28 comments
I thought https://playground.cognition.ai/ was just returning some cached query results, but no, they’re actually spinning up real VMs and running live queries without any authentication or restrictions. That must be costing them a fortune.
Currently, all queries are returning "We're under load and processing too many requests. Please try again later."
So that's how that is going ;)
LLM product managers: Show me what's in the context convenient to where I am prompting. Likely the user knowing and editing the precise context between requests will be a user task for a long time
Kickstarting an exploratory open version here https://github.com/aperoc/op-grep since it doesn't look like they will do it.
Actually I do have a question! How come things as substantial as this were just released and not part of a "wave" ? I quite liked the waves way of doing things! Great work either way.
This is really cool. Thank you for this. I'm a Windsurf user since launch and was VERY pleasantly surprised to see this pop up.
I also enjoyed the tech write-up. It's good to see REAL substantial engineering like this which is both highly impressive and highly productized.
SWE-1 has been being booped up by WindSurf to me lately and I've been impressed - often (enough?) getting me the same answers as GPT5 etc., but almost instantly. Gotta say speed is nice.
nice, what does booped up mean? is this gen z lingo?
ha more like how i talk to my two year old. WindSurf's Cascade sidebar tool (which i use in RubyMine) has a stable of LLMs and it somewhat randomly switches the active one out from time to time. So I get a taste of what different ones are like, it's kind of cool.
Searched for 'hi' and it took 166s to return a response using this model: https://pasteboard.co/oB4VqVC5FGkl.png
Claude Code took 0.1s, Cursor CLI 19s
If you ask a real question, then you might get real results.
hey I'm from the SWE-grep team - feel free to ask me any questions :)
Did you intend to answer them, or you just wanted the questions?
No question just wanted to say good job and thanks as a user. Same with deepwiki and codemaps.
Are you actually using grep here? How much data are you searching?
no - grep is just the closest analogy/use case that we have for it. if we end up releasing the CLI it should be as handy and nobrainer as using ripgrep
idk what you expect from a question about "how much data". its tool based search. its a lot.
I'm just learning about agentic search so I'm a bit adrift.
One of my side projects is a full text index for pattern search, and I'm trying to understand how it might fit with that. You mention tool call overhead, but is that a significant part of the latency in the multi-turn scenario, or is it the coding agent being forced into a serial processing pattern?
this would be useful outside of coding. could you release a benchmark so we can have more models tuned for this?
(coauthor) main charts/evals here https://x.com/cognition/status/1978867021669413252
you can try the https://playground.cognition.ai/ here
i wrote a longer explainer here https://x.com/swyx/status/1978874342743343254 but saving you the click
this was a perspective cut from the blogpost, but let me explain why subagents kill long context
Like you can spend $500m building 100 million context models, and they would be 1) slow, 2) expensive to use, 3) have huge context rot. O(n) is the lower bound.
Cog's approach is something you learn in day 1 of CS50 - divide and parallelize. Embeddings are too dumb, Agentic Search is too slow. So train limited-agency (max 4 turns), natively parallel tool calling (avg parallelism of 7-8, custom toolset) fast (2800tok/s) subagents to give the performance of Agentic Search under an acceptable "Flow Window" that feels immaterially slower than Embeddings.
The benefit of this is threefold:
- 8 ^ 4 toolcalls cover a very large code search space. can compound subagent calls if more needed.
- predictable cost & end to end latency
- subagent outputs "clean" contexts, free of context failure modes like context poisoning and context rot
we originally called this Rapid Agentic Search, to contrast with RAG. but Fast Context rolls off the tongue better.
-- Second perspective --
The Fundamental Equation of Coding Agents is:
Coding Agent Performance = Ability to Read the Right Files * Ability to Generate the Right Diffs
Fast Context is Cognition's first solution for the Read. As codebases get larger and and tasks get more complex, Reads get more important. the average production codebase first query in Cascade is >60% just searching and reading files.
But if this were just about speed, it might not be that exciting. I think there are unappreciated effects in performance as well when you have very good context. In other words:
Context Engineering is Actually Very Important. Too important for humans and hardcoded rules.
The swe-greps are the first dedicated context engineer agent models.
Thanks for the summary. I noticed from the announcement you trained on parallel tool calling to save on serial round tripping. This is awesome.
Most LLM coding is so slow that you're permanently out of flow state, and in 'manager' state right now - I'm interested in a future where you've got enough fast low TTFT support that an engineer could maintain flow state and have sort of super power type productivity at the same time, and this tool makes me think of that.
That is, it looks fast enough to be used as a sort of sidebar info tool, as in "what you're coding might need / refer to these other parts of the codebase" -- effectively increasing an engineer's working memory. Super cool. And obviously useful for an AI engineer as well. Thanks for the writeup!
Any plans to offer this as a tool/MCP server for other coding agents or is it going to be Windsurf exclusive?
we have other things in store that can be used by other coding agents, this one was tuned to use custom fast search tools that kinda wouldnt be useful in other agents
guys please release the benchmark or the benchmark code. like this is just "trust me bro"
well thats what the playground is for! playground.cognition.ai
This link redirects to https://cognition.ai/blog/swe-grep now?
got a lot of traffic and was taken down temporarily for a couple reasons - team got it online again last night
yeah but if people would like to double check the results it would be nice to have the actual benchmark. especially given that your playground is broken...
"We ran into an error processing your request. Please try again"
lol dead thread, cognition begging to grab some traction in this space.
