Getting Started

n8n Workflow & Agent Memory

Two nodes give n8n long-term memory: workflow operations (add, search, recall, list, delete) that an AI Agent can also call as a tool, and a Chat Memory sub-node that makes agent conversations persist across executions, restarts and redeploys. Install n8n-nodes-memorysync from the community-nodes panel.

Two nodes, one package

Node	Plugs into	What it does
MemorySync	Any workflow (and AI Agents as a tool)	Six operations: Add Memory, Add Conversation Turn, Search, Recall Context, Get Many, Delete
MemorySync Chat Memory	The AI Agent’s memory port	The agent’s conversation history lives in MemorySync — sessions survive restarts, redeploys, and weeks between chats

	Mem0	Supermemory	Zep	MemorySync
Community node ships	△ workflow ops only	✗ HTTP-request recipes in docs	△ built-in node deprecated; hard-removed in n8n v3	✓ `n8n-nodes-memorysync`
AI Agent memory sub-node	✗ none	✗	was the removed node	✓ maintained, on the official `@n8n/ai-node-sdk` path
Agent can call memory as a tool	✗ not marked usable	✗	✗	✓ `usableAsTool`
Credential test	✗ none — bad keys fail mid-workflow	—	—	✓ real query at setup time
Retry safety	✗ re-runs re-store	—	—	✓ deterministic idempotency seeds
Runtime dependencies	several	—	—	✓ zero (verified-program rule)

Install

Settings → Community Nodes → Install
Package name: n8n-nodes-memorysync
→ restart is handled by n8n; both nodes appear in the node picker

Create a MemorySync API credential with a key from the dashboard. The credential test performs a real one-result query, so a wrong key fails at connect time — not three steps into a production workflow. n8n Cloud lists only verified community nodes; the verification submission is in progress, so Cloud users should watch the community-nodes panel.

The AI Agent memory sub-node

Chat Trigger ─▶ AI Agent ◀─memory─ MemorySync Chat Memory
                        ◀─model── OpenAI Chat Model
                        ◀─tool─── MemorySync (Search Memories)

Guarantee	How
Sessions survive anything	History lives in MemorySync, not in-process — restarts, redeploys and scale-out don’t lose a turn
Retries never duplicate	Every stored turn carries a deterministic idempotency seed derived from role, session and content
An outage never breaks the agent	getMessages fails open to “no history this run”; the agent still answers
Agents cannot bulk-delete	`clear()` is a deliberate no-op — deleting a customer’s history stays an explicit human action (the workflow node’s Delete operation)
Same memory everywhere	Turns store under the `n8n::<session>` transcript scope — separate history, same shared user memories as every other MemorySync surface

Workflow operations

Operation	Use it for
Add Memory	“Remember this fact” — server-side extraction keeps only durable content
Add Conversation Turn	Verbatim transcript capture with duplicate-proof retries
Search Memories	Semantic lookup — results fan out as n8n items
Recall Context	A grouped, prompt-ready context block for a downstream AI step (no competitor node has this)
Get Many	Newest-first listing for an end user
Delete Memories	Explicit deletion by ID (`m_123`)

curl --request POST https://api.memorysync.io/v1/memory/add_turn \
  --header "X-API-Key: $MEMORYSYNC_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{"tenant_id":"acme","user_id":"customer-42","source":"n8n","text":"human: I always take the window seat","speaker":"human@n8n::support#h<content-hash>","metadata":{"session_id":"n8n::support"}}'

Quota behaviour is explicit: free-tier exhaustion reports stored: false, accepted: true instead of failing the workflow, and evaluation keys surface a truthful 429. Bad metadata JSON and malformed memory IDs raise clear node errors before any network call.

Supported versions

Surface	Requires	Verified on
`n8n-nodes-memorysync` 1.0.0	Self-hosted n8n with community nodes enabled; the Chat Memory sub-node needs an n8n release that ships `@n8n/ai-node-sdk` (current releases do)	23 CI checks: the registry contract (package keyword, dist wiring, zero runtime deps), the fnv1a64 idempotency parity vector, and the memory node driven through the REAL @n8n/ai-node-sdk → LangChain adapter chain — windowed history, session isolation, both metadata envelopes (nested production + legacy flat), duplicate-proof retries, fail-open outages, no-op clear — plus a live production round-trip

Where to go next

Zapier

The same memory in 8,000-app automations.

LangChain Memory

The adapter family the memory node builds on.

Memory Tools

Every operation, argument by argument.

Was this page helpful?

← Previous

ElevenLabs

Zapier