Advanced Features
The features in this section extend a basic agent (see Core Concepts and Building an Agent) with memory, external data access, multi-agent workflows, richer runtime behavior, and platform integrations. Each is optional - use only what your agent needs.
Many of these capabilities are platform-only: they work when the agent is deployed on the NCP platform and are stubbed locally so your IDE can type-check them.
Runtime availability
Memory, Files, Metrics, Artifacts
Raise NotImplementedError - deploy the agent to use them
invoke_llm, send_status, send_content
Raise RuntimeError - only available inside the platform agent loop
Knowledge tools (search_knowledge, peek_knowledge)
Operate against the ingested knowledge/ collection, available when deployed
Do not rely on these platform-only helpers as ordinary local functions.
In this section:
Execution Hooks & Agent Handoffs - observe execution with hooks; transfer the conversation to another agent with handoffs.
Agent Memory - conversation-context management (
MemoryConfig) and a reference store for large results (Memory).Files API - read project and admin files uploaded to the platform.
Knowledge Base Tools - semantic search over ingested
knowledge/documents.Agent Skills - package reusable procedural knowledge as
SKILL.mdfiles.Multi-Agent Composition - delegate to your own agents, platform agents, and orchestration workflows.
Calling the LLM from a Tool - ad-hoc LLM calls with
invoke_llm.Custom Status & Streaming Output - report progress from long-running tools.
Generic SSH/HTTP Connectors - run commands or API calls through registered connectors.
Metrics API - query device, interface, link, and hardware telemetry.
Last updated
