# Assistants & Models

An Assistant stores the instructions and AI configuration used to handle conversations. Create it once, then connect callers through a [Channel](/docs/channels-and-calls). Each call has its own conversation record.

## Instructions and voice

Instructions define the Assistant's role, the information it should collect, and how it should respond. For example, a booking Assistant can ask for a preferred date and use a Tool to check availability.

The Assistant also has a language, timezone, and voice. An optional `firstMessage` supplies its opening greeting. These settings configure the Assistant; they are separate from the transcript and extracted data recorded for each conversation.

## Choose a model architecture

Reecall supports three combinations of speech and language models:

| Architecture | How it processes speech | Model IDs |
| --- | --- | --- |
| Speech-to-speech, or STS | One model receives audio and generates response audio | `stsId` |
| STT + LLM + TTS | Speech-to-text transcribes the caller, a language model generates a text response, and text-to-speech produces audio | `sttId`, `llmId`, `ttsId` |
| STS with external TTS | An STS model handles the conversation, and a separate TTS model produces the voice | `stsId`, `ttsId` |

All three require a `voiceId`. For native STS, choose a voice from the STS model's `voices` array. When using TTS, choose a voice from the TTS model's `voices` array.

STS requires fewer model choices and is the path used in the [Quickstart](/docs/quickstart). Separate STT, LLM, and TTS models let you choose a provider for each stage. Model capabilities, latency, and cost vary by provider and configuration.

The model catalogs are shared resources. Select existing models through the API rather than creating model records for each Assistant. The [Assistant creation guide](/docs/creating-an-assistant-with-the-api) covers catalog endpoints and configuration payloads.

## Information and actions

Use these resources to extend an Assistant beyond its instructions:

| Resource | Purpose | Example |
| --- | --- | --- |
| [Knowledge Base](/docs/knowledge-bases) | Supplies reference content in the Assistant's context | Product information and opening hours |
| [Tool](/docs/tools-and-mcps) | Lets the Assistant perform an action during the conversation | Fetch a booking, transfer a call, or send an email |
| [MCP server](/docs/tools-and-mcps) | Exposes external tools through the Model Context Protocol | A CRM integration with customer lookup tools |

Knowledge Base content is supplied to the Assistant as context. Fetch information that must be current, such as availability or order status, through a Tool.

Caller-specific values can be supplied through [Data Injection](/docs/data-injection). Define a [Data Extraction schema](/docs/conversational-data-extraction) when you need structured fields from the completed conversation.

## Next steps

- [Creating an Assistant](/docs/creating-an-assistant-with-the-api) covers the API configuration.
- [Channels & Calls](/docs/channels-and-calls) explains how callers reach the Assistant.
- [Advanced Settings](/docs/advanced-settings) covers speech detection, audio, and post-processing controls.
