Key Concepts

Understanding these concepts will make everything else in the platform click into place. Each section below covers what the concept is, when you use it, and how it connects to the rest of the system.

Agents

An agent is a voice AI persona. It defines how the AI presents itself during a call -- what it says, how it reasons, what voice it speaks in, and what actions it can take. Every call in zoxaAI is powered by an agent.

Core settings

Every agent has three required settings:

Beyond these three, agents have extensive configuration across five editor tabs (Agent, Engine, Tools, Call, Analytics) covering welcome messages, interruption behavior, silence detection, speech speed, temperature, tool attachments, call duration limits, and more. See the full Agent configuration reference.

Calls

A call is the real-time audio connection between a person and your agent. zoxaAI supports three call types, each suited to a different scenario.

All three call types produce the same run record when they end, so your analytics, transcripts, and recordings are consistent regardless of how the call was initiated.

How to start a call

Runs

A run is the complete record of a single call. Every call -- regardless of type -- creates a run when it ends. Runs are the primary unit of data in zoxaAI: they contain everything you need to evaluate agent performance, debug issues, and extract business value.

Run record fields

Run statuses

Where to find runs

• Agent Runs tab: Shows runs for a specific agent.
• Calls page (sidebar): Shows all runs across all agents. Supports filtering by status, transport, date range, and tags. See Call History.
• API: Query runs programmatically via the API.

Campaigns

A campaign is a batch of outbound calls made to a list of contacts. Instead of triggering calls one by one, you upload a CSV of phone numbers, assign an agent, and zoxaAI works through the list automatically.

Campaign lifecycle

phone_number,first_name,account_id,appointment_date
+14155552671,Sarah,ACC-1234,2026-06-15
+14155552672,James,ACC-5678,2026-06-16

Tools

Tools give your agent the ability to take actions during a call, beyond just speaking. When the LLM determines that a tool is needed based on the conversation context and your instructions, it triggers the tool automatically.

How tools work

1. You define tools with a name, description, and parameters (for HTTP API tools).
2. You attach them to an agent.
3. During the call, the LLM reads the tool name and description to decide when to use it.
4. When triggered, the platform executes the tool (e.g., sends the HTTP request) and returns the result to the LLM.
5. The LLM incorporates the result into its next response.

For detailed configuration, see the Tools reference.

Knowledge Base

A knowledge base lets your agent look up information during a call using retrieval-augmented generation (RAG). You upload documents, zoxaAI indexes them, and when a caller asks something the agent does not know from its system prompt alone, the agent searches the knowledge base and uses the retrieved content to answer accurately.

Supported file types

Retrieval modes

When you upload a document, you choose one of two retrieval modes. This cannot be changed after upload.

How it works during a call

1. The caller asks a question the agent cannot answer from its system prompt alone.
2. The LLM generates a search query based on the conversation context.
3. zoxaAI searches the knowledge base and retrieves relevant content (full document or matching chunks).
4. The retrieved content is injected into the LLM context.
5. The LLM uses the content to generate an accurate, grounded response.

For full configuration details, see the Knowledge Base reference.

Providers

zoxaAI provides managed access to a wide range of AI and telephony providers. You do not need to create separate accounts or manage API keys -- everything is included with your zoxaAI account.

Models are a curated catalog — you pick from the models each provider offers below. Every model carries latency, cost, and quality metadata, and off-catalog model ids are rejected at save time.

LLM providers (4)

TTS providers (6)

STT providers (5)

Telephony providers (8)

For detailed provider configuration and model lists, see:

How everything connects

Here is how the core concepts relate to each other in practice:

Agent
  ├── System prompt, voice, LLM model
  ├── Tools (HTTP API, Call Transfer, End Call)
  ├── Knowledge Base (optional, shared across agents)
  │
  ├── Receives a Call (Web, Phone, or Campaign)
  │     └── Runs the STT → LLM → TTS pipeline in real time
  │     └── Uses tools and knowledge base as needed
  │
  └── Produces a Run when the call ends
        ├── Transcript, recording, duration, cost
        └── Accessible via dashboard or API

A Campaign is an orchestration layer on top of calls -- it takes a list of contacts and triggers one call (and therefore one run) per contact, with concurrency, retries, and scheduling handled automatically.

Next steps