n8n

Invoke Zihin agents from n8n workflows with the Zihin Agent node. The node calls a hosted agent (over @zihin/agent-client), which runs the full loop — tools, memory, governance — and returns the result.

info

The Zihin Agent node ships in n8n-nodes-zihin ≥ 0.9.0 (npm). Update the package if you don't see it.

Installation

Community Nodes (recommended)

1. In n8n: Settings → Community Nodes → Install
2. Search for n8n-nodes-zihin
3. Click Install

npm (self-hosted)

npm install n8n-nodes-zihin

Requires n8n ≥ 1.20 and Node.js ≥ 18.

Credential setup

1. Get your API key at console.zihin.ai → Settings → API Keys (zhn_live_*).
2. In n8n: Credentials → Add Credential → Zihin API.
3. Enter your API key. It is validated against the Zihin API automatically.

The Zihin Agent node

Add the Zihin Agent node to a workflow and configure:

Field	Description
Agent	The agent to invoke — loaded from your tenant (dropdown).
Message	The input sent to the agent.
Session ID	Reuse to continue a conversation; the output returns the session ID for the next call.

Options:

Option	Default	Description
Temperature	—	Sent to the agent, but the /stream endpoint does not apply it (reported in ignored_fields). Configure it in the agent's llm_config.
Max Tokens	—	Same as Temperature: not applied by /stream. Configure it in the agent's llm_config.
Timeout (Ms)	120000	Max total time for the call. Sent to the backend as options.timeout_ms, which only shortens the turn deadline (180 s on the platform). See Agent invocation contract.
Base URL	—	Override for a private deployment.
Allow Custom Base URL	false	Allow a base URL outside the Zihin domain whitelist (your key is sent there).
Include Raw Events	false	Include the raw SSE events in the output (debugging).

Output — the node returns:

{
  "content": "…the agent's answer…",
  "sessionId": "sess-123",
  "model": "anthropic.claude-haiku-4-5",
  "usage": { "inputTokens": 0, "outputTokens": 0, "totalTokens": 0, "costUsd": 0 },
  "sources": []
}

Example workflow

[Trigger] → [Zihin Agent] → [Respond / next step]

1. Add any trigger (Webhook, Chat, Schedule…).
2. Add a Zihin Agent node and connect the trigger.
3. Select the Agent, map the Message (e.g. {{ $json.text }}), and optionally pass a Session ID to keep context.
4. Use content downstream; persist sessionId to continue the conversation later.

Conversation continuity

Store the returned sessionId (e.g. keyed by chat/user) and feed it back into the next call so the agent remembers the conversation.

Endpoints used

Operation	Endpoint
Invoke agent	POST /api/v2/agents/:agentId/stream (SSE)
Agent picker	GET /api/v2/agents

Legacy nodes (deprecated)

Earlier versions shipped a Zihin Chat Model sub-node (a direct LLM gateway over /api/v3/llm/public/call) and standalone media nodes (image / audio / document). These are deprecated in favor of the agent pattern and hidden by default. Existing workflows keep working, but new automations should use the Zihin Agent node.

Troubleshooting

Issue	Solution
"Invalid API Key"	Verify the key is active at console.zihin.ai.
Agent not in the dropdown	Confirm the key's tenant owns the agent; reopen the node to reload.
Timeout errors	The turn deadline is fixed by the platform (180 s) and cannot be extended; Timeout (Ms) only shortens it. For longer tasks use an async trigger. See Agent invocation contract.
Localhost / private host blocked	Enable Allow Custom Base URL (and HTTP for localhost) only in development.