Persistent Memory in
n8n with the Kronvex Node

n8n workflows forget everything between runs. The official n8n-nodes-kronvex community node gives your AI agents a persistent memory layer — install it in two clicks, store and recall context without writing a single line of code.

Why n8n agents forget between runs

n8n's built-in "Window Buffer Memory" node keeps context alive for the duration of a single workflow execution. The moment the run ends — triggered by a webhook, a scheduled cron, or a manual test — that memory is discarded. The next execution starts from a blank slate and has no knowledge of what any user said before.

For one-shot automations this doesn't matter. But for anything that interacts with the same user over time — support bots, sales follow-ups, onboarding sequences, research assistants — it is a fundamental limitation. Kronvex stores memories in a persistent, semantically searchable database hosted in the EU, and the Kronvex Memory node makes this available directly inside n8n with zero configuration overhead.

Install the community node

Community nodes are installed directly from the n8n interface. No terminal, no npm, no Docker rebuild required.

Set up Kronvex credentials

You need a Kronvex API key and an Agent ID before you can use the node. Both are available for free from the Kronvex dashboard — no credit card required.

The credential is now available to all Kronvex Memory nodes in your instance. Use one credential for all your workflows — or create separate credentials for dev and production environments.

Core workflow: AI agent with persistent memory

The minimal pattern for a persistent AI agent in n8n is three nodes. Here is how to configure each one.

Node 1 — Trigger (Webhook or Chat)

Use a Webhook node (for API integrations) or the built-in n8n Chat Trigger. The trigger should output at minimum a userId (or sessionId) and the current userMessage. These two fields are what Kronvex will use to scope and search memories.

Node 2 — Kronvex Memory: Inject Context

Add a Kronvex Memory node immediately after the trigger. Set it up as follows:

This node returns a context string — a pre-formatted block of the most relevant past memories for this user and this message, ready to be inserted directly into an LLM system prompt.

Node 3 — AI Agent

Add an AI Agent node (or an OpenAI node). In the system prompt, reference the context output from the previous node:

You are a helpful assistant with access to the user's history.

{{ $('Kronvex Memory').item.json.context }}

Use the context above to personalise your response.
If no context is available, proceed normally.

Node 4 — Kronvex Memory: Remember

After the AI Agent responds, add a final Kronvex Memory node to persist the new information:

You can add a second Remember node in parallel to also store the user's original message with memory_type = episodic. Both messages will be retrievable in future runs.

Use case: sales intelligence workflow

Here is a more complete example. Your sales team receives a new prospect's LinkedIn profile via a webhook. The workflow checks what Kronvex already knows about this company, generates targeted research questions, and stores the new findings for future reference.

The next time your team encounters the same company, step 2 will immediately surface everything from previous calls, emails and research — scoped to that company's session ID. No spreadsheet hunting required.

Recall vs Inject Context — when to use which

The node exposes two operations for retrieving memories:

Node reference — all operations

Memory resource

Agent resource

Tips and best practices

Session ID — the most important parameter

Without a session_id, all users of your workflow share the same memory pool. A support bot that forgets session scoping will hallucinate context from other users' conversations. Always pass a stable user identifier — email address, Supabase user ID, phone number — as the session ID on every Remember and Recall call.

Memory types for better signal

Use memory types to improve retrieval quality. Store what happened as episodic, store facts as semantic, store recurring patterns as procedural. You can then filter by type in Recall to retrieve only the kind of information relevant to a given step in your workflow.

TTL for transient context

Not every memory needs to live forever. Use the TTL Days parameter to auto-expire short-lived context — support ticket details after 30 days, session summaries after 7 days, temporary user preferences after 90 days. This keeps your memory store lean and retrieval scores high.

Error handling — never break your workflow

Enable Continue on Fail on your Inject Context and Recall nodes. If the Kronvex API is temporarily unreachable, your workflow should still complete — just with an empty context. The AI Agent will respond without memory context rather than the whole workflow failing.

• One agent per environment — create separate Kronvex agents for dev, staging, and production so test data never pollutes production memories
• Summarise before storing — for long AI outputs, add a summarisation step before Remember to keep memory content concise and retrieval fast
• GDPR compliance — all data is stored in the EU (Frankfurt). For deletion requests, use the Agent → Delete operation or the List + Delete Memory operations to wipe a user's session
• Parallel recall — n8n supports parallel branches. Run your Inject Context node in parallel with other setup nodes; Kronvex recall typically completes in under 40ms