Dify integration

Visual workflows. Invisible memory.

Dify lets you build AI applications from a visual canvas. Hebbrix adds the one thing Dify's built-in memory can't: memory that persists across users and sessions, and gets sharper on its own the more your workflow runs.

Setup guide →Get API key

where memory fits

Memory wraps every workflow node

Use the Hebbrix HTTP node before any LLM call to inject context, and after any LLM output to store what was learned. No code inside Dify, only HTTP requests.

User input

Message arrives

→

Hebbrix search

HTTP node · context retrieval

→

LLM node

GPT-4 with context

→

Hebbrix store

HTTP node · save response

→

Output

Response to user

Both Hebbrix nodes are standard HTTP request nodes in Dify. No plugins, no code blocks needed.

Dify workflow editor with Hebbrix search and store HTTP nodes around the LLM node

two api calls

One to retrieve. One to store. That's the integration.

Before the LLM node

http_search.json (Dify HTTP node)

{
  "method": "POST",
  "url": "https://api.hebbrix.com/v1/search",
  "headers": {
    "Authorization": "Bearer {{API_KEY}}"
  },
  "body": {
    "query": "{{user_input}}",
    "collection_id": "{{user_id}}",
    "limit": 5
  }
}
// Pass result into LLM system prompt
// as context_memories variable

After the LLM node

http_store.json (Dify HTTP node)

{
  "method": "POST",
  "url": "https://api.hebbrix.com/v1/memories",
  "headers": {
    "Authorization": "Bearer {{API_KEY}}"
  },
  "body": {
    "content": "{{user_input}} → {{llm_output}}",
    "collection_id": "{{user_id}}"
  }
}
// That's it. Memory grows automatically.
// RL quality checks run in the background.

worth knowing

Why not just use Dify's built-in memory?

Dify's memory is session-bound

Built-in conversation memory lives inside a single session. When a user starts a new chat, it's gone. Hebbrix memory persists indefinitely, across every session, every app version, every user interaction.

5-layer search, beyond recency

Dify retrieves recent messages. Hebbrix finds the most relevant memory no matter when it was stored, weighing semantic search, keyword matching, graph relationships, importance, and recency together.

Memory that learns and improves

After every workflow run, 6 RL quality checks evaluate whether the memories used actually helped produce a good response. The system gets smarter the more your workflow runs.

Add memory to your Dify workflows today

Two HTTP nodes. Ten minutes. Persistent memory for every workflow you've already built.

Start free →Dify guide