Harsh Kumar

Building software. Contributing to open source.
@screenpipe · Community @cognee_ · Maintainer @EpicenterHQ

[Harsh Kumar @thisisharsh7]

Meet Seeva: an AI assistant that sees your screen.

Press a key → It appears
Click "Watch Screen" → It sees what you see
Ask anything → Get help
Press a key → Gone

Works everywhere. Runs on your computer.

https://t.co/uK5TQoM52i

[Harsh Kumar @thisisharsh7]

Got accepted into Claude for open source program by @AnthropicAI.

Back to building🚀

[Vasilije @tricalt]

After Karp's interview, everyone's talking about how using @OpenAI and @AnthropicAI models means giving away your IP.

At @cognee_ we believe you should own your context!

That's why we built COGX: an open format that makes your memory portable and yours. Import it, export it, move it anywhere. No lock-in!

[Linghua Jin 🥥 🌴 @LinghuaJ]

This is huge!! super excited about it. Meet 𝐂𝐞𝐫𝐞𝐛𝐫𝐚𝐬 𝐊𝐧𝐨𝐰𝐥𝐞𝐝𝐠𝐞: one of the most widely adopted internal tools used by humans, automations, and agents, serving 15𝐤 𝐪𝐮𝐞𝐬𝐭𝐢𝐨𝐧𝐬 𝐝𝐚𝐢𝐥𝐲. It establishes a practical system for context engineering from a massive, fast-growing, dynamic enterprise corpus that actually works in production for agents.

In a real enterprise setting, information is generated wherever it is convenient and ergonomic, and agents need to meet data where it lives: Slack, Codebase , Docs, Jira, and internal information that doesn’t live in any of these platforms. Different platforms are tailored for specific domains, optimized through years of product engineering.

This article is an end-to-end guide covering how to ingest from different sources, reconcile them, and put them in the right order with the right amount of precision and recall, and continuously surfacing fresh context that works best for agents.

Every enterprise is trying to figure out how to build a knowledge base for agents these days, and this serves as an instrumental and inspiring guide.

We are big fans of @cerebras and this project, and @cocoindex_io is super honored to support this effort. Let’s go!

[screenpipe (YC S26) @screenpipe]

introducing screenpipe: it records and learns how you work and turns it into a searchable memory, SOPs, and AI agents

open source, local-first, 20K+ GitHub stars, 1,900+ forks, and 130+ contributors

[Paul Graham @paulg]

If you want to start a startup, don't learn "entrepreneurship." Learn how to build things. The hard part of startups is not "entrepreneurship" but product: to know what to build, and to be able to build it.

[Vasilije @tricalt]

Introducing Cognee v1.0: a major breakthrough in agentic intelligence.

It is 145% better than Opus 4.8 and GPT 5.5 at long context memory retrieval.

Cognee allows a 100 BILLION token context window 100,000x more than Claude. It's:

- 6.9x cheaper than GPT 5.5 and Opus 4.8
- Cold starts in 350ms & searches in 260ms

Why this matters:

Today agents forget important context, redo tasks, waste tokens, and slow down as workflows get more complex.

Cognee solves this.

It’s not a place to build agents. It connects to the agents you’ve already built, across any platform, and makes them significantly cheaper, faster, and more accurate.

Here's how it works:

[louis030195 | screenpipe (YC S26) @louis030195]

I just joined Y Combinator Summer 26. I built 200+ open source projects before (without AI), here's my story:

[Harsh Kumar @thisisharsh7]

@louis030195 really inspiring story Louis. glad to be contributing to screenpipe and excited for what's ahead

[Shane Parrish @shaneparrish]

Former podcast guest @kunalb11 named CEO of What's App.

https://t.co/ehIbD3hv66

[Harsh Kumar @thisisharsh7]

@owenthcarey https://t.co/GsTqHmJkil

[Harsh Kumar @thisisharsh7]

@rabanspiegel @David_knp @j_schwxrz @arnestrickmann @miaugladiator1 🔥

[Harsh Kumar @thisisharsh7]

@louis030195 @screenpipe huge congrats louis🔥🎉

[South Park Commons India @spc_india]

Apply to Curiosity 2026 — our exclusive showcase for India's most ambitious deep-tech builders. Compete for a ₹1.25 Cr+ prize pool (powered by @artparkindia) + lots more in credits!

Finalists will demo, science-faire style, before investors, judges & institutional leaders.

Themes: Aerospace , Defence & Robotics, Photonics/Semis, Climate & Energy, Biotech & Medical, Physical AI + more. To apply, you need: a working build/product, prototype, or scaled-down model.

We're selecting teams on a rolling basis.
Deadline: 20th May.

[Harsh Kumar @thisisharsh7]

@jahooma This is awesome.🔥

[Harsh Kumar @thisisharsh7]

@lucaronin @tolariamd 🔥🔥

[Harsh Kumar @thisisharsh7]

@alexdahl145 bruh🔥

[Vasilije @tricalt]

We made Codex remember what your Claude Code did!

@cognee_ now lets you see and share Agent sessions and permanent memory

Showed this to a large audience in NYC,

We had over 1000 people applied to join the event

Check out the video 👇

[Harsh Kumar @thisisharsh7]

@manufact congrats!

[Harsh Kumar @thisisharsh7]

@alexdahl145 @tldraw @duckdb 🔥

[Harsh Kumar @thisisharsh7]

@sama In codex I'd like the option to permanently delete chats (not just archive them).