n8n shipped two patches this week -- 2.39.5 on September 14 and 2.39.6 on September 16 -- fixing credential revocation bugs, Gateway credits logic, and flagging two configurations that will break in v3. CrewAI landed 1.15.22 on September 16 with human feedback tracing, an llm_overlay runtime variable, and OpenRouter embeddings support.
I scan agent framework changelogs weekly because production outages trace back to patches you skipped. Here's what actually shipped, what it means, and what to check in your own setup before the next release drops.
What did n8n 2.39.5 and 2.39.6 change?
n8n released two patches this week: 2.39.5 on September 14 and 2.39.6 on September 16. Between them, the changes cover credential revocation, log streaming, Gateway credits eligibility, and two deprecation warnings that preview what will break in v3. Neither patch introduces new features -- these are maintenance releases that patch holes and set up the v3 migration path.
Version 2.39.5 fixed a bug in the editor where some users couldn't revoke end-user credentials. If your n8n setup uses OAuth or per-user credentials, and revocation was silently failing, this is the fix. The same release added log streaming events for instance reports -- useful if you're piping n8n logs into a centralized observability stack.
Version 2.39.6 is the more important one for self-hosted operators. It fixes Gateway credits eligibility for n8n Cloud users -- a sub-node was checking credits against the wrong parent node -- and adds two deprecation warnings:
- Storage directory rename: n8n is renaming its storage directory in v3. If you've mounted Docker volumes or written backup scripts targeting the old path, you'll see a warning in 2.39.6 logs. Update those paths before v3 drops.
- N8N_DB_PING_TIMEOUT deprecated: This env var controlled database connection ping timeout. It's flagged for removal in v3. The warning is non-blocking today, but plan to remove it from your environment config.
n8n has been on a roughly 2-day patch cadence through September -- versions 2.39.0 through 2.39.6 shipped across nine days. The project sits at over 133,000 GitHub stars and has been in production automation stacks since 2019. If you're self-hosted, audit your configuration against those two deprecated items now.
What did CrewAI 1.15.22 change?
CrewAI 1.15.22, published September 16, ships five production-facing improvements: connection identifier aliases, deployment failure reason recording, automatic human feedback and pause event tracing, a new llm_overlay context variable, and OpenRouter as a supported embeddings provider. None are headline features -- they're the kind of release that removes debugging friction and adds architectural flexibility that wasn't there before.
Aliases as connection identifiers let you reference the same data source or tool by multiple names. In larger crews where the same connection gets referenced across agents using different identifiers, you can define an alias and use it consistently -- no more mismatches across agent definition files.
Deployment failure reason recording closes a debugging gap. When a crew deployment fails, the error reason now lands directly in the trace. Before 1.15.22, you had to reconstruct what went wrong by digging through logs. Now it surfaces cleanly in your observability tooling without extra instrumentation.
Human feedback and pause event tracing is the most practically useful change for HITL workflows. If your crew pauses for human review or a human provides feedback mid-run, those events are now captured in the trace automatically. No manual instrumentation needed. The trace gives you a complete record of when the human was in the loop, what they saw, and what they decided.
llm_overlay is the architectural addition worth tracking. It introduces a context variable for passing model-level overrides at runtime -- different model IDs, temperature values, or other parameters -- without touching your crew definition. For teams running the same crew against multiple clients or cost tiers, this makes multi-tenant agent systems practical. Define the crew once, parameterize the model layer per run.
OpenRouter embeddings extends an existing integration. CrewAI already supported OpenRouter for LLM completions; 1.15.22 extends that to embeddings. If your crew does RAG or semantic search and you route through OpenRouter's unified API instead of hitting providers directly, you can now do that for the embedding step too.
Get the AI Agent Briefing
One email per week. The best AI agent news, tutorials, and tools -- written by someone who actually builds with them.
Subscribe Free
Should you use n8n or CrewAI for your next build?
n8n and CrewAI solve different problems, and that distinction determines which one to reach for. n8n is workflow automation where AI is one node in a broader pipeline -- best when you need to connect SaaS apps, move data between systems, and trigger actions from external events. CrewAI is a Python framework for multi-agent orchestration -- best when the work itself requires multiple agents with distinct roles collaborating toward a shared goal.
The practical split I use: if a workflow needs to pull from an API, transform data, call an LLM, then push results to a database, that's n8n. The AI step is one node, and the surrounding infrastructure is what n8n handles well. If you need a researcher agent pulling source material, a writer agent drafting from it, and a critic agent evaluating the output before it ships, that's CrewAI. The agent collaboration is the point, not a step inside a larger process.
Some teams run both in production. n8n handles orchestration -- triggers, scheduling, SaaS integrations -- and fires a CrewAI crew via webhook when a task needs multi-agent reasoning. n8n has 133,000+ GitHub stars and a visual editor non-developers can use. CrewAI has 30,000+ GitHub stars and a Python-first API agent-product developers tend to prefer. Neither replaces the other at its core use case.
What to check in your setup right now
For n8n self-hosted operators, two action items from this week's releases. First, grep your environment config for N8N_DB_PING_TIMEOUT and remove it -- or mark it for removal before v3. Second, check Docker volume mounts and backup scripts for the old storage directory path. The deprecation warning tells you the rename is coming; update those paths before they become errors in v3.
For n8n Cloud users, the Gateway credits fix is already live in 2.39.6. If you were seeing credit discrepancies on sub-node operations, that was the bug. No action needed on your end.
For CrewAI teams, 1.15.22 is a safe patch-level upgrade. The llm_overlay variable is opt-in -- you don't have to use it unless you're building multi-tenant systems. The tracing improvements are automatic once you upgrade: human feedback and pause events start appearing in traces without any code changes on your end.
What these releases don't answer yet
Neither release addresses a question I keep seeing in self-hosted n8n communities: the v3 timeline. The deprecation warnings are live, but the n8n team hasn't announced a v3 release date. If you're planning infrastructure work, check the n8n GitHub milestones for v3 before scoping your cleanup sprint.
On the CrewAI side, llm_overlay is new and documentation coverage is thin as of 1.15.22. The release notes describe the variable, but documentation on how it interacts with memory, tool calls, and task delegation at runtime is sparse. If you're building multi-tenant systems that depend on llm_overlay, test it in a staging environment before relying on it in production. Don't assume it threads cleanly through every part of the agent execution path without verifying first.
FAQ
What is n8n 2.39.6?
n8n 2.39.6 is a patch release published September 16, 2026. It fixes Gateway credits eligibility for n8n Cloud users and adds deprecation warnings for two configurations that will become breaking changes in v3: the storage directory path and the N8N_DB_PING_TIMEOUT environment variable. Self-hosted users should audit their configurations against both items before v3 ships.
What is CrewAI 1.15.22?
CrewAI 1.15.22 is a patch release published September 16, 2026. Key additions include automatic tracing for human feedback and pause events in HITL workflows, an llm_overlay context variable for runtime model overrides, connection identifier aliases, deployment failure reason recording in traces, and OpenRouter as a supported embeddings provider.
What does llm_overlay do in CrewAI?
The llm_overlay context variable, introduced in CrewAI 1.15.22, lets you pass model-level parameters at runtime without modifying your crew definition. You can specify different model IDs, temperature values, or other parameters per run. This makes multi-tenant setups practical: one crew definition, different model configurations per client or cost tier.
Is n8n 2.39.6 a breaking change?
No. n8n 2.39.6 adds deprecation warnings, not breaking changes. The flagged items -- the storage directory rename and N8N_DB_PING_TIMEOUT -- will become breaking in the upcoming v3 release, but they work normally in the 2.39.x series. The warnings give you runway to update your configuration before v3 drops.
Can you use n8n and CrewAI together?
Yes, and some teams do. n8n handles workflow orchestration -- triggers, data movement, scheduling, SaaS integrations -- and calls a CrewAI crew via webhook or API when a task requires multi-agent reasoning. The tools complement each other: n8n provides the surrounding process infrastructure, CrewAI handles the agent collaboration layer inside it.
Get the AI Agent Briefing
One email per week. The best AI agent news, tutorials, and tools -- written by someone who actually builds with them.
Subscribe Free