they gave frontier models a 1 billion token context window: you can now run Fable [1B], Opus [100M], GPT 5.6 [500M] across your entire codebase - it's a graph based tool that can traverse databases with 400M+ of nodes in ms - only one prompt to install - &... it's open source (link in video) Introducing Polygresshow more

Check it out on

everyone check this out

hit up the founders! @daleverett & @daltonmeon

this is false information. They didn't increase context window. They just figured what to put in the context window effectively to feel like it's large. But that's highly debatable.

Alright I’m in trying this now

maybe you should

1 billion token and this costs how much?

basically no additional usage costs, you can try their open source repo or pay for their tooling on their site

interesting, will check it out! thanks

ngl u can actually learn from this launch video 🤯

LFG! I’ve been using it and it’s genuinely amazing

I’d like to see benchmarks that show improved agent memory, performance, recall, context consumption and decision making using this. Could be neat to implement but the movement towards world models is more relevant then implementing extremely large context windows.

here are some, you can checkout their site for more. they're releasing some other benchmarks soon

how does this differ from

Fat asians innovating i love to see it

Not quite grasping the value proposition here, is it just a faster ontology / semantic model / knowledge graph implementation? Because I've worked with enterprise data volumes and the issues they describe are real but only for vanilla toolchains. A couple of config knobs and the problem is gone.

@evokoa_ai Gemma 4 でも 1M くらいは余裕⁉️

best educational launch I've ever seen

@daleverett 🐐

you can just lie

this is not 1B context 🥱🥱🥱

super interesting, i will try this out! do you think there's any reason why openai, anthropic, or any of the big model providers wouldn't implement this architecture? there's always tradeoffs so i'm curious what the tradeoffs are here

A billion-token window sounds impressive, but retrieval quality is the real test. Feeding an entire codebase is easy; consistently identifying the five files that actually matter is where graph-based context could become genuinely useful.

No way, a billion token context window over an entire codebase is actually wild.

just double tapping this, really well done the videoraphy here. enjoyed the whole vid, till the end, and have pull it up multiple times since @daleverett, @daltonmeon, @damienhe 🔥🔥

yo thats really a movie type shii video. love it! congrats guys!

🔥🔥🔥

It's a managed postgres service + an extension.. Social media strategy excessive claims is making this look very 'snake oil'

@daleverett don't forget us in SG when you become famous!!

W

heyo @daleverett send it

W @daleverett

the word “context window” will only be around for another 1-2yrs

my coding agent got infinite context now!!!

let's GOOOO!!

Fellow vibe coders - do not waste your time learning context engineering

this is insane

Pretty cool

This is cool 😎

Cool