The REST node
Anything with an HTTP API is reachable as a step: URL, headers, body and operation set on the node, credentials pulled from the secrets you keep on the Configure tab.
Connectors
A connector is a node. Drop it into a workflow, call it like any other step, under the same guardrails and the same audit trail.












What ships prebuilt
Each family is a set of documented nodes in the agent editor, not a sample script — and the same node in every deployment.
Query the systems your records live in as a step inside a workflow, with the result available to the next node — instead of exporting data to somewhere the model can reach it.
The enterprise search and knowledge platforms your organisation has already standardised on, used as a retrieval source under the same grounding checks as any other.
Read from and write to the object stores and document libraries where policies, contracts and procedures are kept, including writing a generated document back.
Issue trackers, code hosting, boards and team messaging — where work is requested, and where the answer has to land for anyone to see it.
Mail, calendars, documents and files, so an agent can prepare and file work in the tools people already have open.
Mail over your own SMTP server and scheduled file drops over SFTP — the transport layer a regulated organisation still runs on.
Fetch and search public sources as an explicit, governed step — screened on the way in like any other untrusted content, never an implicit fallback when the corpus comes up empty.
The audit stream exports to object storage, to Splunk, to Datadog, or to your own collector — so the record lives in your security telemetry, not in a vendor console.
The node library
Each card is a node in the agent editor's function library: drop it on the canvas, fill in the query and the connection, and the result is a variable the next node can read.
MySQLConnects to a MySQL database, runs the SQL query you give it and returns the rows — as plain sentences or as CSV — for the next node to use.
When nothing prebuilt fits
A connector catalogue is finite and your estate is not. These are why a missing connector is an afternoon rather than a roadmap request.
Anything with an HTTP API is reachable as a step: URL, headers, body and operation set on the node, credentials pulled from the secrets you keep on the Configure tab.
Your agents are exposed as MCP tools, so an MCP client your team already runs can call a governed agent as a step — same guardrails, same run record.
A model is configured the way a connector is: a card, with the functions it serves ticked. Swap one and nothing built on top moves.
Start from something
Templates you open, point at your data and change. Every node is visible, and the moment you edit one it is your agent.
Slide 1 of 10
Connecting an AI system to your systems of record is where a security team stops nodding, so it is worth being precise.
No. A connector runs where NeuralSeek runs — for a regulated deployment, inside your boundary, over your network. The only call that can leave is the model call, and a model you host closes that too.
On the Configure tab, as named secrets. A workflow references a secret by name and never shows its value — it is inserted from a menu in the editor, not pasted into the agent.
Yes, through the same nodes that read. What a step sent and what came back is in the run record, and a saved agent only runs when something calls it — an API request, or a schedule you set.
Talk to a NeuralSeek expert about how it fits your stack, where your data has to live, and the governance your auditors already expect.