Skip to main content

Guide: Add Dify Memory with Hindsight

· 6 min read
Ben Bartholomew
Ben Bartholomew
Hindsight Team

Guide: Add Dify Memory with Hindsight

If you want Dify memory with Hindsight, the cleanest setup is the Hindsight Dify plugin. It adds three tools — Retain, Recall, and Reflect — that drop into any Dify workflow alongside your other LLM, search, and tool nodes. That gives Dify long-term memory across runs instead of forcing every workflow execution to start stateless.

This is a good fit for Dify because Dify is a visual workflow builder with a growing tool and plugin ecosystem, but its workflows are stateless across runs by default. The plugin uses the same node model as everything else in Dify: you place a Retain node to store content, a Recall node to pull relevant context before an LLM step, or a Reflect node to ask a synthesizing question over a bank.

This guide walks through installing the plugin, adding your Hindsight credentials, wiring the three tools into a workflow, and a quick verification flow so you can confirm memory is actually being used. Keep the docs home and the quickstart guide nearby while you work.

Quick answer

  1. In Dify, go to Plugins → Install Plugin and install the Hindsight plugin.
  2. Open the Hindsight plugin and add your API URL and API Key credentials.
  3. Add a Recall node before an LLM step to pull relevant context.
  4. Add a Retain node to store new content into a bank.
  5. Verify that a later run recalls what an earlier one stored.

Prerequisites​

Before you start, make sure you have:

  • A Dify instance where you can install plugins and build workflows
  • A reachable Hindsight backend, Hindsight Cloud or a self-hosted server
  • A Hindsight API key from the dashboard (optional for self-hosted unauthenticated instances)

Step 1: Install the plugin​

In your Dify dashboard go to Plugins → Install Plugin, then choose one of:

  • Marketplace — search for Hindsight (once published)
  • GitHub — install from vectorize-io/hindsight (path hindsight-integrations/dify)
  • Local — upload the .difypkg archive

After install, the Hindsight plugin appears under Tools in the workflow editor.

Step 2: Add your Hindsight credentials​

Open the Hindsight plugin and add credentials:

  • API URL — defaults to https://api.hindsight.vectorize.io (Cloud); change it for a self-hosted server
  • API Key — your hsk_... key (optional for self-hosted unauthenticated instances)

If you do not have a key yet, sign up for Hindsight Cloud on the free tier and grab one from the dashboard, or self-host a server.

Step 3: Wire the tools into a workflow​

The plugin gives you three tools you place like any other Dify node.

  • Retain — store free-text content in a bank. Fields are Bank ID (the memory bank to store in, auto-created on first use), Content (the free-text to retain), and optional Tags (comma-separated). Hindsight extracts facts asynchronously after the call returns.
  • Recall — search a bank for memories relevant to a query. Fields are Bank ID, Query (natural language), Budget (low / mid / high), Max Tokens (cap on returned tokens), and optional Tags. It returns a results array.
  • Reflect — get an LLM-synthesized answer over the bank. Fields are Bank ID, Query (the question to answer), and Budget (low / mid / high). It returns a text field.

A typical shape is a Recall node before an LLM step to surface prior history, then a Retain node after the run to store the new content back into the same bank.

For lower-level behavior, read Hindsight's recall API and Hindsight's retain API.

What you get​

Because the tools are ordinary Dify nodes, you can drop memory into chatflows, workflows, and agent apps without leaving the visual builder:

  • Customer-support assistant — every closed ticket retains the resolution. Every new ticket starts with a Recall against the bank to surface similar past issues, then passes that context to your LLM node to draft the first reply.
  • Sales-call coach — a call summary is retained after each call. Before the next prep call, a Recall on the prospect's name pulls every prior touchpoint into the daily prep doc.
  • Knowledge-base agent — uploaded documents are retained, and the chatflow uses Recall instead of vector-DB-only retrieval to get fact-extracted, deduplicated, time-aware results.

Verify that memory is working​

A good test sequence is:

  1. run a workflow that ends with a Retain node writing to a bank
  2. store a distinctive fact — for example a decision or a customer detail
  3. run a second workflow with a Recall node reading the same Bank ID
  4. query for the earlier fact
  5. confirm the earlier fact appears in the results array

If the Recall node surfaces what the earlier run stored, the setup is working.

Common mistakes​

Mismatched Bank IDs​

Recall and Retain only share memory when they use the same Bank ID. If a later Recall returns nothing, check that it points at the bank the earlier Retain wrote to.

Testing Recall immediately after Retain​

Retain extracts facts asynchronously after the call returns. If you Recall the instant a Retain finishes, the facts may not be searchable yet.

Missing or wrong credentials​

If the tools error, confirm the API URL points at your backend and the API Key is set (required for Hindsight Cloud).

Treating Reflect like Recall​

Recall returns a results array of memories; Reflect returns a synthesized text answer. Use Recall to feed an LLM step and Reflect when you want the synthesized answer directly.

FAQ​

Do I need Hindsight Cloud?​

No. A self-hosted Hindsight server works too — point the plugin's API URL at it, and the API key is optional for unauthenticated instances.

Where does the plugin appear in Dify?​

After install, the Hindsight plugin shows up under Tools in the workflow editor.

How is memory scoped?​

By Bank ID. Each bank is an isolated memory store, and you choose which bank each tool node reads from or writes to.

Which tool should I use?​

Use Retain to store content, Recall to pull relevant memories before an LLM step, and Reflect to get an LLM-synthesized answer over a bank.

Next Steps​