flowise-nodes-niadra brings two nodes for Flowise:
- Niadra Customer Memory (Memory): connect it to an agent’s Memory input (Tool Agent, Conversational Agent). Before every model call it hands the agent the customer’s context as a system message right after the agent’s system prompt, followed by the chat’s own messages (from the memory you connect as Chat History, or the last 100 kept in the process). What the agent adds is recorded as the customer’s and the agent’s turns.
- Niadra Customer History (Tools): connect it to an agent’s Tools input. It gives the model
search_customer_history,get_customer_timelineandopen_history_item(andsearch_agent_memorywith Agent Memory on), bound to the chat’s customer. No tool has a parameter that names a customer.
Install
Flowise loads nodes from its components package. Build and copy, frompackages/flowise-nodes-niadra in the TypeScript SDK repository:
@niadra/sdk 0.3.0, which reaches npm when it is published. Tested with @langchain/core 1.2 through the same prompt the Tool Agent builds.
The five primitives
Limits
- Both nodes take the customer’s handle (bind it to a variable or the override config, never to text the model writes), the channel, the verification level the chat proved and the key (or the
NIADRA_API_KEYenvironment variable). The Flowise chat session is the Niadra conversation. - When Niadra is slow or down, the agent gets the chat’s messages alone.
- The node interface types and the memory node’s layout come from Flowise (Apache 2.0), credited in
NOTICE.
Next steps
n8n
the n8n community node.
Langflow
the three Langflow components.

