we’re launching Codag (Y Combinator S26) - the compression and control layer for agent tools. we started with logs. then customers pulled us into searches, tests, builds, and api responses. the pattern was always the same: agents burning context on repetitive output, with teams unable to see where the... tokens went. Codag now cuts input-token usage ~3x at answer parity and gives teams visibility across their agent stack. agents are becoming the primary users of software - companies are adding agents, models, providers, and MCP servers faster than they can measure or control them. most cannot answer which tools their agents invoke, what those calls cost, or where the waste is. we’re building the layer between them and everything they read. 90+ engineering orgs already use Codag today to reduce that spend and see agent tool usage across provider. find out more and book a demo atshow more

@ycombinator Context bloat from tool returns is such an underrated problem. Agents getting stuck in loops because 80% of their window is filled with raw stdout/stderr logs happens way too often. Compressing tool output before it hits the prompt is a huge win

@ycombinator yep 100% that’s what we’re targeting

@ycombinator CONGRATULATIONS FOR MAKING AI CHEAPER

@ycombinator congratulations to the team

@ycombinator thanks! i (the whole team) appreciate it

@ycombinator that's not just about logging or i/o efficiency it's about how you define 'output' in a world where an agent can produce endless possibilities. codag's job is to figure out which ones matter.

@ycombinator agreed, super hot space

@ycombinator 3x input token compression at answer parity is the number that matters. Most agent tooling logs spend after the fact, you're compressing before the call reaches the model. How's Codag handling schema drift on MCP servers, hash pinned or diffed live against output shape?

@ycombinator nice

@ycombinator 🔥🔥🔥

@ycombinator congrats 🎉

@ycombinator thanks man!

@ycombinator y'all got discord community set up yet?

@ycombinator i don't, probably not going to be soon since we're not open core rn. i'll consider it

@ycombinator got it, if you're looking to launch one anytime i am excited to help. ps: I'd advise we kick community running Day-one even if it's a 20 power user group then full blown community whenever you feel is the right time. No pressure either way Micheal 🙌🏻

@ycombinator Y'all hiring?

@ycombinator goat

@ycombinator Congrats!!!

@spei04 @ycombinator super cool michael, congrats!

@ycombinator how can i try this? can i grt get access to this? would love to give my feedback if that helps. i just wanna try this out.

@ycombinator self-serve is on the 6 month roadmap, it's just me and i'm focused on b2b for now. i'll post when its out! tysm tho

@ycombinator @grok 这个纯研究还是有工业意义，具体工业场景视角看意义是什么，有开源数据集或者开源项目代码吗？从多个数据源交叉验证，不要只看新闻媒体一面之辞。帮我排除没意义的垃圾商业营销推广、诈骗、夸张博眼球、虚假新闻 以及自吹自擂，自嗨，无病呻吟。

@ycombinator awesome!

@ycombinator GOAT

i've hit this exact thing. the part that kept biting me wasn't pulling the right context, it was catching when the agent pulled *wrong* context and burned half the window on it. curious how you decide what to compress vs what to surface fully — that's where the trust question lives for me.

@ycombinator Similar to my CSM!

@ycombinator Compression layers work until the agent needs the one detail you decided was noise last week. Curious how Codag handles retrieval when the "irrelevant" output turns out to matter.

@ycombinator HUGE CONGRATSSSSS

@ycombinator Let's go Michael! Such a no-brainer product for engineering orgs

@ycombinator thanks jae! means a lot coming from you

@ycombinator Awesome product. Congrats Michael!

@ycombinator logs are the right place to start, agents burn a stupid amount of context on stack traces they cant act on

@ycombinator Bet

@ycombinator Huge congrats Mike! Let’s get ittt

@ycombinator DUBSS

@ycombinator congrats man, lfg!

@ycombinator How does Codag handle context compression without losing the crucial semantic meaning or intent that the agent needs for subsequent tool calls? Would love to know how you maintain answer parity.

@ycombinator OSS?

@ycombinator This is the right place to sit. Once you’re already in the tool path, the useful question is who gets to invoke, not just how cheap the result was to read. Compression helps the model. Control is still a yes or no before the call.

@ycombinator tool level token visibility is the missing view

@ycombinator Congrats on the launch I'm a motion designer and checked out your launch video. The product is cool, but I think it could feel much more premium with better motion, a stronger hook, and branded animation. If you ever need help with launch videos, DM me.

@ycombinator Comprimir la salida de las tools antes de entrar al contexto es la palanca clave, los agentes queman casi todo releyendo logs y dumps enormes.