Design Studio

Design Studio is where your team sets up the system before it runs: define agent behavior, author runbooks and network goals, configure skills, and set the guardrails every agent operates within.

Agent Categories

Every agent in Agent Studio belongs to one of a defined set of categories, fixed the moment it’s created. The category decides how much you can customize it, which guardrails apply, and who’s allowed to touch its behavior.

System Agents: the core agents NetBrain ships (Deep Diagnosis, Assessment, Change, Companion). Their description and system prompt are locked, so behavior stays the same across every deployment.
Custom Agents: troubleshooting agents your team builds from a baseline. Every field, including the system prompt, is yours to edit, starting from Diagnosis, Change, Assessment, Companion, or a blank General baseline.
Function Agents: single-purpose building blocks like Golden Config, Path, and Intent, selectable from the agent picker. Read-only to customers.
Subagents: delegation helpers like Intent Assistant and ADT Assistant that nest inside other agents. Also read-only.
System and Function agents stay current automatically through Knowledge Cloud updates.

Intent Authoring

This is where you define and version Network Goals and Spec, the codified rules for what your network is supposed to do. Network Goals and Spec is the broader framework for defining, versioning, and enforcing network rules; this page covers the authoring surface where you build them.

Runbook Authoring

Build, test, and publish a governed workflow right here. Once published, it becomes a Runbook Template the Change Agent, the Companion Agent, or an engineer can reuse instead of starting from scratch. Automation Tools covers how a published runbook executes.

Automation and Skills Configuration

Skills authoring lives here too: teach NetBrain’s agents your network’s own language, no model retraining needed. Automation Tools covers the authoring workflow and Skills Assistant in full.

Access Guardrails

Administrators hold one guardrail that no agent configuration can override. Put a device in Excluded devices, and its sensitive data, Config, CLI, and NCT, never reaches the LLM. This is enforced at the tool level, not through agent-specific settings.

Command Exclude restricts which CLI commands agents can run, for example limiting ping and traceroute to approved IP ranges, or blocking commands entirely on sensitive devices.
RBAC and permissions decide who can view, build, or approve inside Design Studio. Agent customization stays a controlled capability, not an open door.