n8n Workflow & Agent Memory

How it works

1. Install n8n-nodes-memorysync from n8n’s community-nodes panel.
2. Connect your MemorySync API key once as a credential — it is tested with a real query the moment you save it.
3. Drop a MemorySync node into any workflow to add, search, recall, list or delete memories.
4. Optionally plug MemorySync Chat Memory into an AI Agent’s memory port, and hand the MemorySync node to the agent as a tool it calls on its own.

Before you start

1. A MemorySync API key. Create one in the dashboard under Settings → API Keys, inside the project whose memories you want to use.
2. A self-hosted n8n instance. Installing community nodes from npm is a self-hosted feature; n8n Cloud only lists nodes n8n has verified (our verification is in progress).
3. Owner access to that instance — only instance owners can install community nodes.

Install the nodes

Installing in Docker images (optional)

Running n8n from your own image? Preinstall the package instead of clicking through the panel:

RUN cd /home/node/.n8n/nodes && npm install n8n-nodes-memorysync

# Community nodes must stay enabled (default: true)
ENV N8N_COMMUNITY_PACKAGES_ENABLED=true

Your first memory workflow

A two-step smoke test: Manual Trigger → MemorySync (Add Memory) → MemorySync (Search Memories).

1. 1Add a MemorySync node. Set Operation to *Add Memory*, End User ID to alice, and Text to Alice is vegetarian and never eats mushrooms.
2. 2Add a second MemorySync node. Set Operation to *Search Memories*, the same End User ID alice, and Query to what does Alice eat?
3. 3Select Test workflow. The search step returns the stored dietary fact as a regular n8n item you can map into any later node.

Give an AI Agent persistent memory

The MemorySync Chat Memory node plugs into the AI Agent’s *Memory* port. The agent’s conversation history then lives in MemorySync instead of process memory, so sessions survive restarts, redeploys and scale-out — and the same user’s durable memories stay available to every other MemorySync surface. The transcript is kept in the session’s conversation history, apart from the user’s memories: MemorySync extracts durable facts from the user’s turns, and only those facts appear as memories (Search, Recall, Get Many, the dashboard).

1. 1Build the agent as usual: Chat Trigger → AI Agent, with your chat model connected.
2. 2Under the agent’s Memory port, select MemorySync Chat Memory.
3. 3Set Session ID to your conversation key — for the Chat Trigger that is {{ $json.sessionId }} — and End User ID to the person chatting, such as {{ $json.userId }} or an email address.
4. 4Leave Context Window Length at 10. That is how many previous turn pairs the agent sees on each run.
5. 5Optional: under Tools, also attach a MemorySync node set to *Search Memories*. The node is marked usable as a tool, so the agent can look up long-term facts on its own.

Operations

The MemorySync node groups six operations under the Memory resource. Every one is scoped by End User ID, so one user’s memories can never leak into another user’s workflow run.

Add Memory

Add Conversation Turn

The turn is kept in the session’s conversation history, apart from the user’s memories. A user turn is also sent to fact extraction, and only the durable facts in it become memories; an assistant turn is kept in the history and nothing is extracted from it. The node outputs accepted, turnId (the history turn), memoryId (the same turn in the m_… form the Delete operation accepts) and alreadyExists (true when a retried execution sent the same turn again).

Search Memories

Recall Context

Recall Context returns one grouped, prompt-ready context block — paste it straight into a downstream AI step’s system prompt instead of assembling search results yourself.

Get Many

Delete Memories

Choosing an End User ID

The End User ID is a stable string you pick to identify whose memories these are: your app’s internal user ID, an email address, or a UUID. Use the same value across Add, Search and Recall or recall returns nothing. It also powers billing attribution and the per-user isolation guarantee.

Quotas and plan limits

Quota behaviour never breaks a workflow. On free and paid plans, hitting the monthly limit makes Add Memory report stored: false, accepted: true and reads return empty results — the workflow keeps running. Evaluation keys instead surface a truthful 429, so you find out during testing, not in production. Bad metadata JSON, and a Memory IDs list without a single valid ID, raise clear node errors before any network call; entries that are not memory IDs are skipped.

Billing counts requests, not results. Add Memory and Add Conversation Turn are one add per execution, whichever role the turn has. The Chat Memory sub-node saves the user’s message and the agent’s reply as two separate turns, so one exchange is two adds. It reads the history once when the agent loads its memory, and n8n reads it once more after the exchange is saved, to show the conversation in the execution log, so one AI Agent run is two adds and two retrievals. Search Memories, Recall Context and Get Many are one retrieval each, whatever they return. Deletes are free.

Troubleshooting

• The nodes don’t appear in the picker — community nodes install on self-hosted n8n only, and only instance owners can install them. On n8n Cloud, watch the community-nodes panel; our verification is in progress.
• The credential test fails — the key is wrong, revoked, or was created outside a project. Regenerate it in the dashboard under Settings → API Keys, inside a project.
• Search returns nothing right after an add — check the add node’s output first: low-value text is skipped (status: skipped) by design. Also confirm both nodes use the same End User ID.
• A retried execution didn’t store a duplicate — that’s intentional. Turns carry deterministic idempotency seeds, so replays report alreadyExists: true instead of double-writing.
• Add Conversation Turn text doesn’t show up in Search or Get Many — by design: the turn lives in the session’s conversation history, and only the durable facts extracted from user turns appear as memories. Give extraction a few seconds.
• MemorySync Chat Memory is missing from the Memory port, or says "update n8n" — the sub-node needs n8n 2.16.0 or later (the first release that ships @n8n/ai-node-sdk). The MemorySync workflow node keeps working on older releases. Update n8n and reopen the picker.

Supported versions

How it compares

Where to go next