Semantic search improves our agent's accuracy across all frontier models, especially in large codebases where grep alone falls short. Learn more about our results and how we trained an embedding model for retrieving code.

Nov 5, 2025 · 5:31 PM UTC

65
104
74
1,420
0
cursor.com/blog/semsearch

Improving agent with semantic search · Cursor

Built to make you extraordinarily productive, Cursor is the best way to code with AI.

cursor.com
2
7
3
135
Replying to @cursor_ai
@grok what is grep? What is semantic search?
1
Replying to @cursor_ai
Is cursor still using turbopuffer?
2
2
Replying to @cursor_ai
Don't get, is it already available ?
1
Replying to @cursor_ai
hybrid search is all you need :) tigerdata.com/blog/vector-se…

Vector Search Isn't the Answer to Everything. So What Is? A Technical Deep Dive

TigerData’s Jacky Liang argues that vector search alone isn't sufficient for many AI applications since it provides similarity when users need exact relevance.

tigerdata.com
3
1
60
Replying to @cursor_ai
Semantic search is definitely a step up from traditional methods.. but am wondering how it balances precision with the risk of overfitting or even hallucinating results in newer models. Large codebases are tricky beasts to tame.. and while embeddings are powerful, they're not flawless. How does it ensure the agent doesn't miss edge cases or misinterpret intent in complex queries?
2
2
Replying to @cursor_ai
you weren’t using semantic retrieval before?
1
2
Replying to @cursor_ai
I like your semantic search BUT; i'd argue its advantage isnt increasing accuracy but rather speed. and with speed comes more budget for more investigation
1
4
Replying to @cursor_ai
Now we just need 5 years of research/development just get it to stop saying "You're Absolutely Right!"
Replying to @cursor_ai
Thank god
Replying to @cursor_ai
1
15
Replying to @cursor_ai
Is Cursor down? My chat won't load. The rest of the new UI seems to be working.
1
1
1
Replying to @cursor_ai
I noticed with 2.0 that I can't use the @ symbol to call on context. Is that because Cursor wants to push people towards relying solely on agentic search workflows?
Replying to @cursor_ai
Everyone building AI coding tools but Cursor actually focused on the boring stuff that matters. Finding the right code is half the battle. Make that faster and everything else gets easier. Smart move.
32
Replying to @cursor_ai
Look at this visual, isn’t it insane: you can say “obviously search by meaning, not by text, is better” - but even obvious things can be lost and will need to be found again over time. Especially, as it’s said in the docs - when the codebase is large
8
Replying to @cursor_ai
@cline joins the chat? I believe their take is that just grep is better
1
Replying to @cursor_ai
grep pain is too real
1
Replying to @cursor_ai
@cline boys in shambles
4
Replying to @cursor_ai
很高兴看到语义搜索的进展！这将极大提高开发效率。
Replying to @cursor_ai
does that mean we dont need serena mcp anymore?
Replying to @cursor_ai
I'm not convinced semantic search is a replacement for expert-level intuition yet.
1
Replying to @cursor_ai
This is HUGE. Imagine never having to dig through endless lines of code again. Semantic search is the future for dev productivity!
Replying to @cursor_ai
cooked!!
3
Replying to @cursor_ai
Ah, semantic search is a game-changer, Cursor! That's a clever way to tackle those massive codebases, I must say.
Replying to @cursor_ai
Semantic Search is a Great Addition!
Replying to @cursor_ai
That’s huge 🔥 Semantic search is such a game-changer, especially when dealing with massive repos where keyword search just can’t keep up. 👍🏻 Great updates from @cursor_ai lately
Replying to @cursor_ai
"new embedding model what do you think"
12
GIF
Replying to @cursor_ai
You’re shipping too fast guys I can’t keep up
Replying to @cursor_ai
Excellent feature. I have been waiting for semantic search from day one. Thank you 🫡
Replying to @cursor_ai
If you want a Rust-based hybrid semantic + BM25 search (MCP + CLI) that runs with local embedding models optimised for agents, try ck (“seek”) it’s wild watching agents taking it for a spin! GitHub: github.com/BeaconBay/ck • Docs: beaconbay.github.io/ck/

GitHub - BeaconBay/ck: Local first semantic and hybrid BM25 grep / search tool for use by AI and...

Local first semantic and hybrid BM25 grep / search tool for use by AI and humans! - GitHub - BeaconBay/ck: Local first semantic and hybrid BM25 grep / search tool for use by AI and humans!

github.com
1
13
Replying to @cursor_ai
Hello, please add a “Summarize Chat Now” manual option to free up some context while keeping the same thread and maintaining the current context.
1
Replying to @cursor_ai
If Grep isn’t doing a great job for your codebase they consider, to take a look at how you are organizing code Cuz grep be like:
1
GIF
Replying to @cursor_ai
Is Cursor semantic search by default?
Replying to @cursor_ai @srush_nlp
...maybe make basic search work inside a conversation first ffs @cursor_ai
2
Replying to @cursor_ai
@grok how does semantic search actually helps when we have a larger code base. Help me understand here a bit brief.
1
Replying to @cursor_ai
Do you have more details about the model?
1
Load more