n8n-nodes-unify-llm

Use the full @atom8ai/unify-llm stack inside n8n with a LangChain-style experience:

• Multi-provider inference via connected AI model sub-nodes
• Structured output with JSON schema
• Guarded generation checks
• Runnable chain-style orchestration
• Ensemble router intelligence (BayesianUtilityRouter, ParetoNavigatorRouter, PrimRouter)
• Persistent vector retrieval with local cache, Qdrant, or Pinecone backends

Included

• Unify LLM node with resources:
  • Orchestration
  • Vector Store
• Provider credential nodes (for use by model sub-nodes):
  • Unify LLM OpenAI API
  • Unify LLM Anthropic API
  • Unify LLM Gemini API
  • Unify LLM Ollama API

Operations

Orchestration

• Generate
• Generate Structured (Schema JSON)
• Guarded Generate (advanced runtime options)
• Chain Generate
• Chain Structured
• Quickstart Ask
• Route Only
• Routed Generate

Vector Store

• Upsert Documents (persist vectors/documents into selected backend)
• Similarity Search (query against previously persisted vectors)

Supported vector backends:

• Persistent Local Cache (workflow static data namespace)
• Qdrant (REST API)
• Pinecone (REST API)

Build

• npm install
• npm run build
• npm pack

Publish to npm (community-node ready)

Before publishing:

• Ensure you are authenticated (npm whoami).
• Run quality checks (npm run lint, npm run build).
• Preview package contents (npm pack --dry-run) and confirm dist/nodes/UnifyLlm/unifyLlm.svg is included.

Publish:

• First release (unscoped): npm publish --access public
• Recommended CI release with provenance: npm publish --access public --provenance

After publishing:

• Verify package page: https://www.npmjs.com/package/n8n-nodes-unify-llm
• Submit for verification in n8n Creator Portal: https://creators.n8n.io/nodes

Notes

• Connect an AI model node to the Unify LLM Chat Model input (ai_languageModel).
• Provider credentials are configured on the connected model node, not on the Unify LLM root node.
• If multiple models are connected, router operations can use ordered model inputs.
• For vector workflows at scale, run Vector Store > Upsert Documents first and then Vector Store > Similarity Search.
• Persistent Local Cache stores vectors in workflow static data, while Qdrant and Pinecone use their respective remote indexes.