Skip to content
One platform. From customer conversation to completed work.Take a closer look
NODEHUB AI AGENTS

Give AI a job worth doing.
And the context to do it well.

Bring instructions, knowledge, and selected tools together in a reusable agent. Make it part of a workflow your team can understand, test, and improve.

Workflows Customer request to coordinated response Illustrative configuration
A connected path from request to follow-up Select a step to explore
01 / INCOMING REQUEST02 / HUMAN REVIEW & FOLLOW-UP
Customer requestConversation received
Find the accountGet account record
Draft
UrgentStandardNeeds review
Needs escalation?AI decision
ApprovedKeep internal
Review escalationHuman approval
Notify the teamNotify the team
Create an issueCreate an issue
Update the accountUpdate account
Keep the contextAdd internal note
OpenAIGPT-4.1
KnowledgeSupport handbook
AirtableRead an account record
SlackRead channel history
LinearRead an issue
GitHubSearch repository issues
Example configuration
Scroll the canvas to explore every step.

An illustrative support workflow using Desk + Automation. Agents can also run in Automation with your connected services.

FROM A PROMPT TO A PROCESS

A useful agent has more than instructions.

Choose what it knows, what it can do, and where a person takes over. Keep those choices together as a version you can use again.

01

Start with the right context.

Give the agent a focused task, an available model from your AI account, and the knowledge sources that help it answer. Supply the workflow data the task actually needs.

02

Choose its tools deliberately.

Select supported operations and verified connections. Set execution limits, review requirements, and an expected output before you publish the agent.

03

See how the work happened.

Inspect the recorded run, tool calls, and outputs. Review a failed or uncertain action before retrying, then improve the next version with what you learned.

AI WITH A PLACE IN THE PROCESS

Let it prepare the next step.
Keep your team in the loop.

Gather context, summarize a request, or help route work. Put a human approval before the external action that needs review, and keep predictable steps as explicit rules.

01 / TRIGGERS

Start with a signal.

A customer message. A scheduled check. An event in another app. Give each one a clear next step.

02 / INTELLIGENCE

An agent with a job.
And the tools to do it.

Bring your model, knowledge, and selected operations into the same workflow.

Meet your next agent
03 / HUMAN JUDGMENT

You decide where AI stops.

Ask for approval before the next action. Give your team an explicit place in the process.

04 / LOGIC

Every path, intentional.

Conditions and AI decisions route each request to the next appropriate action.

05 / VISIBILITY

Keep the trail.

Inspect workflow runs and step results. Follow the work back to what actually happened.

06 / CONNECTIONS

Your stack belongs here.

Use supported service actions, HTTP requests, and agent tools to connect the work.

BETTER TOGETHER

The workflow does the work.
Desk keeps the context.

Bring customer conversations into the same workspace as the process. Connect the request, the decision, and the follow-up without rebuilding the story in another app.

Explore NodeHub Desk
DeskThe customer context
AutomationThe work that follows
THE PRACTICAL DETAILS

Before you
get started.

Can I build agents without using NodeHub Desk?

Yes. Automation supports agents for work across connected services. Workflows that read or change native NodeHub tickets, customers, or companies need Desk + Automation. The support example above uses both products.

Can I choose my own model?

Choose from the models available through the AI accounts configured in your workspace. Provider support, account permissions, and model availability determine what you can select. Model usage is billed separately by your provider.

Can an agent use every connection in my workspace?

No. An agent uses its selected tools and configured accounts. Review each operation and its permissions before publishing. A verified connection confirms account access, not that every possible action should be allowed.

Does installing an example start an agent?

No. A Hub workflow installs as a private draft. Configure its model, knowledge, and tools; test the published agent version; then review the workflow before enabling it.

What data goes to the model provider?

The configured request can include instructions, selected knowledge, supplied workflow context, and tool results. Review your provider's terms and your workspace configuration before sending sensitive information.

Start with one useful job.

Build the agent. Test the result. Connect the next step.

Get started with Automation