If your Zep-based n8n workflow stopped working, you are not alone. Hundreds of n8n users ran into the same wall after Zep shut down v1 and relaunched as a hosted cloud service with a completely different API.
This post explains exactly what happened, why your workflow broke, and three alternatives that actually work today.
What Happened to Zep v1
Zep launched as an open-source, self-hostable memory server. It had a simple REST API — you would POST conversation turns and GET back relevant context. n8n users integrated it via HTTP Request nodes and it worked well.
In late 2024, Zep deprecated v1 and launched Zep Cloud — a fully managed service with a new authentication model, new endpoints, and a new pricing structure. The v1 self-hosted option reached end-of-life.
This broke every existing n8n workflow that relied on Zep v1 in three specific ways:
1. API Endpoint Changes
Zep v1 used paths like /api/v1/sessions/{session_id}/memory. Zep Cloud uses a completely different URL structure under api.getzep.com. Every HTTP Request node pointed at the old endpoint started returning 404.
2. Authentication Model Change
Zep v1 used a simple z-api-key header. Zep Cloud requires a JWT token obtained from a separate auth flow. Any workflow with a hardcoded API key stopped authenticating.
3. Data Loss
If you were self-hosting Zep v1, your memory data lived on your own server. Migrating to Zep Cloud required re-importing all stored memories — there was no automatic migration path.
Zep Cloud's free tier has strict rate limits and requires credit card verification. The pricing model changed significantly from v1, which was free to self-host.
3 Working Alternatives
Option 1: retainr (Recommended — 30 Seconds to Set Up)
retainr was built specifically for n8n. Install the community node, add your API key, and you have persistent semantic memory without touching the HTTP Request node.
Settings → Community Nodes → Install → n8n-nodes-retainr
The difference from Zep: retainr's API never breaks between major versions. The /v1/memories endpoint is immutable — any workflow you build today will still work in two years.
Two nodes replace all your Zep HTTP calls:
- Store Memory: save context after each AI response
- Search Memory: retrieve relevant context before the next AI call
Namespace per user (e.g., user:alice) to scope memory — no schema design required.
Option 2: n8n Postgres Chat Memory Node (Built-in, Requires DB)
n8n ships a Postgres Chat Memory node in the AI Agent module. It stores conversation history in a Postgres table you manage.
Setup:
- Spin up a Postgres instance (Supabase free tier works)
- Connect n8n to it via the Postgres credential
- Drop the Postgres Chat Memory node into your AI Agent
Limitations:
- Stores full conversation history, not semantic search
- Returns last N messages, not most relevant context
- You manage the database schema, backups, and scaling
- No TTL expiry on old memories
Use this if you already run Postgres and want zero external dependencies.
Option 3: Manual HTTP Memory Pattern (No Dependencies)
If you cannot install community nodes (n8n Cloud restricts this on lower plans), you can replicate basic memory with n8n's built-in nodes.
Pattern:
- Use a Code node to build a simple key→value store in the workflow's Static Data
- Serialize the last N conversation turns as JSON
- Inject the JSON into your AI prompt
Honest assessment: Static Data is global — it is not scoped per user. This only works for single-user automations. For multi-user workflows, you need an external store.
If you are on n8n Cloud and cannot install community nodes, use retainr via the HTTP Request node directly. Two calls: GET /v1/memories/search before the AI step, POST /v1/memories after. No community node required.
Migrating Your Zep Workflow to retainr
If you had a working Zep v1 workflow, here is the 1:1 replacement map:
| Old Zep call | retainr equivalent |
|---|---|
POST /api/v1/sessions/{id}/memory | POST /v1/memories with namespace: session:{id} |
GET /api/v1/sessions/{id}/memory | GET /v1/memories/search?namespace=session:{id}&q={query} |
DELETE /api/v1/sessions/{id}/memory | DELETE /v1/memories?namespace=session:{id} |
The data model maps cleanly. Zep stored messages arrays; retainr stores content strings. Before importing, join your Zep message arrays into a single string per turn:
role + ": " + content
Then POST each turn to retainr. The semantic index rebuilds automatically.
Why retainr Does Not Break
Zep broke because they pivoted their product from open-source to SaaS. The API was rebuilt from scratch for the new business model.
retainr is an API-first product from day one. The rule in our codebase: existing fields on /v1/memories are never removed or renamed. We add fields; we never break them. When we need a breaking change, we increment the version (/v2/) and keep v1 running.
You can read the commit history — this rule has held since the first commit.
Give your AI agents a real memory
Free plan includes 1,000 memory operations/month. No credit card required.
Replace Zep in 30 seconds — free plan available →Frequently Asked Questions
Is Zep v1 completely dead? Yes. The v1 GitHub repository is archived and the self-hosted distribution is no longer maintained. Security patches will not be backported.
Can I still use Zep Cloud with n8n? Yes, but you need to rebuild your HTTP Request nodes with the new JWT auth flow and new endpoint structure. There is no native n8n node for Zep Cloud.
Does retainr support the same session-scoped memory as Zep? Yes. Use namespace: session:{session_id} to scope memories per conversation session, or namespace: user:{user_id} to scope per user across sessions.
What about Zep's graph memory feature? Zep Cloud added a knowledge graph layer in v2. retainr uses flat semantic search (pgvector). If you need graph-structured memory with entity extraction, Zep Cloud is worth evaluating — but for most n8n automation use cases, flat semantic search retrieves relevant context with less complexity.
Add memory to any n8n workflow
Working with a specific app? These guides cover the exact setup:
- n8n WhatsApp AI Memory — Persistent memory for WhatsApp bots
- n8n Telegram AI Memory — Long-term memory for Telegram bots
- n8n Slack AI Memory — Persistent memory for Slack bots
- n8n Discord AI Memory — Persistent memory for Discord bots
- n8n Gmail AI Memory — Per-sender memory for Gmail AI workflows
- n8n HubSpot AI Memory — CRM-aware AI memory for HubSpot
- n8n Zendesk AI Memory — AI memory for support ticket workflows
- n8n Airtable AI Memory — Persistent memory for Airtable workflows