Graperoot

"We have got 95%+ benchmark on longMemEval by using a memory layer on long-running agents." Hey Everyone, we created a memory infrastructure for long-running agents, and it beats mem0 and other big tools on LongMemEval benchmarks. If you are working on long-running agents, I would love to get feedback or suggestions over DM Website: swafra.vercel.app Github: lnkd.in/gz6HbQtU

@graperootdev link: graperoot.dev already used by ~5k users across 90+ countries! without any paid marketing, just the real solution to the real problem.

switched from claude max to claude pro same output quality @graperootdev did that

After spending $15k in 4 months we finally pushed to create GRAPEROOT on our github dashboard :) @graperootdev has saved over $250k for almost 200 developers(this is only number who have opted in for telemetry). Github : github.com/kunal12203/cod… You can see AI Tag LoL🙂

Delhi isn't exactly Silicon Valley. but last weekend in a cafe, met someone running Claude Code, walked up and pitched @graperootdev he already knew it. his whole team was using it. imagine Bengaluru, SF 🥰 this is what 5,000 users feels like in real life. @krishna37930189

800 daily users is big number! I got to know when my cofounder @ShivGRoot randomly met a person in a cafe at Delhi. He was using Claude code, then shiv went to pitch our product lol (GTM things) and he turned out our active user. That was really crazy!!

GrapeRoot builds a local context graph so AI agents retrieve only what they need, not your entire codebase. better context → better outputs, lower token waste. graperoot.dev

3 months ago: "what if AI agents stopped rereading the same codebase?" today: ~5,000 devs · 90+ countries · $220K saved AI coding isn't model-constrained anymore. it's context-constrained.

Water level live on: graperoot.dev/leaderboard

main website: graperoot.dev github(Open source): github.com/kunal12203/cod…

We started this as a cost saving mission for people using Claude code but we literally saved almost a reservoir of water 38 million Litres!!!!

Fable 5 released but we are still saving upto 80% of cost using @graperootdev because model can become more smarter but at any point they need context to work on, without memory of your codebase or documents it is still a stateless model. That’s what we solve at lowest cost possible.

Was exploring multiple things in last few days and now back to complete what we started @graperootdev Let’s goo! Releasing Biggest and efficient context layer for any CodeBase and production setups very soon

Star the repo at github.com/kunal12203/cod… Let’s complete that 1k goal :)

Developers saved almost a reservoir of water using @graperootdev LoL Calculation: 0.1mL of water needed for 1 token and we have saved 370B tokens We have reached 60k pip installs and developers are loving it, if you want to try out go at: graperoot.dev

What do you actually do to debug any issue?

I let @graperootdev fix some hard issues on Open Source Go repos live on Github and surprisingly, it uses $0 to find the issue and context related to that and used Claude to fix that without using that extra exploration cost. #graperoot

@ycombinator @FrancisPSantora And we're venture firm :)

@FrancisPSantora we're also a software company!

YC now has their own Docsend clone. Just tried it. Works great. Every YC company saves thousands.

I saved 80$ per month using this in Claude Code. Tool: grape-root.vercel.app After tracking token usage I noticed most tokens weren’t used for reasoning they were used for re-reading the same repo files on follow-up turns. Added a small context routing layer so the agent remembers what it already touched. Result: about $80/month saved in Claude Code usage. Honestly felt like I was using Claude Max while still on Pro. Try yourself and thank me later!