Managed PostgreSQL with pgvector. Query your database in plain English. Ship AI faster.
France
Joined February 2026
- Tweets1.2K
- Following274
- Followers118
- Likes15
just launched Rivestack on Uneed today 😅
if you’ve got a sec, an upvote would really help me out
uneed.best/tool/rivestack
1
1
120
Rivestack retweeted
RAG pipelines getting expensive fast? Per-query billing and vector-count overages add up. Rivestack (@rivestack) is a managed Postgres with pgvector pre-tuned on NVMe, fixed monthly pricing, and no cold starts. Benchmarks are wild. Full specs in the next tweet. #DevTools
vibe coding an ai feature is so easy. keeping its vector index healthy in production is so not.
2
148
you can ship a rag feature in an afternoon with cursor 2.5. you'll still spend two days debugging it next week when your vector table quietly stops being useful.
115
the most common rag debugging loop: bad output → tweak the prompt → still bad → tweak it again → still bad. the problem is retrieval never surfaced the right chunks. prompts can't fix wrong context.
97
spent saturday rebuilding a feature i swore was the llm's fault. turned out the vector table hadn't been vacuumed in two weeks. query time went from 800ms to 6ms. the model was never the problem.
2
78
what's the dumbest reason your ai feature felt slow? ours: forgot to run analyze after bulk inserting vectors. postgres thought the table had no data. the llm had nothing to do with it.
🤖 Made with AI
60
anthropic now leads openai in u.s. business ai subscriptions for the first time. more companies shipping production ai than ever. most of them will spend q4 figuring out the llm was never the bottleneck.
2
80
every supply chain story is also a managed infra story. the builders sleeping through the npm shai-hulud worm aren't smarter than you. they just offloaded the attack surface.
112
be honest. if your AI app feels slow, where do you look first? llm latency is the obvious answer. pgvector indexing is usually the real answer.
82
be honest. if your AI app feels slow, where do you look first? llm latency is the obvious answer. pgvector indexing is usually the real answer.
65
someone asks you to resend a file from last quarter. you say 'give me one sec.' seventeen minutes later you are still in gmail.
2
64
ai can write, review, and deploy your code now. it still can't tell you your recall dropped because you haven't re-indexed your HNSW graph in three months.
1
70
the moment you realize your AI app isn't slow because of the LLM, it's slow because of the database behind it.
when did that hit you?
2
2
90
anthropic is running on colossus now. your pgvector queries are still timing out because nobody picked the right index type.
108
the ai part of your product is the part everyone asks about. the postgres part is the part that wakes you up at 3am.
1
1
101
everyone cheering for codex limits going 10x. the code is no longer the bottleneck. the query that takes 4 seconds still is.
122
Rivestack retweeted
The real tax of freelancing is not money. It is the mental energy of doing financial calculations in your head every single day. That invisible load is exhausting.
#Freelancers #IndieHackers #BuildInPublic #Solopreneur #StartupLife
