> ## Documentation Index
> Fetch the complete documentation index at: https://docs.lehar.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Billing & Credits

> How usage becomes credits, and how to read your balance and ledger.

Billing is **usage-based** and denominated in **credits**. A voice call is charged as a flat **per-minute** line; WhatsApp is charged **per message**. You read your balance and history from two customer endpoints; the internal cost model is not exposed.

<Warning>
  **A negative credit balance blocks new work.** Credit enforcement is on by default: starting a new session, campaign dial, single call, inbound answer, or agent-test run while your balance is **below zero** is rejected with `402 insufficient_credits`. A balance of exactly **0 is allowed** — the call that takes you negative still runs — and a call already in progress is **never** interrupted: live calls and open conversations always finish. Overdraft allowances and per-workspace exemptions are set by your administrator.
</Warning>

## How a voice call is priced

Every voice call — outbound phone, inbound phone, or web — is charged as a single flat line, **`voice_billing_minutes`**, priced per **call type** (`outbound_phone`, `inbound_phone`, `web`). This is the only line that carries credits for a call.

* **Pulse rounding.** The billed duration is rounded **up to the next whole pulse**. The default pulse is **60 seconds** (per-minute billing); your administrator can lower it — down to **1 second** (per-second billing).
* **Free-below threshold.** A call shorter than a **minimum billable duration** is free (billed at zero quantity). The default is **0 seconds**, so every answered call is charged; your administrator can raise it.

The underlying components (session, telephony, model tokens, speech-to-text, text-to-speech, recording) still appear on your usage **as line items with quantities**, but their credit charge is **0** — the flat `voice_billing_minutes` line replaces them. The detail stays visible without double-charging.

## What gets metered

A finalized **voice call** produces usage lines across these components. The flat `voice_billing_minutes` line carries the charge; every other line is recorded at **zero credits** (the **Credited** column). [WhatsApp](#whatsapp) is billed differently.

| Component                                                            | Unit      | Credited                           |
| -------------------------------------------------------------------- | --------- | ---------------------------------- |
| `voice_billing_minutes`                                              | minute    | **Yes** — the flat per-call charge |
| `agent_session_minutes`                                              | minute    | No — usage only                    |
| `webrtc_minutes` / `telephony_minutes` / `telephony_carrier_minutes` | minute    | No — usage only                    |
| `llm_input_tokens` / `llm_output_tokens` / `llm_cached_tokens`       | token     | No — usage only                    |
| `stt_minutes`                                                        | minute    | No — usage only                    |
| `tts_characters`                                                     | character | No — usage only                    |
| `recording_minutes`                                                  | minute    | No — usage only                    |
| `observability_events`                                               | event     | No — usage only                    |

Token counts are **reported** by the model; minutes and characters are **estimated** from timestamps and transcript length. A phone call (inbound or outbound) meters the same duration on two telephony legs — a platform leg and a carrier leg — except when the call used your own SIP trunk (BYON), where the carrier leg is skipped. Realtime audio-to-audio models skip the separate speech-to-text / text-to-speech lines.

## WhatsApp

WhatsApp is metered **per message**, not per session. Each outbound message emits one **`whatsapp_messages`** event, priced by conversation **category** — `marketing`, `utility`, `authentication`, or `service`. WhatsApp has no flat per-call line, so the zero-credit rule above doesn't apply: both the `whatsapp_messages` events **and** the reply LLM's `llm_input_tokens` / `llm_output_tokens` / `llm_cached_tokens` lines carry credits here. Metering is idempotent per message, so a retried send is never double-charged.

## Credits & ledger

Your **credit balance** is the running sum of your ledger. Entries are signed:

| Entry type         | Sign | Source                                  |
| ------------------ | ---- | --------------------------------------- |
| `grant` / `refund` | +    | Added by an administrator               |
| `adjustment`       | ±    | Manual correction — may add or deduct   |
| `debit`            | −    | Charged for a call, session, or message |
| `reversal`         | +    | Reverses a debit when usage is re-rated |

Charging is **idempotent** — a call's usage is charged once. If usage is later corrected, the prior debit is reversed and re-charged, never added on top.

## Read your usage

| Method | Path                          | Scope                           |
| ------ | ----------------------------- | ------------------------------- |
| GET    | `/billing/me/summary?from&to` | `billing:read` (customer admin) |
| GET    | `/billing/me/ledger?from&to`  | `billing:read` (customer admin) |

`me/summary` returns a `summary` object — `customer_name`, `currency` (INR), `credits_granted`, `credits_used`, `credits_balance`, `total_tokens`, `total_minutes`, and `unpriced_usage_events` — alongside a `breakdown` array carrying each component's `quantity` and `credits`. The breakdown is where the flat `voice_billing_minutes` charge and the zero-credit detail lines appear. `me/ledger` returns your signed entries with a running `balance_after`.

```bash theme={null}
curl "$LEHAR_BASE_URL/billing/me/summary" -H "Authorization: Bearer $LEHAR_LOGIN_TOKEN"
```

<Note>
  These endpoints require a **customer-admin bearer session** and expose credits only — never internal cost. Rate cards and cross-customer accounts are administrator-only and aren't part of the public API.
</Note>