n8n-nodes-memorysync

MemorySync nodes for n8n — long-term memory for your workflows and a persistent brain for n8n's AI Agent.

Two nodes in one package:

| Node | What it does | | --- | --- | | MemorySync | Workflow operations: Add Memory, Add Conversation Turn, Search, Recall Context, Get Many, Delete. Marked usableAsTool, so an AI Agent can call it autonomously. | | MemorySync Chat Memory | An AI Agent memory sub-node (AiMemory output): the agent's conversation history persists in MemorySync across executions — sessions survive restarts, redeploys, and weeks between chats. |

Install (self-hosted n8n)

Settings → Community Nodes → Install → n8n-nodes-memorysync

(n8n Cloud surfaces only verified community nodes; verification submission is in progress.)

Credentials

Create a MemorySync API credential with your API key from app.memorysync.io → Settings → API Keys. The built-in credential test makes a real (tiny) query, so a bad key fails at setup time — not mid-workflow.

The AI Agent memory sub-node

Chat Trigger ──▶ AI Agent ◀──memory── MemorySync Chat Memory
                        ◀──tool────── MemorySync (Search / Add)

• Session ID (e.g. {{ $json.sessionId }}) scopes the transcript; the same session resumes its history on the next execution.
• End User ID isolates everything per user.
• Context Window Length counts turn pairs handed to the agent each run.
• Turns store verbatim with deterministic idempotency seeds — a retried execution can never duplicate a conversation.
• clear() is a deliberate no-op: an agent can never bulk-delete a customer's stored history. Deletion stays an explicit human action.
• A MemorySync outage degrades to "no history this run" — the agent still answers; nothing throws.

Requires n8n 2.16.0 or later for the Chat Memory sub-node — the first release that ships @n8n/ai-node-sdk, which it is built on (declared as n8n.aiNodeSdkVersion + a peer dependency, so n8n only offers the package to instances that can run it). On an older n8n the package still loads and the MemorySync workflow node works; only the sub-node reports "update n8n" when used. Built and tested against n8n-workflow 2.37.4 and @n8n/ai-node-sdk 0.27.4.

Operations cheat-sheet

Resource Memory → one of six operations:

| Operation | Use it for | | --- | --- | | Add Memory | "Remember this fact about the customer" (server-side extraction applies) | | Add Conversation Turn | Verbatim transcript capture with duplicate-proof retries | | Search Memories | Semantic lookup — results fan out as items | | Recall Context | A grouped, prompt-ready context block for a downstream AI step (unique to MemorySync) | | Get Many | Newest-first listing for a user | | Delete Memories | Explicit deletion by ID (m_123) |

Semantics worth knowing

• Every operation is scoped by End User ID — header-enforced isolation.
• Free-tier quota exhaustion is silent by design: adds answer accepted-without- storing, reads answer empty. The node reports stored: false, accepted: true instead of failing your workflow. Evaluation keys surface strict 429s.
• Transcript turns store under the n8n::<session> scope — separate history, same shared user memories as every other MemorySync surface.
• Node code never reads your credential. Every request goes through n8n's httpRequestWithAuthentication; the credential's own authenticate hook adds the base URL and the API key header.

Development

Built with @n8n/node-cli in strict (n8n Cloud-eligible) mode.

npm install
npm run lint    # n8n-node lint — n8n's community-node rules, default config
npm test        # n8n-node build + 31 contract/behaviour tests (node --test)
npm run dev     # n8n-node dev — local n8n with this package linked, hot reload

Version history

| Version | Changes | | --- | --- | | 1.0.9 | @n8n/ai-node-sdk is loaded when the Chat Memory sub-node runs, not when the package loads — on n8n releases older than 2.16.0 (no SDK) the MemorySync workflow node and the credential keep working and the sub-node reports a clear "update n8n" error, instead of the whole package failing to load. | | 1.0.8 | n8n verification review, round 2: @n8n/ai-node-sdk declared as a peer dependency with n8n.aiNodeSdkVersion; build/lint/dev via @n8n/node-cli (strict mode, default ESLint config); node code no longer reads credentials — requests use httpRequestWithAuthentication and the credential's authenticate hook supplies base URL + key; operations grouped under a Memory resource. Existing workflows keep working unchanged. | | 1.0.7 | Codex files use only documented keys and the supported Data & Storage category (review round 1). | | 1.0.6 | User-Agent version derived from package.json instead of a typed string. | | 1.0.5 | Codex placement updates. | | 1.0.4 | Distinct dark-mode icon variants; unconditional error wrapping. | | 1.0.3 | Verification-scanner findings addressed. | | 1.0.0 – 1.0.2 | Initial releases: MemorySync node (six operations, usableAsTool) and MemorySync Chat Memory sub-node. |

License

MIT