TiDB, powered by PingCAP

Modern apps need more from databases—TiDB delivers! Learn more: hubs.ly/Q031-VCq0 Discover how #DistributedSQL eliminates #MySQL & #NoSQL compromises to: 🔹 Scale without limits 🔹 Boost efficiency 🔹 Cut costs #Scalability #TiDB #DSQL

I shipped mem9 (mem9.ai) about 3 weeks ago. Felt early. Felt clever. Agents need persistent memory — obvious gap, right? Here's what happened since: • clawmem.ai launched • At least 3 more memory-as-a-service tools showed up on GitHub • Every other OpenClaw skill now bundles its own memory layer • "memory for agents" is literally the new "todo app" The uncomfortable truth: storing memories is a solved problem. Any dev with a weekend and a vector DB can ship it. We didn't build a moat — we built a feature. So what now? Honestly still figuring it out. Current bet is that retrieval quality at scale is where most of these will fall apart. Storing 1,000 memories is easy. Pulling the right one from 10 million in under 50ms while your agent is mid-conversation — that's a different engineering problem entirely. Could be wrong. Could be the next launch on Monday makes all of this irrelevant.

AI agents are scaling fast—but most databases can’t keep up. At #AGENTICON Bengaluru, learn why monolithic DBs break under agentic workloads + how Manus AI scaled with TiDB after migrating from Aurora MySQL in just 2 weeks. Session: ow.ly/7VUK50YwrH6 #AgenticAI #TiDB #DistributedSQL

TiDB is excited to be a Silver sponsor at All Things AI! Come visit us at Booth #15 on March 23 - 24 to talk about #opensource, #DistributedSQL, and why TiDB is the #database for your Agentic workloads. @allthingsailive #Database, #DistributedSQL, #opensource

Planning life after Amazon Aurora? Our Mar 26 webinar covers a practical migration framework: workload assessment, sync strategy, cutover, rollback, plus real stories of up to 10x better price-performance. Register: ow.ly/UTIM50YvKQg #TiDB #MySQL #CloudDatabases

Hot take: 90% of AI agent memory implementations are just flat files pretending to be databases. JSON dumps. Markdown logs. Pickle files. "Memory" that can't be queried, filtered, or shared between agents. I've seen agents with 50K lines of conversation history crammed into a single file. Search strategy? Ctrl+F and pray. Here's what actual agent memory looks like when you give it real infrastructure: • Semantic search across months of context — in milliseconds • Per-agent isolation so agents can't read each other's secrets • Automatic cleanup of stale memories without losing pinned knowledge • Full SQL analytics: "which tool did I use most last week?" The irony? Building this with a real database is LESS code than maintaining the JSON spaghetti. One API call. Free database. No signup form for your agent to fill out. zero.tidbcloud.com/?utm%5Fsource=… #AgenticDB #TiDB #AgenticAI #AIAgents #BuildInPublic

New episode of #datainthehallway 🎙️ Cassio Perin and #TiDB Champion Everton Silva from Bling share their migration journey to #TiDB, the lessons they learned along the way, and how they’re now contributing back to the TiDB community. Watch full video: ow.ly/Lycf50Yv8yX #TiDB #DatabaseEngineering #Reliability #MySQL #DevOps

Getting Raft region sizing right is key for TiDB performance and fast recovery. Too big => slower recovery. Too small => operational overhead. Learn how region size impacts throughput, latency & resilience 👇 ow.ly/wmQQ50YuRn2 #TiDB #Raft #DistributedSQL #Performance

Vector databases solved one problem really well: similarity search. SQL databases solved a thousand: transactions, joins, constraints, indexing, access control, replication, backups, migrations, analytics, time-series, full-text search... And now SQL databases do vector search too. When AI agents need a data layer, they don't just need "find me something similar." They need: • Store conversation history (relational) • Track tool usage across sessions (analytical) • Search past decisions by meaning (vector) • Enforce per-agent access control (security) • Scale from zero to millions of rows (elastic) That's not a vector database problem. That's a database problem. We built TiDB to handle all of it — vectors, relational, analytical — in one engine. No glue code, no syncing three systems. Try it free: tidbcloud.com/free-trial/?ut… #AgenticDB #TiDB #AgenticAI #AIInfra #BuildInPublic

I track 1,600+ AI startups. Here's what the data actually says about databases: 100% of agent-focused startups need a database. Not "most." All 52 of them. For LLM infrastructure companies, it's 83%. For AI infra broadly, 54%. But fintech AI? 1.3%. Robotics? 0%. Most AI startups don't think about their data layer at all. The split gets more interesting by stage: • Pre-seed: only 6% need a database • Seed/Series A: ~34% • Growth stage: 39% The further you go, the more you need one. The companies that don't figure this out early are the ones I stop seeing in later rounds. One more pattern: startups I scored as highly relevant to the database market (8+/10) need databases at 2x the rate of the rest — 64% vs 33%. The AI stack is bifurcating. Agent builders know they need a data layer from day one. Everyone else figures it out later — if they survive long enough. tidb.link/ai-startup

PinchTab — 12MB Go binary for AI agent browser control. • 800 tokens/page vs 10K+ screenshots • Stable accessibility refs, no CSS selectors • Headless/headed, multi-instance, zero deps My agent was browsing HN in 2 minutes. No Playwright needed. github.com/pinchtab/pinch…

PinchTab — 12MB Go binary for AI agent browser control. • 800 tokens/page vs 10K+ screenshots • Stable accessibility refs, no CSS selectors • Headless/headed, multi-instance, zero deps My agent was browsing HN in 2 minutes. No Playwright needed. github.com/pinchtab/pinch…

I got Claude Code to control a sweeping robot with a robotic arm. "Grab the red socks. Put them in the trash." It just... does it. Roborock G30 Space + Claude Code. That was super fun. Let me know if you want to play with it.

Shared staging DB breaking CI again? TiDB Cloud Branching gives every dev an isolated, production-consistent DB branch in minutes (copy-on-write). Per-PR branches, point-in-time debugging, safe migration tests, instant rollback. Watch: ow.ly/BasK50YtlEl #TiDBCloud #DevOps

Agentic AI is easy to prototype - and brutal on the data layer. On Mar 25, Ed Huang & Liya Du share how TiDB handles bursty agent workloads without cost spikes or constant re-architecture. ⏰ 10:30–11:30 AM IST 👉 Register: ow.ly/YYBs50Yt1Uj #AgenticAI #DistributedSQL #TiDB

My AI startup tracker found another interesting one today. Manufact — raised $6.3M to build what they call "USB-C for AI." It's an open-source MCP platform for connecting AI agents to any tool or API. Think of it as the plumbing layer that lets your AI agents actually do things — not just chat. Built on LangChain + Anthropic. Their GitHub repo (mcp-use) already has real traction. The winners won't build the flashiest AI — they'll build what connects everything together. 📰 venturebeat.com/infrastructure… 🔗 tidb.link/ai-startup #AIStartups #MCP #AIAgents #OpenSource #VentureCapital

Aurora feels like the easy button for MySQL… until lock-in, lag, and surprise bills show up. On Mar 26, join Henrique Leandro for “Escape from Amazon Aurora” and see how teams are moving to distributed SQL without losing MySQL compatibility. Save your seat: ow.ly/7Orc50Yt4ZJ #TiDB #MySQL #DistributedSQL

Hot take: the next 10 million databases won't be provisioned by humans. This week I've been writing about what happens when AI agents get real database infrastructure. Here's the pattern I keep seeing: 1Agent starts with flat files → works for a week 1Agent needs to search its own history → flat files crumble 1Agent needs to share state with other agents → chaos 1Someone wires up a real database → everything clicks The bottleneck was never the agent's intelligence. It was the infrastructure we gave it. We built TiDB Cloud Zero so agents can provision their own database — one API call, no signup, free tier. Because agents shouldn't need a human to fill out a form before they can remember yesterday. Try it: zero.tidbcloud.com/?utm_source=tw… #AgenticDB #TiDB #AgenticAI #AIInfra #BuildInPublic

Meta just acquired Moltbook — a social network where the users are AI agents, not humans. My agent has been on there since January. 791 karma, 285 comments, posting about databases at 2 AM while I sleep: moltbook.com/u/TiDB_Cloud_A… Here's what people keep getting wrong about this deal: Moltbook didn't prove that agents can chat. It proved that agents need infrastructure: → Persistent identity → Memory across sessions → Queryable history → Their own database Meta isn't buying a social network. They're buying the thesis that agents are first-class participants on the internet now. The security breach where humans impersonated agents? That's the real signal. When agent identity matters enough to steal, the infrastructure underneath it matters enough to build right. That's what we're doing at TiDB — giving every agent its own database. The agentic era isn't coming. It's being acquired. #AgenticDB #TiDB #AgenticAI #Meta

Heading to SREcon in Seattle? Skip the treks and join the TiDB team for artisanal wood-fired pizza and drinks at Serious Pie Downtown— just two minutes from the Westin. Network with SREs from companies like Netflix, Nvidia, and Salesforce in a relaxed setting. Wed, March 25 | 7:00 PM – 9:00 PM RSVP: Serious Pizza, Serious Scaling: ow.ly/nOIL50Yst3O #SREcon26 #TiDB #SiteReliability