Agent Native: Ultra‑Fast Analytical Database Paradigm for Agents

At the "Agentic AI Summit" in Shenzhen, Ma Ruyue, CEO of Feilun Technology and founder of Apache Doris, shared the concept of "Agent Native" data infrastructure, arguing that databases must evolve from merely machine‑readable to truly Agent‑ready.

Background: Why Agents Need a New Data Foundation

Agents act both as heavy data consumers (frequent queries) and producers (massive trace, prompt, token streams).

Four fundamental challenges arise:

Core proposition: databases must transition from "Machine‑Readable" to "Agent‑Ready".

Solution Overview: SelectDB Agent Native Capabilities

Ultra‑Fast Engine : Benchmarked with ClickBench, TPC‑H, and TPC‑DS, delivering sub‑second query latency to support dozens of queries per reasoning step.

Unified Engine (One Engine, All Data) : Native mixed‑mode retrieval across structured, JSON, full‑text, and vector data, eliminating the need for multiple systems.

Agent Native Stack :

MCP Server – Direct integration with leading agents such as Claude Code, Codex, and Cursor.

Semantic Layer – Defines metrics and dimensions so agents move from "write SQL" to "understand business".

CLI + Skill – Encodes capacity planning, table design validation, and slow‑query diagnosis into executable workflows for AIOps.

Agent Observability – Litefuse open‑source platform creates a closed loop: trace collection → storage → analysis → evaluation.

Compatibility with the Langfuse SDK enables integration with over 100 AI ecosystem components (LangChain, Dify, OpenAI SDK, etc.).

Cloud Elasticity – Compute‑storage separation, second‑level elastic scaling, and pay‑as‑you‑go billing via Alibaba Cloud SelectDB Serverless.

Implementation Challenges and Mitigations

Latency Amplification : Multi‑query bursts are handled by the ultra‑fast engine and adaptive optimizations to keep response times in the sub‑second range.

Data Silos : A single engine manages all data modalities, providing native mixed‑mode search (One Engine, All Data).

Semantic Gap : The semantic layer together with MCP Server lets agents first comprehend business logic before generating queries.

Black‑Box Decisions : Litefuse captures full trace data, storing and analyzing it to close the observability loop.

Load Spikes : Storage‑compute separation and serverless elasticity absorb sudden inference peaks.

Case Study: The "JieJie XingChen" project built an Agent Trace platform on SelectDB, achieving a complete data‑to‑evaluation pipeline.

Future Roadmap

Agentic Analytics – moving from "human‑asks‑data" to "agent‑driven autonomous analysis".

Ecosystem Expansion – standardizing the MCP protocol and launching a community skill marketplace.

Open‑Source Collaboration – advancing the Litefuse community roadmap.

Conclusion

The next evolution of databases is from pure analytical engines to real‑time Agent‑centric analysis platforms. Speed is the baseline, unification the foundation, Agent‑Native the core, and observability the guarantee, turning data into a resource that serves agents directly.