Description

n8n-nodes-milvus-db

A fork of n8n’s built-in Milvus Vector Store node that adds an explicit
Database selector. The collection dropdown is scoped to the selected database.

Upstream n8n has no database option — the connection always lands in Milvus’s
default db. The feature request was closed as not planned; a community PR has
been open and unmerged since 2025 (currently n8n-io/n8n#28082).

What’s different from the built-in node

| | Built-in vectorStoreMilvus | This node vectorStoreMilvusDb |
|—|—|—|
| Database | always default | dropdown, listed from listDatabases() |
| Collection | all collections in default | scoped to the selected database |
| Credential | milvusApi | milvusApi (the same one — no re-entry) |
| Modes | load / insert / retrieve / retrieve-as-tool | identical |

It reuses the existing milvusApi credential type, so your current Milvus
credentials appear in the dropdown untouched. Both nodes can coexist — existing
workflows on the built-in node keep working.

How the database actually gets applied

LangChain’s Milvus constructor spreads clientConfig first, then overrides
address / username / password from the top-level args. So database set
inside clientConfig survives and pins the gRPC connection to that db. Every
subsequent call (listCollections, search, insert) is scoped by it.

Build

npm install
npm run build     # -> dist/nodes/VectorStoreMilvusDb/

Local development

mkdir -p ~/.n8n/custom && cd ~/.n8n/custom && npm init -y
cd /path/to/this/repo && npm link
cd ~/.n8n/custom && npm link n8n-nodes-milvus-db
n8n start

Search the nodes panel for Milvus Vector Store (DB) — search the node
name, not the package name.

Deploy to Kubernetes / ACK

docker build --build-arg N8N_VERSION= 
  -t /n8n-milvus:0.1.0 .
docker push /n8n-milvus:0.1.0

Then point the deployment at the new image. Nothing else to configure —
N8NCUSTOMEXTENSIONS is baked in.

Two things that will bite you

1. PVC shadowing. If you mount a volume at /home/node/.n8n, anything you
copied to ~/.n8n/custom at build time is hidden by the mount and the node
silently won’t appear. That’s why this image uses /opt/n8n-milvus/custom,
outside any volume.

2. node_modules inside the custom directory. n8n’s CustomDirectoryLoader
globs */.node.js under each custom dir and loads every match — it does not
read package.json. Dependencies contain at least one file matching that
pattern (brotli-wasm/index.node.js). Keeping node_modules one level above
the custom dir sidesteps it while still resolving normally.

Optional: shrink the image

The bundled dependency tree is ~560 MB, mostly @langchain/community. The n8n
image already ships @langchain/community, @langchain/core,
@zilliz/milvus2-sdk-node and @n8n/ai-utilities. Confirm the path:

docker run --rm --entrypoint sh n8nio/n8n: -c 
  'ls -d /usr/local/lib/nodemodules/n8n/nodemodules/@n8n/ai-utilities'

If it’s there, you can drop the COPY --from=builder /build/node_modules line
and instead set:

ENV NODEPATH=/usr/local/lib/nodemodules/n8n/node_modules

This also removes the duplicate-LangChain-instance risk (two copies of
@langchain/core in one process). Test insert and retrieve before shipping —
if document objects cross the boundary and something does an instanceof check,
this is where it would surface.

Version pinning

@n8n/ai-utilities and n8n-workflow must match the versions your n8n runs.
Mismatched n8n-workflow copies produce confusing TypeScript structural errors
at build time. Check with:

docker run --rm --entrypoint sh n8nio/n8n: -c 
  'cat /usr/local/lib/nodemodules/n8n/nodemodules/@n8n/ai-utilities/package.json | grep version'

Currently pinned: @n8n/ai-utilities@0.24.3, n8n-workflow@2.31.3,
@langchain/community@1.1.27, @langchain/core@1.2.0.

Permissions

Listing databases requires the credential’s Milvus user to have database-level
read privileges. If it doesn’t, the dropdown falls back to offering default
only — you can still type a database name using the By Name mode.

Icon

milvusdb.svg is a generic placeholder. Swap in your own if you want the real
Milvus mark; check its trademark terms before redistributing it.