Open it, edit it
An agent is saved by name, loaded back into the editor, and changed like any other workflow — the same object your own team would build from scratch.
See agent workflowsNeuralWorks
A workspace assistant built from governed agents.
The employee-facing product on the NeuralSeek platform: four jobs in a working day, each a mAIstro agent you can open and edit inside your boundary.
Four jobs
One workspace, organised around four jobs.
The structure of the product, not a list of promises. Each job is a set of agents; what a set does is what you configure.
Meetings
Preparing for the ones on your calendar, and following up on the ones that have happened.
Email
Sorting what arrives, surfacing what needs you, and drafting what goes out — for you to send.
Documents
Drafting and editing documents from your governed data, in the platform's editor or exported to the format you work in.
Actions in systems of record
Turning a request in plain language into an update in the systems your organisation runs on — a ticket, a record, a message.
Where you work
Every agent on one screen.
The Agents screen groups the agents by when they run — before a meeting, after one — and gives each its own switch and a Test button.
Each row is an agent: what it does, whether it runs on its own, and the connection it writes to. The rows you read are the rows you switch on.
neuralworks › agents
Built from agents
Every quiet thing it does is an agent you can open.
No hidden model behind NeuralWorks. Each function is a mAIstro agent: readable as template code, editable in the visual editor, run on your terms.
Every one of these is a platform feature documented for mAIstro. NeuralWorks is what they look like assembled into a working day.
Run it on a schedule
The Agent Scheduler runs an agent at the minutes, hours, days and months you set, so the recurring jobs happen without someone starting them.
Read about the Agent SchedulerRun on your cadence, not its own
Which agents run on their own is a switch on each one; the rest run when you call them. Below a confidence you set, a fallback can notify a team or open an issue instead of answering.
See how agents are runGuarded on the way in
Prompt-injection protection, a profanity filter and personal-data masking are nodes an agent runs before a model sees the input.
See the guardrails
In the product
An action in a system of record.
A short capture from the agent editor: four nodes, and the output panel underneath.
A calendar event is updated, a mail is sent, the model is called, and a slide deck is written — four nodes in one agent, each visible, each recorded step by step. That is an action before it runs unattended.
Four nodes, one agent
What the security review sees
What your CISO gets.
A workspace assistant touches the calendar, the inbox, the documents and the systems of record. The guarantees come from the platform underneath.
Nothing here is specific to NeuralWorks. It is the platform's guarantee, and it applies to these agents because they are platform agents.
The boundary
It runs where the platform runs: managed, in your own cloud tenant, on your own cluster, or with no route to the public internet at all.
See where your data is allowed to liveThe model
Your choice, and swapped without a rebuild — so the model your contracts require is a configuration, not a constraint on the product.
See how it is deployedSensitive data
Personal data is found and masked before the model call, with the built-in patterns and your own under your control.
See the guardrailsThe audit trail
Every run is recorded step by step and can be inspected afterwards, so the answer to “what did it do?” is a record rather than a recollection.
See what is recorded
See it running inside your own boundary.
Talk to a NeuralSeek expert about how it fits your stack, where your data has to live, and the governance your auditors already expect.