# TiDB Product Knowledge Reference

> Source of truth for product naming, capabilities, tier positioning, and approved metrics.  
> Skills load this before writing any content referencing TiDB product specifics.  
> **Update after every major release.**

Last updated: 2026-07-24  
Sources: pingcap.com official pages (fetched 2026-07-24)

---

## Current positioning

**Primary tagline:** Scale Your AI Agents Without Compromise  
**Category:** AI-Native Distributed SQL Database  
**One-sentence positioning:** TiDB is the AI-native distributed SQL database that unifies vectors, transactions, and analytics in one engine, enabling AI application builders to scale agent workloads without managing separate data systems.

**Core value pillars (in priority order for current messaging):**
1. **Unified engine** — Vectors, transactions, and analytics in one SQL call
2. **Elastic scale** — Autoscaling compute and storage; adapts to unpredictable agent workloads
3. **ACID consistency** — Agent state stays consistent across concurrent workflows
4. **Developer ready** — Standard SQL; MCP server, pytidb, LangChain, LlamaIndex, DSPy integrations

---

## Official product tiers (2026 — CORRECTED NAMES)

> ⚠️ "TiDB Cloud Serverless" is the **deprecated** name. Do not use it.

| Tier | Official name | Status | Target user | Entry motion |
|------|--------------|--------|------------|-------------|
| Managed — free/dev | **TiDB Cloud Starter** | GA | Developers testing and building; free tier, scales to zero | PLG; free |
| Managed — production | **TiDB Cloud Essential** | Preview | Growing teams; production-grade; autoscaling | PLG → SLG |
| Managed — enterprise growth | **TiDB Cloud Premium** | Preview | Enterprise workloads needing unlimited growth, advanced security, instant elasticity | SLG |
| Managed — enterprise | **TiDB Cloud Dedicated** | GA | Enterprise; dedicated resources; business-critical | SLG |
| Self-hosted | **TiDB Self-Managed** | GA | On-prem / private cloud; OSS contributors | OSS download; SLG for support |
| Hyperscale elastic | **TiDB X** | GA (powers Starter + Essential) | Hyperscale AI / enterprise; underlying elastic architecture | Not a standalone SKU; the engine behind Starter/Essential |

**Note on TiDB Cloud Premium:** Added to the tier lineup between Essential and Dedicated (confirmed live on pingcap.com/tidb/cloud/, 2026-07-24). Verify current Preview/GA status before citing externally — Preview tiers can flip to GA without a CHANGELOG note here if this file isn't re-checked.

**Note on TiDB X:** The product page says "Powered by TiDB X" for Starter and Essential tiers. TiDB X is the elastic architecture, not a separately purchasable tier. Do not market it as a tier — reference it as the underlying technology.

---

## Key technical components

| Component | Name | What it does |
|-----------|------|-------------|
| Storage engine | **TiKV** | Distributed key-value store; handles OLTP writes via Raft |
| Analytical engine | **TiFlash** | Columnar replica; handles OLAP queries; real-time replication from TiKV |
| Change data capture | **TiCDC** | Replication and CDC |
| Backup tool | **BR (Backup & Restore)** | — |
| Kubernetes operator | **TiDB Operator** | Manages TiDB on Kubernetes |
| Cluster manager | **TiUP** | Deployment and cluster management |
| Analytics bridge | **TiSpark** | Spark integration for analytical workloads |

---

## Scale and reliability specs (from official sources)

| Metric | Value | Source |
|--------|-------|--------|
| Max cluster capacity | Up to 500TB per cluster | pingcap.com/solutions/modernize-mysql-workloads |
| Throughput | 2 million QPS at millisecond latencies | Same source |
| Availability | 99.999% uptime during planned outages | Same source |
| DB sprawl reduction | 80% less vs single-node deployments | Same source |
| Customers | 4,000+ global enterprises | pingcap.com/compare/mysql-vs-tidb |
| Cost vs MySQL | 52–67% savings (8,000 QPS, 500GB, aws-us-east-1) | pingcap.com/solutions/lower-infrastructure-costs |
| AI adoption signal | 90% of new daily TiDB Cloud clusters created by AI agents | pingcap.com/compare/best-database-for-ai-agents |

**Usage rule:** Always cite source and workload context when quoting these numbers externally.

---

## Approved customer proof points (validated from official pages)

### AI / Agentic
| Customer | Use case | Approved claim | Source |
|---------|---------|---------------|--------|
| Manus | AI agent platform | "From Viral Launch to 1M+ Database Tenants" — migrated to TiDB Cloud in two weeks; TiDB now powers context persistence behind Manus's agent swarms and full-stack app generation | pingcap.com/case-study/manus-agentic-ai-database-tidb/ |
| Kimi | AI agent hosting platform (K2.6) | "Millions of Agent-Created Databases, Provisioned in Under a Second" — sub-1-second database provisioning per agent-built site; 10M+ tenants supported on TiDB Cloud's multi-tenant architecture; zero per-site compute cost at idle | pingcap.com/case-study/kimi-2-6-agent-hosting-platform-tidb-cloud/ |
| Plaud | AI hardware (smart recording device) | "From Two Data Stores to One Unified Database" — migrated from MySQL + Amazon S3 to TiDB Cloud; eliminated S3 retrieval latency and unlocked online DDL for 2M+ users across 170 countries; 10x QPS improvement under peak load; 13 TiDB Cloud clusters in production | pingcap.com/case-study/how-plaud-eliminated-s3-latency-limitless-scale/ |
| Dify | AI workflow platform | "Consolidated backend infrastructure; unified vectors, documents, and relational data" | pingcap.com/ai/agentic-ai |
| Rengage | Marketing automation | Real-time user context for agent decisions | Same |
| Toprism | Business intelligence | Hybrid semantic + SQL queries | Same |

### Fintech
| Customer | Use case | Approved claim | Source |
|---------|---------|---------------|--------|
| Plaid | OLTP / SaaS | "99.99% uptime with increased developer velocity"; zero-downtime upgrades | pingcap.com/solutions/fintech + saas |
| Chainbase | Web3 infrastructure | "Reduced infrastructure costs by 50%" | pingcap.com/solutions/fintech |
| SB Payment Service | Payment scaling | Zero-downtime scaling for growing payment systems | Same |

### SaaS / Enterprise
| Customer | Use case | Approved claim | Source |
|---------|---------|---------------|--------|
| Atlassian | Cluster consolidation | "750+ Postgres clusters → 16 TiDB clusters" | pingcap.com homepage |
| Catalyst | Performance | "60x faster performance" | pingcap.com/solutions/enable-operational-intelligence |
| ZTO Express | Logistics | "25% reduction in per-order costs" | Same |
| Conga | SaaS HTAP | "Transactional and analytical workloads with exceptional performance" | pingcap.com/solutions/saas |

### E-commerce / Scale-out MySQL
| Customer | Use case | Source |
|---------|---------|--------|
| Pinterest | Graph service, 10x P99 latency improvement | pingcap.com homepage |
| Rakuten, Flipkart, Zhihu, Ninjavan, ZTO | E-commerce scale | pingcap.com/solutions/e-commerce |
| LINE, Shopee, Bolt, Streak | MySQL modernization | pingcap.com/solutions/modernize-mysql-workloads |
| Square, Bolt | Fintech / general | Logo display only; no specific metrics approved |

**Rule:** Never cite a customer metric that isn't in this table. If a customer is "logo display only," don't attach a number to them.

---

## Core capability claims: what to say vs. what not to say

| Correct | Prohibited | Reason |
|---------|-----------|--------|
| "MySQL protocol compatible — see compatibility docs for known differences" | "MySQL compatible" / "fully MySQL compatible" | Documented incompatibilities exist; validate in POC |
| "Rolling upgrade with minimal per-node disruption" | "Zero downtime" | Node restarts occur |
| "Scales horizontally to [tested config]" | "Scales infinitely" | All systems have practical limits |
| "TiFlash replication introduces small lag (typically seconds)" | "Real-time analytics" without lag disclosure | TiFlash has sub-second to few-second replication lag |
| "HNSW-based vector indexing with SQL hybrid queries" | "Vector database" (standalone) | TiDB is a unified system, not primarily a vector DB |
| "MySQL protocol compatible: Snapshot Isolation by default" | Implying SERIALIZABLE isolation | TiDB default is Snapshot/REPEATABLE READ, not SERIALIZABLE |
| "Distributed transaction coordination introduces latency tradeoffs" | Hiding this trade-off | Acknowledged in all official compare pages |

---

## AI developer integrations (approved)

| Integration | Description |
|------------|-------------|
| **MCP (Model Context Protocol)** | TiDB MCP server — a tool-calling interface that lets agents query TiDB via standard tool calls. MCP is a communication protocol only; it does not provide state governance, identity management, or audit trails. Do not claim MCP solves state portability. |
| **pytidb** | Python SDK for AI workloads |
| **LangChain** | Vector store + document retrieval integration |
| **LlamaIndex** | RAG pipeline integration |
| **DSPy** | Prompt optimization framework integration |
| **OpenAI / Gemini / Amazon Bedrock / Ollama** | Embedding and LLM partners |

---

## Agent State Stack — composition and messaging

The "Agent State Stack" is the GTM framing for TiDB Cloud Zero + mem9 + drive9 as a composed system for agent state infrastructure. These are not a single product — they are layered use cases on the same SQL substrate.

| Layer | Product | Role | Target |
|-------|---------|------|--------|
| **Substrate** | **TiDB Cloud Zero** (zero.tidbcloud.com) | Instant, MySQL-compatible, serverless SQL. Independent from any harness, portable. Entry point for queryable agent state without operational overhead. | Any developer who needs structured, queryable agent state |
| **Memory use case** | **Mem9** (mem9.ai) | Managed persistent memory API for agents and tools, built on TiDB Cloud. Shows how the substrate becomes production-grade cross-session memory. | Coding agents, AI assistants needing cross-session user memory |
| **Artifact use case** | **Drive9** (drive9.ai) | Workspace/artifact/document storage layer, built on TiDB Cloud. Shows how the substrate extends to file-like agent-generated content. | Agents that produce, read, and revise documents or code artifacts |

**Composition rule:** Zero is the substrate; mem9 and drive9 are *use cases on top of it*, not separate databases. In GTM content, position Zero as the foundation and mem9/drive9 as demonstrations of what that foundation enables.

**⚠️ Internal roadmap watch (not public, do not use in external content):** Internal GTM drafts (2026-06 to 2026-07, tidb_family/, renamed from tidb_family_knowledge/) describe a proposed repositioning of this stack ahead of a "TiDB Cloud FS" Public Preview (~Oct 2026): Starter reframed as a broader "unified storage layer" (working name "TiDB Storagebase"), drive9 folded in as a "TiDB Drive" capability rather than a standalone product, and explicit direction to stop new investment in mem9 as a product (keep it only as a reference implementation). As of 2026-07-24 none of this is live — pingcap.com and zero.tidbcloud.com still describe the stack exactly as documented above (Zero + mem9 + drive9, drive9.ai and mem9.ai both live as named products). Re-verify against pingcap.com before citing; do not adopt "Storagebase," "TiDB Cloud FS," or "TiDB Drive" externally until confirmed live.

**What the SQL substrate is and isn't:**
- SQL substrate is **strongest for control-plane / metadata-plane state**: task history, permissions, user profiles, audit trails, recovery checkpoints — structured records that make an agent's operating history inspectable and portable across harnesses.
- It is **not** the universal carrier of all agent state. Embeddings/vector retrieval, conversation buffers, and raw file artifacts typically live elsewhere. Don't claim the SQL substrate handles all state types.

---

## Reference projects (TiDB in use)

| Project | What it is | GTM relevance |
|---------|-----------|--------------|
| **OSSInsight** (ossinsight.io) | GitHub analytics platform tracking 10B+ events; built by PingCAP on TiDB | Proof of TiDB HTAP at scale; developer community tool |
| **mem9** (mem9.ai) | Managed persistent agent memory API, powered by TiDB Cloud | Shows SQL substrate as production-grade cross-session memory |
| **drive9** (drive9.ai) | Workspace/artifact storage for agents, powered by TiDB Cloud | Shows SQL substrate extended to agent-generated files and documents |
| **TiDB Cloud Zero** (zero.tidbcloud.com) | Instant, serverless TiDB — the substrate layer of the Agent State Stack | Entry point for Starter tier; free, no signup, MySQL-compatible |

---

## Company context

| Item | Correct |
|------|---------|
| Company name | PingCAP |
| Product name | TiDB |
| Founded | 2015 |
| Open-source license | Apache 2.0 |
| GitHub | github.com/pingcap/tidb |
| TiKV GitHub | github.com/tikv/tikv (CNCF graduated project) |
| Gartner rating | 4.9 stars on Gartner Peer Insights |