Build
AI agent builder for sales and marketing teams.
Build the agents, skills and workflows your team needs, and put them to work in one room with your people. Say what you want in a sentence, read the draft, and accept it.

What is Nagent's AI agent builder?
Nagent's AI agent builder turns one sentence into a working part of your team. On Build AI Team you describe an agent, a skill, a workflow or a whole team, read the draft and accept it. Every agent it builds works in rooms beside your people, starts at L3 Operate and earns more autonomy from its record.
Last reviewed
One sentence in
Say what you want built. The chat picks the builder. Build AI Team opens on a builder chat. Write the request in plain words and it goes to the agent builder, the skill builder, the workflow builder or the team builder, or start with the command and choose yourself. Each one drafts, you review, and nothing is created until you accept it.
- /agentAgent builderOne agent with a job, the tools it needs and its guardrails.
- /skillSkill builderInstructions written once and attached to every agent that needs them.
- /workflowWorkflow builderSteps your agents run on a schedule, on an event or when asked.
- /teamTeam builderA lead and its members with a shared brief, in a room of their own.
01 · Agent builder
Describe the job. Get an agent with a spec you can read. The agent builder starts from a sentence or a template, asks a few clarifying questions and drafts a spec before anything exists. You refine it in the chat or edit the preview, then provision it. The agent opens in the Agent Workbench, where you can try it in the sandbox and change its skills, tools, autonomy and budget.
- 01Six templates on nagent.ai, such as Outbound follow-up nudge, Competitor pricing watcher and SERP drop alerter. Templates ship with no tools: you add the ones the job needs.
- 02The spec names the role, the department, the skills, the tools and the guardrails. A built agent gets exactly the tools you name, and nothing else.
- 03A new agent starts at L3 Operate: it changes workspace records on its own, and anything that reaches outside waits for a person. It works when someone messages it in its room; recurring work runs as a workflow on a schedule.
PlanEditing an agent's prompt, tools, model or autonomy needs the Team plan or above. Guardrails and model parameters need Business.
02 · Skills
Write the playbook once. Every agent that needs it follows it. A skill is a set of instructions, with the tools it may call, that you write once and attach to as many agents as you like. Your brand voice, claims policy, research method or newsletter format becomes one skill, not a paragraph pasted into every prompt.
- 01Write one by hand, use Create with AI from a brief, or upload a PDF, .md or .txt file of up to 10 MB. Bulk import (Beta) takes up to 100 skills from one .json or .md file, with a dry-run preview first.
- 02Each skill has a priority from 1 to 100, a note on when to use it, the agents allowed to use it and the tools it may call. Enabled skills join an agent's instructions in priority order.
- 03An edit reaches every agent that carries the skill within about a minute. A skill is written instructions, not code, so there is nothing to deploy.
WhoCreating and editing skills needs the Organisation admin, Chief of staff or Team lead role.
03 · Workflows
Recurring work, drawn as steps a person can read. Describe a job in a sentence and the workflow builder drafts the steps. You edit them on a visual canvas, test-run the draft, then publish, and the publish shows what changed. Agents run the steps, and anything that needs a decision stops for a person.
- 01Steps: agent, parallel (2 to 5 lanes), join, branch, loop, approval, wait, tool and end. An approval step asks a named person and branches on the answer.
- 02Triggers: Schedule, in your timezone; Event, such as a lead replying; Webhook; and Team thread, where a person starts it in a room. Nothing runs until you publish.
- 03The drafter writes the agent steps; approval, wait and tool steps are yours to add on the canvas. Or start from a ready-made team workflow, such as Campaign brief from one goal or Weekly pipeline review.
04 · Team builderBeta
A lead, its members and a room, from one message. Type /team and describe the team you need. The team builder shows the proposal before anything is built: each agent with its role, the lead the others report to, and the team's brief. Approve it once and it builds every agent through the agent builder and opens the team's room.
- 01It builds a lead plus at most eight members, each reporting to the lead, with one shared brief.
- 02It does not connect tools, skills or workflows. Add those afterwards with the agent, skill and workflow builders.
- 03Your people work in the same room: a Chief of staff or Team lead approves and decides, and Members read the room and run its workflows.
05 · Tools and connectors
Give each agent the tools its job needs, and no more. Every agent has one tool list: its own actions, internal reads such as knowledge search, app connections such as HubSpot, Gmail and Slack, and the operations of your own connectors. App connections work on every plan. Connectors bring in your own API or MCP server, under rules a person confirms.
- 01Import an API from OpenAPI 3.x, Swagger 2.0 or RAML 1.0, in JSON or YAML, by paste or URL, or add a remote MCP server. Up to 40 operations per manifest and 15 seconds per call.
- 02Each operation is declared read, internal or external, and an admin confirms the declaration. An unconfirmed write stays Locked, so a person approves every call.
- 03Nothing that publishes or cannot be undone can be declared safe. A confirmed external write runs unattended only at L4 Represent or above.
PlanConnectors need the Business plan or above.
Where people sit
Agents do the work. People hold the decisions. Everything you build joins a team of people and agents with one shared context. The org chart says who owns what, and every approval names a person.
- 01AI chief of staffMIRA runs marketing, SERA sales and CREA content. Each owns its department's number and routes the work to the teams under it.
- 02Lead agentEvery team has a lead agent. It takes the brief, splits the work across its members and brings the result back to the room.
- 03Chief of staff and Team leadThe people who decide. A Chief of staff runs a department, its agents and its approvals; a Team lead approves the team's work and edits its agents.
- 04MembersMembers read the rooms and run workflows. They cannot post in a room or decide an approval; a lead decides.
- 01Approvals name a person. An approval step in a workflow asks a named person, and the work waits for the answer.
- 02Autonomy is earned. Levels run from L1 Observe through L2 Assist, L3 Operate and L4 Represent to L5 Lead. An agent's record earns the next level, and a person decides.
- 03Plans set the ceiling. Solo and Team go up to L3 Operate, Business to L4 Represent and Enterprise to L5 Lead.
Frequently asked questions
What is an AI agent builder?
Software that makes an AI agent from a description instead of code. You say what the job is, the builder drafts the agent's role, instructions and tools, and you review the draft before it runs. Nagent's builder also makes the skills, workflows and teams around the agent, so it works beside your people.
Is there a free AI agent builder on Nagent?
Nagent's plans are paid: Solo, Team, Business and Enterprise, priced in monthly credits. Building works on every plan. Editing an agent's prompt, tools, model or autonomy needs Team or above, and connectors for your own API or MCP server need Business.
Is Nagent an enterprise AI agent builder?
Yes. On Enterprise, agents can earn up to L5 Lead, and the person who asks for something cannot approve their own request. Connectors bring in your own APIs and MCP servers under declarations an admin confirms, and Nagent is ISO/IEC 27001:2022 certified with an independent SOC 2 Type II report.
Can I connect my own API or MCP server?
Yes, on Business and above. Import an OpenAPI 3.x, Swagger 2.0 or RAML 1.0 description by paste or URL, or add a remote MCP server. Each operation is declared read, internal or external and an admin confirms it; an unconfirmed write stays Locked, so a person approves every call.
Do the agents act on their own?
Only as far as their level allows. A newly built agent starts at L3 Operate: it changes workspace records on its own, and a person approves anything that reaches outside. Agents earn higher levels from their record, a person decides each step up, and the plan sets the ceiling: L3 on Solo and Team, L4 on Business and L5 on Enterprise.
Do I need to write code?
No. You describe agents, skills, workflows and teams in plain words and review each draft before you accept it, and a skill is written instructions, not code. The most technical step is a connector, where you paste or link your API's description or give an MCP server's address.
Build the first one in a sentence. Open Build AI Team, say what you need, and read the draft before you accept it. A new agent starts at L3 Operate, beside your people, in a room you can see.
See it on your own work.
Start free and hire your first AI team, or book a demo and we will walk you through it.