SQLite Vector Embedding Database Plugin (UE5)

A lightweight SQLite plugin for Unreal Engine that bundles and requires the sqlite-vec extension for vector similarity search.
It’s a code-only plugin (no Content assets) for running SQL and K-NN vector search inside UE (AI memory / embedding retrieval, etc.).

• Publisher: haozena
• Supported UE: 5.6 (Win64)
• Module: SQLiteVecPlugin
• Third-party: SQLiteCore (core, required), SQLiteVec (required; integrated with the plugin and loaded dynamically)
• Content: "CanContainContent": false

---

1) Installation

1. Close Unreal Editor.
2. Copy the entire SQLiteVecPlugin/ folder to:
   • Recommended: [YourProject]/Plugins/, or
   • Engine/Plugins/ (engine-wide install).
3. Verify required third-party files are included with the plugin:
   • Source/ThirdParty/Include/ → sqlite3.h, sqlite3ext.h, sqlite-vec.h
   • Source/ThirdParty/Bin/ → SQLiteCore.dll, SQLiteVec.dll and their LICENSE files
4. Launch UE → Edit > Plugins → enable SQLite Vector Embedding Database Plugin.
5. On first package/build/run, the Build.cs copies the needed DLLs to the runtime folder (usually YourProject/Binaries/Win64).

---

2) Quick Start (Blueprint workflow)

The plugin auto-registers/loads the sqlite-vec extension in StartupModule. All nodes below are provided by this plugin.

Step A: Open a database

• OpenDatabase(FilePath) → DBHandle (int32)
  Example: Saved/SQLite/demo.db (created automatically if missing).

Step B: Create a vector table (Create SQLite Vec Table if not exists)

• Call CreateVecTable(DBHandle, OutError, TableName="items", Dim=1536).
  This creates a vec0 virtual table; the vector column name is embedding (handled internally).

Step C: Insert a vector (Insert Vec Row)

• Call InsertVecRow(DBHandle, TableName, RowId, VecData(float[]), OutError).
  • The vector length must match the Dim used at table creation; use PadFloatArray(InArray, TargetLength=Dim, FillValue=0.0) to pad/truncate if needed.

Step D: Similarity search (Vector Query (K-NN))

• Call VectorQuery(DBHandle, TableName, QueryVec(float[]), K, OutResults, OutError).
  • OutResults : TArray<FVectorSearchResult> contains Id and Distance.

Step E: Close the database

• CloseDatabase(DBHandle)

For regular SQL, use ExecuteSQL / QuerySQL.

---

3) C++ example (equivalent to the Blueprint flow)

#include "SQLiteVecBPLibrary.h"

int32 Handle = USQLiteVecBPLibrary::OpenDatabase(TEXT("Saved/SQLite/demo.db"));

FString Err;
bool Ok = false;

// A. Create vector table (note param order: DBHandle, OutError, TableName, Dim)
Ok = USQLiteVecBPLibrary::CreateVecTable(
    Handle, Err, TEXT("items"), 1536);

// B. Insert a vector row
TArray<float> Vec; Vec.SetNum(1536);
// ... fill Vec ...
Ok = USQLiteVecBPLibrary::InsertVecRow(Handle, TEXT("items"), /*RowId=*/1, Vec, Err);

// C. K-NN query
TArray<float> Query; Query.SetNum(1536);
// ... fill Query ...
TArray<FVectorSearchResult> Results;
Ok = USQLiteVecBPLibrary::VectorQuery(Handle, TEXT("items"), Query, /*K=*/5, Results, Err);

// Close database
USQLiteVecBPLibrary::CloseDatabase(Handle);

---

© 2025 haozena. All Rights Reserved.