Agentic
Agentic is where an organisation builds its own AI agents in the NeonNow Admin app. An agent is given instructions and tools, tested in a playground, and published when it is ready. Published agents answer phone calls on the front line, or work through tickets, messages and data in the background, and every conversation is recorded so it can be reviewed afterwards.
What You Can Do with Agentic
- Describe an agent in plain words and let IQ draft it, or configure every field by hand
- Answer inbound calls with a voice agent on its own phone number
- Automate text and data work with an agent that runs when your systems call it
- Give agents abilities by connecting them to your own systems as tools
- Test a draft by text or voice before anyone outside the team reaches it
- Publish when ready, while the version callers reach stays unchanged until then
- Review every session, with its transcript, the tools it used, a score and a summary
- Hand conversations to people on your team when an agent should not finish them alone
- Connect AI apps such as Claude, ChatGPT or Cursor, so they can read and build agents for you
How the Parts Fit Together
An organisation holds one or more environments. Each environment holds its own agents and its own settings, so work in one never touches another. An agent combines instructions that say how it should behave with tools it can use to act in your systems. A Frontline agent also has a Knowledge setting. Every agent has a draft that is edited and tested, and once published, a live version that callers and integrations reach.
Instructions
Instructions are the plain-language brief an agent follows: what it is for, what to check before acting, and what it must never do. For a voice agent they also include the greeting a caller hears first. See The Agent Editor.
Knowledge
A Frontline agent has a Knowledge switch, labelled Knowledge base agent in the editor, described as "Allow the agent to consult the configured knowledge base." It is on by default and only turns that setting on or off. Background agents have no knowledge setting. Facts an agent needs, such as opening hours, go in its instructions or come from a tool. See Giving an Agent Knowledge.
Tools
A tool connects an agent to one of your systems so it can act, for example looking up a booking or moving it to another day. The agent decides when to use a tool, and each use is recorded in the session. See Tools and Knowledge.
The Agent Lifecycle
Every agent moves through the same loop. Changes are made to the draft, tested, and published. Live sessions then show how the agent performs, which feeds the next round of changes.
Key Terms
Other pages link back to these definitions rather than repeating them.
Environment
An environment keeps a set of agents and their settings together. Each one has a type, Sandbox, UAT or Production, which labels what it is for: experimenting, testing with the team, or serving customers. Settings such as the Twilio connection belong to one environment, so a Sandbox can use a test account while Production uses the real one. See Environments.
Agent
An agent is an AI worker built from instructions and tools. It lives in one environment, and it is either a Frontline or a Background agent. See Agents.
Frontline and Background Agents
A Frontline agent is a voice agent. It answers inbound calls, either on a phone number assigned to it or when your own telephony reaches it by its agent ID. A Background agent works on text and data behind the scenes, such as summarising support tickets or qualifying leads from a form. It runs when another system, a flow or another agent calls it. The type is chosen when the agent is created.
Draft and Live
Every agent has a draft, which is where all editing happens. Publishing copies the draft into the live version, and the live version is the one callers and integrations reach. Editing a published agent changes the draft only, so callers keep getting the live version until the changes are published. See Publishing and Versions.
Session
A session is one conversation or one piece of work an agent handled. For a Frontline agent a session is a Call, and for a Background agent it is a Run. Each session keeps its transcript, the tools the agent used, a score and a summary. See Sessions.
Live and Evaluate Sessions
A Live session is real traffic: a customer calling, or one of your systems sending work. An Evaluate session is a test conversation held in the playground or in the builder and editor test panels. In an agent's Sessions list, Evaluate sessions carry the IQ mark, a Playground badge and the word "Evaluate" in their details, and the All, Live and Evaluate filters keep test traffic apart from real traffic.
Playground
The playground is a test conversation with an agent, where you play the customer. It is the Playground tab on a published agent's page, and it can test the live version or the draft. A Frontline agent is tested by voice and a Background agent by text. Every playground test is recorded as an Evaluate session, never as a Live one. Before an agent's first publish, testing happens in the builder's Test It Out or Try It Out step, or the editor's Test panel, instead. See Playground and Testing Before You Publish.
IQ
IQ is the AI assistant built into the Admin app. In Agentic it drafts an agent from a plain description, asks follow-up questions to fill the gaps, plays the caller during tests, reviews the results and suggests fixes. In the agent editor it sits in the Ask IQ panel. See Building with IQ.
Human Escalation
Human escalation hands a conversation from an agent to a person on your team when the agent should not handle it alone. The person picks it up in the workspace app. It is optional, set up per environment by an Organisation Admin, and runs on the organisation's own Twilio account. See Human Escalation and The Workspace.
Connected Apps
A connected app is an AI tool, such as Claude, ChatGPT or Cursor, that an Admin or Organisation Admin has connected to Agentic. Once connected, the app can be asked things like "Show me my agents in Production" or "Why did my agent's score drop?", and with build access it can create and change agents too. See Connected Apps.
Who Can Use Agentic
Agentic is available to Admins and Organisation Admins. Both can build, test and publish agents and manage environments. Settings is for Organisation Admins only. Other roles do not see Agentic in the sidebar. See Roles and Permissions.
Documentation Structure
| Section | What it covers |
|---|---|
| Getting Started | Who has access, turning the product on, the first environment, and the Getting started checklist |
| Quickstart | A first agent from description to publish |
| Environments | Sandbox, UAT and Production, and moving between them |
| Agents | Building, editing, publishing, phone numbers and deleting agents |
| Testing and Monitoring | The playground, sessions and metrics |
| Connected Apps | Connecting AI apps such as Claude, ChatGPT or Cursor |
| Settings | Twilio and human escalation, for Organisation Admins |
| The Workspace | The app people use to handle escalated conversations |
| Roles and Permissions | What Admins and Organisation Admins can each do |
| Knowledge Base | Short answers to common questions |