Vector Store Question Answer Tool generates invalid schema in v2.6.3 - breaks OpenAI and Anthropic

[AlphaBeez]

Bug Description

The bug:
After upgrading from n8n v2.4.7 to v2.6.3, the Vector Store Question Answer Tool (toolVectorStore v1.1)
generates invalid tool schemas that both OpenAI and Anthropic APIs reject.

Error with OpenAI:
Invalid schema for function 'protocol_knowledge': schema must be a JSON Schema of 'type: "object"', got 'type:
"None"'.

Error with Anthropic:
tools.0.custom.input_schema.type: Field required

To Reproduce

To Reproduce:

1. Create workflow with: Chat Trigger → AI Agent (v3.1) → OpenAI/Anthropic Chat Model
2. Add Vector Store Question Answer Tool (v1.1) connected to AI Agent
3. Connect PGVector Store (v1.3) + Embeddings + Chat Model to the Vector Store Tool
4. Activate and send a message that triggers the tool

Expected behavior

Expected behavior:
Tool should generate valid JSON schema with type: "object" for function calling.

Workaround:
Downgrade to n8n v2.4.7 where it worked.

Environment:

• n8n v2.6.3 (npm, self-hosted)
• PostgreSQL
• Node versions: toolVectorStore 1.1, agent 3.1, vectorStorePGVector 1.3

Debug Info

Debug info

core

• n8nVersion: 2.6.3
• platform: npm
• nodeJsVersion: 24.13.0
• nodeEnv: production
• database: postgres
• executionMode: regular
• concurrency: -1
• license: enterprise (production)
• consumerId: 70d2c345-ccff-4fc0-9ea2-0552f008b06a

storage

• success: all
• error: all
• progress: false
• manual: true
• binaryMode: filesystem

pruning

• enabled: true
• maxAge: 720 hours
• maxCount: 10000 executions

client

• userAgent: mozilla/5.0 (macintosh; intel mac os x 10_15_7) applewebkit/605.1.15 (khtml, like gecko) version/26.2 safari/605.1.15
• isTouchDevice: false

Generated at: 2026-02-04T04:32:32.663Z

Operating System

Ubuntu Linux 6.8.0-86-generic

n8n Version

2.6.3

Node.js Version

v24.13.0

Database

PostgreSQL

Execution mode

main (default)

Hosting

self hosted

[n8n-assistant]

Hey @AlphaBeez,

Thank you for reaching out! We’ve received your issue and are looking into it. To help us track this internally, we’ve created a Linear ticket with the reference: "GHC-6655".

We’ll keep you updated as we make progress, but if you have any additional details or context to share, feel free to add them here - it’s always helpful!

Thanks for your patience and for bringing this to our attention.

---

For INTERNAL use only: GHC-6655

[AlphaBeez]

Still broken on v2.6.4 (tested today, Feb 8 2026).

This affects all LLM providers — both OpenAI and Anthropic reject the malformed schema. The toolVectorStore v1.1 generates input_schema without the required "type": "object" field because VectorStoreQATool extends LangChain's base Tool class (simple string input) rather than DynamicStructuredTool.

Root cause is in packages/@n8n/nodes-langchain/nodes/tools/ToolVectorStore/ToolVectorStore.node.ts — the VectorStoreQATool from @langchain/classic/tools doesn't provide a proper JSON schema. The fix would be either:

1. Wrapping the tool in a DynamicStructuredTool with an explicit z.object({ input: z.string() }) schema
2. Or patching the schema serialization to ensure type: "object" is always present

This is a complete blocker for anyone using Vector Store tools with AI Agents — no workaround exists within n8n's built-in nodes. The only option is downgrading to v2.4.7.

[AlphaBeez]

Update: Applied a local patch on our npm install that confirms the fix.

Correction to my earlier analysis: I initially reported that Postgres Tool nodes with descriptionType: "manual" were also affected. After patching and testing, this was a misdiagnosis — it was always the Vector Store Question Answer Tool causing the error. Postgres Tools with manual descriptions work fine once the Vector Store schema is corrected.

The patch (one-file fix):

In packages/@n8n/nodes-langchain/nodes/tools/ToolVectorStore/ToolVectorStore.node.ts, after creating the VectorStoreQATool, wrap it in a DynamicStructuredTool:

import { DynamicStructuredTool } from '@langchain/core/tools';
import { z } from 'zod';

// After creating vectorStoreTool and setting chain...
const wrappedTool = new DynamicStructuredTool({
    name,
    description,
    schema: z.object({
        input: z.string().describe("A fully formed question")
    }),
    func: async (toolInput) => vectorStoreTool._call(toolInput.input),
});
return wrappedTool;

This produces a proper JSON schema with "type": "object" that both Anthropic and OpenAI accept. All existing RAG logic (vector store, embeddings, QA chain with topK) stays identical.

Tested and verified on n8n v2.6.4 — all workflows with Vector Store + AI Agent working correctly.