Oceabase Releases TedDB: Open Source AI Native Hybrid Search Database for Multi-Fodel Rag and AI Agents

AI applications rarely deal with a single clean table. They mix user profiles, chat logs, JSON Metadata, embeds, and sometimes location data. Most teams are responding to this with a patchwork of OLTP databases, vector stores, and search engines. A ship release looking forOpen source AI GETABASASE database (under Apache 2.0 license). IntudDB is described as a traditional AI database that combines relational data, vector data, text, json, and GIS in one engine and exposes hybrid visits and data operations.

What is PUSTYB?

looking for it is planned as a lightweight, embedded version of the ocean base engine, aimed at an AI application rather than a general purpose distributed deployment. It works as a single node database, supports embedded mode and client or server mode, and is always compatible with MySQL and SQL Syntax drivers.

In the Matrix of Power, IntudDB is tagged as:

• Embedded database is supported
• Standalone database supported
• Distributed layout is not supported

While a full seafood product includes a distributed case.

From a data model perspective, Wanted accessories:

• Details related to standard SQL
• Vector search
• Full Text Search
• JSON data
• Spatial GIS data

All within a single storage and indexing layer.

Hybrid search as a key feature

The main feature of OceanBase SUPS is Hybrid Search. This combinatorial vector search is based on semantic refueval, full test keyword Retroueval, and scalar filters in one query and one ranking step.

Hybrid search is implemented using a system package called dbms_hybrid_search with two entry points:

• Dbms_hybrid_search.search returns the results as Json, sorted in parallel
• Dbms_hybrid_search.get_sql which returns the SQL concrete string used for execution

A hybrid search method can run:

• Pure vector search
• Full Full Search
• Hybrid search

It can also push filters related to storage joins. It also supports reranking techniques such as weighted scores and regression fusion and can connect to a large language model based on RAND RANDERS.

By bringing back Augmented Generation Generation (Rag) and Agent memory, this means you can write a single SQL query that performs sql matching on Emeddings, proper nouns or proper nouns.

Vector and full text details

At its core, he wants to reveal a Today's Vector and a stack full of text.

For vectors, you want:

• Supports dense vectors and sparse vectors
• It supports Manhattan, euclidean, inner product, and cosine distance
• It provides memory index types like HNSW, HNSW sq, HNSW BQ
• It offers various types of Disk based including IVF and IVF PQ

Hybrid Vector Index Show how to store the raw text, let IntdDB call the embedding model automatically, and then the system stores the corresponding vector index without a separate pipeline to get a different power.

For text, Ufuna offers full text search with:

• keyword, phrase, and boolean queries
• BM25 Compatibility position
• multiple tokenizer methods

The key point is that the full content of the text and veterses is the first class and is combined in the same query editor with scalar indexes and GIS indexes, so the hybrid search does not need an external input.

AI works within data

looking for It includes built in ai function expressions that allow you to call models directly from SQL, without a separate application service to listen to every call. The main functions are:

• AI_EDBED to convert text to embed
• The AI
• AI_Rerank reranks the list of students
  AI_PROMPTT to combine prompt templates and prompt values ​​into an AI_COMPLETE JSON object

Model Metadata and Endpoints are managed by the DBMS_I_SERVICE package, which allows you to register external providers, and configure buttons, all on the data side.

Multimodal data and workloads

looking for It is designed to handle multiple data streams in one place. It has a multimodal data and Index layer that includes vectors, text, Json, and GIS, and a multi-compute layer for loading all hybrid, full text and scalar scenarios.

It also provides JSON indexes for Metadata queries and GIS indexes for geographic conditions. This allows questions like:

• Find similar documents
• Sort by Json Metadata by employer, region, or category
• Consoain by Spatial Range or Polygon

without leaving the same engine.

Because the sentdy is derived from the engine aceache, inheritance and acid transaction, column hybrid line, and execution of columns, although the maximum distribution is distributed, although the maximum distribution is always a function of the complete data of the sea.

Comparison table

Key acquisition

1. Traditional AI for hybrid search: sEEKDB combines vector search, full text search and relative filtering in SQL with one SQL Interface and DBMS_Hybrid_search, Agent Works can run multiple signal returns for one query instead of installing multiple engines.
2. Multimodal data for one engine: NEDDB stores and indexes relational data, vevectors, text, JSON and GIS in the same engine, allowing AI applications to highlight documents without storing separate information.
3. In Rag's Database AI functions: With AI_EI_EI
4. Single node, embedded friendly design: TentDB is a one-way, MySQL-compatible engine that supports embedded and standalone modes, while distributed, large-scale deployments still run CentDB for local workloads, edges and embedded AI services.
5. Open source and TOOLS ECOSYSTEM program: TentDB is open source under Apache 2.0 and integrates with a growing ecosystem of AI tools and frameworks, with Python support through PyseekDB and agencies, so it can serve as a unified data plane for AI applications.

Look It's a waste and The design. Feel free to take a look at ours GitHub page for tutorials, code and notebooks. Also, feel free to follow us Kind of stubborn and don't forget to join ours 100K + ML Subreddit and sign up Our newsletter. Wait! Do you telegraph? Now you can join us by telegraph.

AsifAzzaq is the CEO of MarktechPost Media Inc.. as a visionary entrepreneur and developer, Asifi is committed to harnessing the power of social intelligence for good. His latest effort is the launch of a media intelligence platform, MarktechPpost, which stands out for its deep understanding of machine learning and deep learning stories that are technically sound and easily understood by a wide audience. The platform sticks to more than two million monthly views, which shows its popularity among the audience.