Kaustubh

Engineering @AgnoAgi • Built India’s first AI powered vending machine • Research @AucklandUni • Open source ML ❤️ • CS @IIITRanchi '24

[Kaustubh @kaus_mos]

*Stares at the camera

[Kaustubh @kaus_mos]

@HelloVyom @pritisinghhhh @AgnoAgi @smallest_AI @harshitajain561 Lol no just a tool integration

[Kaustubh @kaus_mos]

@pritisinghhhh @AgnoAgi @smallest_AI @harshitajain561 Oh hello? I merged it. Need cholla bhatoora too 😢

[smallest.ai @smallest_AI]

Smallest AI is now in @AgnoAgi

Give any Agno agent a real voice: broadcast-quality speech via our Text-to-Speech models - Lightning v3.1 and Lightning v3.1 Pro - in three lines of code. Install @AgnoAgi and get started:

pip install agno

Get building agents that talk back!

[Ashpreet Bedi @ashpreetbedi]

Introducing the Pro Agent Builders series. Inside information on how experts are building agents.

First post: automating the grueling work of improving agents. Prompt engineering, tool construction, testing. The work that eats your evenings is now an overnight loop that runs without supervision.

Recursive auto-improvement (RAI) is a convergent process where a coding agent turns your agent's spec and usage data into simulations, and fixes the agent until every probe passes.

My favorite version is the overnight run. I'll kick off 300-500 probes, go to bed, wake up to a better agent.

I rarely deploy an agent without running this loop first.

[shafu @shafu0x]

if it touches money you absolutely need to read the code

[Ashpreet Bedi @ashpreetbedi]

New 📽️: Watch me build an agent platform using only coding agents 👇

Build agents, improve them, deploy them, secure them, connect them to your AI Apps... without opening your editor

[the tiny corp @__tinygrad__]

The limiting factor of software engineering has never been code production. It is complexity management.

[Ashpreet Bedi @ashpreetbedi]

Agno is proudly open-source and co-signs the Open Weights and American AI Leadership letter.

Open weights, Open models, Open platforms

https://t.co/3Oyi29O8Cm

[Guillermo Rauch @rauchg]

Vercel proudly co-signs the Open Weights and American AI Leadership letter.

Open source, data, protocols & research enable the technological wonders we enjoy every day, enriching our lives and our world.

Open weights are the logical next frontier: https://t.co/SMeMS2lINu

[Ashpreet Bedi @ashpreetbedi]

Most people go to the pub on a friday, ashpreet's building a fucking FILESYSTEM for agents

https://t.co/kDD7imHYQJ

(not for use in production)

[jack @jack]

the government of india does not like technologies like bitchat and wants it taken down

[Kaustubh @kaus_mos]

@emperoryp @Meta @Spotify 😂😂

[Kaustubh @kaus_mos]

@monali_dambre That somehow makes perfect sense!

[Monali @monali_dambre]

https://t.co/f7x2MsLqHX

[Kaustubh @kaus_mos]

@dpradhanbjp @RahulGandhi @INCIndia Thodi sharam karle

[Ashpreet Bedi @ashpreetbedi]

Introducing Agno AgentOS, a FastAPI application for serving agents as an API, an MCP server, and through chat apps like Slack, Telegram and WhatsApp.

AgentOS covers the valley of death between an agent definition and a live service. It gives your agents a durable runtime, multi-user security with RBAC, background execution, checkpointing, session management, tracing, evals, guardrails, and more.

In the video below, a single AgentOS is running multiple parallel streams (resumable across connection drops), serving multiple MCP clients, and handling multiple processes reading and writing the same data.

A true work of performance art, AgentOS is the perfect backend for any agentic application.

The best part: you can get your own with one prompt. Go to https://t.co/oooZJm0u42, pick your cloud, and hand the 2 line prompt to your favorite coding agent.

Free and Open Source. Enjoy!

[Priti @pritisinghhhh]

The India we grew up reading about - Inquilab Zindabad!

[Ashpreet Bedi @ashpreetbedi]

Introducing Agno Environments, our first step towards RL for agents.

Today we're releasing part 1 of our most requested feature: verification and data generation. Run your agent K times, score every attempt, and export the ones that passed as a fine-tuning dataset.

Up next: fine-tune on the dataset, then re-run to measure gain.

I'm open-sourcing everything, including the first set of 70+ code examples: https://t.co/snx7UBDzPM

Agents → Environments → Fine-tuned Model → Better Agents. One step down, two more to go.

[Ashpreet Bedi @ashpreetbedi]

https://t.co/nGIFfPPBgn