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

# How Callaro Calls Work

> Follow the full lifecycle from dial to script, action execution, and post-call logging.

# Every Callaro call moves through a predictable lifecycle.

<Steps>
  <Step title="Dial or resolve inbound">Callaro either places a call from an approved number pool or receives an inbound call that is mapped to the correct tenant, number, and agent context.</Step>
  <Step title="Load agent configuration">The runtime loads the active agent, voice, script rules, knowledge sources, and action permissions.</Step>
  <Step title="Run adaptive conversation logic">The model follows the script, branches based on caller responses, and retrieves knowledge when the response requires grounding.</Step>
  <Step title="Execute actions safely">When the conversation meets the right conditions, Callaro can update CRM fields, log notes, or create a calendar event.</Step>
  <Step title="Persist outcomes">On call completion, the platform writes the transcript, recording metadata, extracted fields, tool calls, timing, and billing context.</Step>
</Steps>

## Lifecycle observability map

| Lifecycle stage         | Primary signal                             | Typical failure to watch                |
| ----------------------- | ------------------------------------------ | --------------------------------------- |
| Dial or inbound resolve | Provider acceptance, call start timestamps | Number health/routing mismatch          |
| Agent load              | Config + permissions load success          | Missing/invalid agent dependencies      |
| Conversation run        | Transcript turn progression                | Prompt drift or tool-call errors        |
| Action execution        | CRM/calendar action result                 | Third-party auth or rate-limit failures |
| Outcome persistence     | Session + export availability              | Partial-write or delayed indexing       |

## Operational recommendations

* Alert separately on dial failures, action failures, and persistence failures.
* Use webhook + API reconciliation for high-confidence downstream processing.
* Keep retries idempotent by request/session identifiers.

<Note>
  This lifecycle is designed so teams can choose between push and pull models. You can subscribe to webhooks, query call logs directly, or use both together.
</Note>