context.field_name}}` | Prefixed, since the payload template can also reference `initial_context` |
| [Run record](/developer/webhooks#payload-context-variables) (API / dashboard) | `gathered_context` object | Returned after the run completes, alongside `recording_url` and `transcript_url` |