n8n brings standalone agents into its workflow platform

n8n supports standalone AI agents alongside traditional workflows, combining models, instructions, tools, triggers and configurable memory. Users can describe an agent to its AI Assistant or configure it manually, then test a draft before publishing. Existing workflows, 500+ integrations, custom code and Model Context Protocol servers can become tools agents call. Unlike the AI Agent node, which operates within a fixed workflow, standalone agents maintain their own identity, sessions and triggers.

Agents share the platform’s permissions and execution infrastructure. Credentials are scoped to individual tools, sensitive actions can require human approval, and published versions remain separate from draft edits. Session and execution logs capture inputs, tool calls, outputs and errors. Each conversational turn counts as an execution, while workflow-tool and sub-agent calls do not count separately. Session memory is enabled by default; optional memory across sessions requires an OpenAI credential.

Agents are available on n8n Cloud and self-hosted deployments, although the feature remains behind a preview flag during active development. Current limitations include no always-on monitoring of event streams, no open-ended multi-agent orchestration beyond bounded sub-agent delegation, and no discovery or invocation by external systems. Agents do not automatically learn or improve themselves. A file-based knowledge base is available on n8n Cloud.