Dify + MemorySync

Overview

Dify’s native memory is per-conversation: the LLM-node memory toggle and conversation variables reset with every new thread, and the knowledge base is static documents — nothing remembers the USER across conversations or apps. The MemorySync plugin adds exactly that as four curated tools. Recall Context returns a prompt-ready memory block to wire into any LLM node; Remember stores facts with deterministic idempotency seeds (re-running a node never duplicates); Search Memories returns scored JSON with tolerant parsing and a hard result cap; Forget Memory deletes exactly one memory, loudly — and there is deliberately no delete-everything tool (a community Mem0 fork exposes delete_all_memories to the LLM: a one-call account wipe). User scoping resolves automatically — parameter, then Dify’s own runtime user, then a deterministic default — where the community Mem0 plugin takes user_id as a free-form parameter and silently mixes every end user when it isn’t wired. Calls run under a 10s budget with structured, branchable error JSON instead of 30-second node hangs and opaque error blobs.

Setup status and requirements

Supported version

memorysync plugin 0.0.1 (.difypkg)

Last setup review

2026-08-24

Permissions

A MemorySync API key with read and write scopes.

Limits

Requires Dify 1.14+ (plugin runner Python 3.12). Install from local package file; marketplace listing pending review.

Capabilities

• Recall Context: prompt-ready memory block for any LLM node
• User scoping auto-resolved from Dify’s runtime — zero wiring required
• Conversation-scoped metadata (dify::<conversation_id>)
• Idempotent Remember — node re-runs converge on one row
• Scored Search with tolerant parsing and a ≤25 result cap
• Loud single-id Forget; deliberately NO delete-everything tool
• 10s budget + structured branchable errors — no 30s node hangs
• Cross-surface: memories flow to and from every MemorySync adapter

Quick Start

Chatflow wiring:

[User Input]
     │
[Recall Context]   ← MemorySync tool node
     │  {{recall_context.text}}
[LLM Node]         ← memory block in the prompt
     │
[Remember]         ← store what the user shared
     │
[Answer]

Use Cases

Explore More