Overview
NCP SDK
The NCP SDK lets developers build, package, and deploy custom AI agents that run on the Network Copilot platform. An agent is a small Python project that pairs your own tools (Python functions the model can call) with an LLM, and can use platform capabilities such as data connectors, memory, files, knowledge search, and telemetry.
What you can build
Custom agents with your own tools for querying APIs, running commands, and processing data
Agents that use platform data connectors and platform-hosted agents as building blocks
Agents that apply memory, read platform files, search a knowledge base, use skills, stream progress, and read telemetry
Capabilities
Tools, multi-tool agents, and parallel tool execution
Data connectors and MCP servers
Agent memory - conversation context management plus a reference store for large results
Files API - read project and admin files uploaded to the platform
Knowledge base search
Agent skills
Execution hooks and agent handoffs
Multi-agent composition and orchestration
Calling the LLM from within a tool
Custom status and streaming output
Generic SSH/HTTP connectors
Metrics and telemetry
Deployment paths
There are three distinct ways an agent reaches the platform - keep them separate:
CLI deploy -
ncp deployuploads a packaged agent and runs it containerized (isolated from the platform process).Admin onboarding -
ncp onboard(admin-only) integrates a custom agent within the platform process (no container isolation) and exposes it to users through role-based CustomAgents permissions in Admin > Roles & Permissions.UI upload - agents can also be uploaded through the platform UI. Submissions by users with the Submit Agents role go through administrator approval before becoming available. This UI path is the upload-and-approval workflow (see Agent Onboarding).
Resources
PyPI package: https://pypi.org/project/ncp-sdk/
SDK examples: https://github.com/AvizNetworks/ncp-sdk-examples
Sample agents: https://github.com/AvizNetworks/ncp-sdk-agents
Last updated