Community

Blog Events Webinars Tutorials Forum
  1. Blog
  2. Events
  3. Webinars
  4. Tutorials
  5. Forum
Create Account
×
Community Blog AgenticDB: Born from Alibaba Cloud AnalyticDB, Built for AI-Native Enterprises

AgenticDB: Born from Alibaba Cloud AnalyticDB, Built for AI-Native Enterprises

This article introduces AgenticDB, an AI-native data foundation built on AnalyticDB for managing AI agent context and backend services.

The AI paradigm shift has driven the rapid growth of AI-Native enterprises. These enterprises include AI startups, one-person companies, and innovation teams within traditional enterprises. They share the following characteristics and demands: First, they use AI Agents as a core capability to build new product moats. As Large Language Models (LLMs) become ubiquitous, AI context (including memory, knowledge, Agent Skills, etc.) becomes a valuable asset. Therefore, building an AI context foundation compatible with both Single-Agent and Multi-Agent architectures is critical. Second, they leverage AI to boost efficiency so humans can focus on business while AI achieves unmanned O&M. This requires the data foundation to Support context fencing for Agent sub-Job streams, automatic start and stop, and fallback guarantees when AI makes faults.

As Agent systems grow increasingly complex, the patch-based evolution of traditional data products has become inadequate. AgenticDB is a data foundation built on Alibaba Cloud AnalyticDB for PostgreSQL as its core, integrating AI application backend services and context Management. It helps enterprises quickly launch AI products to explore the marketplace, accumulate AI context to strengthen User stickiness and product competitiveness, accompany AI-Native enterprises through rapid growth, and ultimately stride into a new Future in the AI era.

1. Product Positioning

For AI-Native enterprises at different stages of development, AgenticDB leverages its High-Performance retrieval-augmented generation DPI engine and global data lake warehouse to build an enterprise context hub. On top of a unified underlying technology stack, it provides different Version Features, ensuring AI-Native enterprises can launch quickly while also providing migration Solutions for advanced stages.

Launch (Edition):

  • Build an AI Agent system based on one-stop backend services.
  • No need to worry about underlying implementation or node specifications. Quickly access AI context services and efficiently integrate AI Agents.

Advanced version:

  • Finer-grained context management and data sandbox protection ensure that Agents can write with confidence, prevent data leaks, and support compliance audit.
  • Provides extremely elastic dedicated resources. Launches instances in seconds when the Agent needs them, and automatically detects traffic to stop in seconds.

1

🔗 Data sandbox:https://www.alibabacloud.com/help/en/analyticdb/analyticdb-for-postgresql/data-sandbox-management

🔗 Context service: https://www.alibabacloud.com/help/en/analyticdb/analyticdb-for-postgresql/context-service

2. Common Scenarios

2.1 Vertical Domain AI Applications

Scenario description: Focus on vertical scenarios to achieve efficient intelligent interaction with low resource consumption through flexible deployment and cost control for AI scenario exploration.

Core challenges: Fast iteration requires rapid deployment and publishing. Unpredictable traffic peaks are encountered, while operational costs are extremely sensitive, demanding the best cost-effectiveness.

AgenticDB Solutions:

  • Application Setup:

    • Rapid Startup: Complete Version iterations in hours with a one-stop backend service and go published within half a day.
    • Traffic-Aware Dynamic Scale-out: Built-in AutoScaling detects traffic changes in real time and automatically scales out within seconds to handle peak loads, ensuring zero Latency for Users.
    • Scale-to-Zero for Maximum Cost Reduction: Automatically scales in to zero instances during no-traffic periods, significantly reducing idle resource fees and maximizing operational cost savings.

  • Context Service:

    • Deployment-Free Service: An out-of-the-box context service. Developers only need an API key to connect to the Agent.
    • Zero O&MCost: The fully managed infrastructure frees developers from database Management, allowing them to focus on product logic and commercialization.

2.2 Enterprise Agent Platform

Scenario: An enterprise agent platform is typically a complex system composed of multi-agents.

  • Context fencing: Massive sub-job streams use data sandboxes to achieve context environment fencing, preventing mutual interference. Create two fencing environments and run A/B Test comparisons in parallel in the fencing environments.
  • Context Shared: Accumulate common context (including memory, agent skills, and knowledge) and share it across agents.

Core JavaScript Challenge: In large-scale multi-agent systems, complex sub-job streams cause intertwined contexts, requiring both Strict fencing mechanisms and Shared Feature capabilities.

AgenticDB Solutions:

  • Context Management

    • Provides mature enterprise-level retrieval-augmented generation/GraphRAG and long-memory solutions. Supports hierarchical context management and efficient retrieval, significantly improving agent cognitive capabilities.

  • Data Sandbox:

    • Builds independent data fencing environments at the same storage cost using Copy-on-Write technology, providing an independent environment for each subtask stream to ensure parallel execution without interference.
    • Supports context sharing with authorization.
    • Support data travel for different time segments.

2.3 Vibe Coding

Scenario description:

  • Collaborative developer: Each session has an independent data branch to ensure fencing between environments. Create separate Dev and Prod environments to achieve developer and Test separation.
  • Safeguard against "database deletion management events": Prevent data corruption or contamination caused by AI in extreme cases from delivering a devastating blow to enterprises.

Core JavaScript Challenge: Each Vibe Coding application requires an independent fencing environment base, causing Cost to surge dramatically. Traditional databases lack version management capabilities comparable to Code.

AgenticDB Solution:

  • Branch Management:

    • Provides data branch management capabilities, making data management as flexible and efficient as Git manages code. You can create environments such as Dev, Test, and Prod for collaborative development.
    • In the event of data corruption or contamination, you can quickly reset or regenerate a new branch.

  • One-Stop Backend Service:

    • Provides one-stop backend service capabilities. Developers can use natural language to build application backends with one click, without worrying about infrastructure O&M, and focus on business logic innovation.

3. Core Feature Introduction

3.1 Context Service

Launch phase:

  • Zero-threshold activation: No deployment required. Ready to use out of the box, eliminating the burden of infrastructure O&M.
  • Multi-tenant data isolation: A workspace-based multi-tenant data isolation architecture ensures data independence across projects.
  • Agent integration: Use RESTful APIs to easily connect self-built agents and mainstream AI ecosystem tools.

Advanced lossless migration: As your business scales and requires high concurrency, strict fencing, and advanced customization, you can seamlessly migrate from the managed service pattern to dedicated resources. The same underlying architecture ensures consistency of data formats and interface protocols. The migration procedure requires zero code changes and zero data loss, achieving a smooth transition from getting started to advanced at the lowest cost.

3.2 Context Management

Supports memory data classification

  • Episodic memory: A UNIX timestamp event log that records "when and what happened".
  • Semantic memory: Connects to enterprise-level knowledge bases.
  • Procedural memory: Stores Agent Skills to provide data support for agent iteration optimization.

Enterprise knowledge base

  • GraphRAG: Provides knowledge graph and retrieval-augmented generation build, management, and application capabilities.

  • Update frequency:

    • Static/semi-static knowledge: updated daily/weekly;
    • Dynamic knowledge: updated by minute/hour;
    • real-time knowledge: updated by second/sub-second.

Context assembly: Assembles contexts through ContextBlock to implement class file system management, with modular encapsulation of retrieval-augmented generation, memory, and skills. This enables contexts to be managed, retrieved, and shared like files, improving the modularity, collaboration, and security of agents.

Agent bootstrapping: Agent skills in procedural memory undergo iterative optimization, such as removing redundant steps and fixing fault logic.

3.3 Data Sandbox

High-speed branch Management: Creates branches within milliseconds for the same copy of data based on Copy on Write technology. Supports multilayer sub-branch creation. Each branch has an independent environment and fencing Status, ensuring parallel jobs do not interfere with each other. Supports deriving new branches from the main branch (MAIN) at any specified point in time, perfectly adapting to parallel exploration and hypothesis authentication scenarios for agents.

Agile Reset and recover: Provides fine-granularity Fault Tolerance mechanisms. When branch data is contaminated or logic errors occur, you can one-click Reset the Status from the parent branch, or directly discard the current branch and recreate it. This greatly reduces agent trial-and-error Cost and ensures business continuity.

Time-Travel: Supports point-in-time "data travel" queries. Multi-Agent systems can obtain data snapshots and Status context at any historical moment to implement cross-time Status comparison, review, and decision optimization.

Full-link compliance Audit: Built-in multi-Version data retention mechanism that completely records data change history. Supports full data backtracking and operation log audit to meet enterprise-level Security compliance requirements and ensure that every agent decision is documented.

3.4 Context Hub

3.4.1 One-Stop Backend Service

Supports Serverless one-stop backend service. AI agents can directly invoke resources such as databases, Auth, Storage, and Edge Functions without manual configuration or debugging. Fully managed and O&M-free.

  • Creation in Seconds: AgenticDB can complete an instance cold start within seconds when the agent receives a user instruction or triggering event.
  • Auto Scaling: Supports AutoScaling. Users can configure elastic resource ranges to automatically scale resources based on traffic.
  • Stop in Seconds: Once the agent completes a job and becomes idle, or no interaction is detected for an extended period, the system automatically suspends the instance and releases compute and disk storage resources. Data is automatically archived to low-cost storage, reducing costs to near zero.
  • Secure Model Invocation: Places token invocations on the server-side to prevent key leakage.
  • SSO Authentication: Builds enterprise SSO logon capabilities based on the OAuth 2.0 protocol.

3.4.2 Massive Data Retrieval-Augmented Generation DPI Engine

Supports hybrid indexes combining graph and partition indexes (NOVA Disk), with built-in quantization compression algorithms such as PCA, PQ, and RaBitQ, and asynchronous write + merge query read/write pattern. Outperforms mainstream open source competitors across multiple dimensions including query retrieval, index build, real-time write/update/delete performance, storage cost, and efficiency.

  • Massive data scale: Supports 100 billion+ vector data scale with distributed extension.
  • Extreme cost optimization: Only 1/10 the cost of open source products at equivalent performance requirements.
  • Ultra-high build efficiency: Building tens of billions of vector indexes takes only 30 minutes.
  • Millisecond-level response: Ann search with Recall 95%+ at sign Top 5000 completes within 100 milliseconds, supporting high concurrency real-time query scenarios.

3.4.3 Global Data Lakehouse

  • Data lakehouse: Leveraging the resource pooling and massive Object Storage Service capabilities of cloud-native infrastructure, the data lakehouse deeply integrates Massively Parallel Processing (MPP) database technology and Serverless technology to achieve compute-storage separation, second-level Auto Scaling, and efficient one-write-multiple-read capabilities, helping enterprises significantly reduce the total cost of ownership in massive data analytics and high-concurrency read scenarios.
  • AI Table: The powerful engine behind DingTalk AI Table, applicable to scenarios such as massive data, high-frequency changes, multi-table joins, and real-time computing. It integrates multiple core technologies, including Streaming View, Laser vectorized engine, Beam multi-version row-column hybrid engine, hybrid transactional and analytical processing, and Massively Parallel Processing (MPP) parallel computing, supporting hot tables with tens of millions of rows and providing robust support for complex business scenarios.

4. Conclusion

As the AI Agent era arrives, the AI-Native agile enterprise development pattern is becoming an industry trend. Traditional data infrastructure constrains AI-Native enterprises from realizing their potential, rapidly expanding into the marketplace, and building secure, isolated, traffic-aware AI context infrastructure after reaching maturity. AgenticDB focuses on context storage, security, management, and application to provide full-lifecycle product capabilities for AI-Native enterprises at different development stages, helping them build a product moat in the AI era. AgenticDB is now officially open for invitational preview. We sincerely invite AI-Native enterprises and organizations to use it and build the new data infrastructure for the AI era. We believe that more highly secure, user-aware super AI applications will emerge in the future, reaching every corner of society and industry.

Database AnalyticDB AnalyticDB for PostgreSQL Vector database Data Lakehouse AI Agent AI-native database AgenticDB AI Context Management AI Table
0 1 0
Share on

Read previous post:

PolarDB-X Optimizer Tips | Incremental Prediction of Time Columns in Statistics

Read next post:

Fully-managed PolarDB-X Mem0 Service: Enabling Infinite AI Memory Extension and Dual-Use Integration

ApsaraDB

633 posts | 185 followers

You may also like

Comments

ApsaraDB

633 posts | 185 followers

Related Products

More Posts by ApsaraDB

See All