Agents

Agents enable NovaAI to perform multi-step tasks by combining language models with external tools, APIs, and workflows. They can reason through complex problems, make decisions, and execute actions to accomplish a goal.

Overview

Unlike standard chat models that only generate text, Agents can plan, use tools, retrieve information, and interact with external systems before returning a final response. This makes them ideal for building autonomous assistants and workflow automation.

Note: Agents can use multiple tools during a single request, depending on the task and available permissions.

How Agents Work

Rendering diagram…

Create an Agent

1const agent = await client.agents.create({
2  model: "nova-agent-1",
3  instructions: "You are a helpful travel assistant.",
4  tools: ["web_search", "calculator"]
5});
6 
7console.log(agent.id);e

Agent Properties

model

string

required

Agent model identifier.

instructions

string

required

Defines the agent's behavior.

tools

array

optional

Tools available to the agent.

memory

boolean

optional

Enables conversation memory.

Common Use Cases

Customer support assistants
Research and web search
Workflow automation
Travel and booking assistants
Data analysis
Internal knowledge assistants

Tip: Give your agent clear instructions and only enable the tools it actually needs. Simpler agents are generally faster, easier to manage, and more predictable.

Next Step

Continue to Streaming to learn how NovaAI delivers responses in real time as they're generated.