so that they can collaborate, trade and negotiate.
Example: “Should I invest in NVIDIA tomorrow?”
Imagine you want a collaborative result not a single agent/team output.
You spin up *5 different AI agents*, each running in a different system, diffrnet auth and paywall:
- One langchain agent reads *NVIDIA’s latest earnings & presentations*
- One agno agent analyzes *competitors* (AMD, Intel, etc.)
- One crew agent reads *market & macro reports*
- One openai agent tracks *recent news & filings*
- One adk agent combines everything and gives a final recommendation
Today, connecting this is messy.
Each agent is a script. Every connection is custom glue code.
## What Bindu does here
With Bindu:
- Each agent gets a *simple URL*
- Agents can *call each other directly*
- The final “decision agent” just calls the other four
- No framework lock-in, no custom wiring
- A common context - all the agents can share.
That’s it.
## So what is Bindu?
*Bindu makes AI agents behave like small services.*
Once an agent is on Bindu:
- it can be called like an API
- other agents can use it
- you can reuse it across projects
- you don’t care where or how it’s running
Agents stop being isolated scripts and start becoming building blocks.
## Why we built it
While building agent-based products, 278 difrrent frameworks we kept hitting the same wall:
Agents are getting smarter, but *they don’t work together easily*.
We didn’t want another agent framework.
We wanted a simple way to connect agents that already exist.
So Bindu focuses on one thing:
*making agents easy to connect and reuse.*
If you’re building multi-agent systems and feel like you’re rewriting the same wiring over and over, I’d love to hear your thoughts.
I'm not too excited by Phi-4 benchmark results - It is#BenchmarkInflation.
Microsoft Research just dropped Phi-4 14B, an open-source model that’s turning heads. It claims to rival Llama 3.3 70B with a fraction of the parameters — 5x fewer, to be exact.
What’s the secret? Synthetic data.
-> Higher quality, Less misinformation, More diversity
But the Phi models always have great benchmark scores, but they always disappoint me in real-world use cases.
Phi series is famous for to be trained on benchmarks.
I tried again with the hashtag#phi4 through Ollama - but its not satisfactory.
To me, at the moment - IFEval is the most important llm benchmark.
But look the smart business strategy of Microsoft:
have unlimited access to gpt-4
the input prompt it to generate 30B tokens
train a 1B parameter model
call it phi-1
show benchmarks beating models 10x the size
never release the data
never detail how to generate the data( this time they told in very high level)
claim victory over small models