frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Show HN: BlaBlaBlAI – An open-source chat where LLMs are aware of each other

https://github.com/Transhumai/BlaBlaBlAI
3•ansorre•1h ago
Hi HN,

working with AI today involves a lot of copy and paste. Single LLMs are rarely enough to solve real problems, so we constantly switch contexts, ask multiple LLMs, and let them correct each other.

To break free from all that, I built BlaBlaBlAI and released it as Apache 2.0 open source.

The idea is simple: no more 1-on-1 chats with LLMs, but chats with as many LLMs and as many humans in the same chat as needed. Think WhatsApp groups, but with LLMs too, and a reasonable set of tools to coordinate them (rules, muting, cost attribution).

I can imagine how many types of businesses could be built on top of a chat system like this. Obviously, having worked on this app for some time, I have a thousand ideas. But today I’m interested in emphasizing just three things:

1) Using many LLMs in the same chat instead of one has been, for my productivity, more disruptive than the invention of LLMs themselves. This is why I’m sharing this.

2) Ideas like this tend to evolve in one of two ways: either they fragment into many mediocre clones, or they converge into one or a few strong open-source implementations. I’m explicitly aiming for the second, which is why, for me, open source is a technical necessity, regardless of any potential business built on top.

3) I’m used to the concept “Make something people want”, so this message is meant to understand whether people actually want BlaBlaBlAI. I know how much I want this (very much), so this is really about understanding whether BlaBlaBlAI will evolve just for me or for many.

That’s it. I also made a video to show an example of use. The use case is absolutely genuine and real, it happened to me. Who hasn’t received less-than-ideal responses from LLMs? In fact, that’s precisely the point of BlaBlaBlAI: there’s no single perfect LLM, but many LLMs working together can make a real difference.

Final details and notes:

- Open-source code on GitHub (the app works) - Multi-LLM chat - Markdown support in messages - Demo video showing real interactions between agents

Missing for now:

- Hosted version - Easy setup (but an exhaustive README.md to run it exists) - Onboarding

Links:

- GitHub repo: https://github.com/Transhumai/BlaBlaBlAI - Demo video: https://youtu.be/cYnIs_9p99c - Intro blog post: https://blablablai.com/blog/posts/welcome/ - Landing: https://blablablai.com/

I’ll be around all day to answer questions.

Thanks for reading.

Comments

DavideDL•1h ago
this is incredibly powerful
ansorre•16m ago
Thanks Davide!
Emarginated73•28m ago
Great stuff!!

It is definitely an idea discussed quite a bit on X (and in the broader AI community) as i read around.

The core argument is that no single LLM is truly best-in-class at absolutely everything and never can it be (in the short/medium term).

Different models have relative strengths (e.g., one excels at reasoning, another at creative writing, coding, fact-checking, or domain-specific tasks like DeFi or medical analysis).

So, routing tasks to specialized LLMs (or combining/ensembling them) often beats forcing one generalist "do-it-all" model to handle every step.

ansorre•15m ago
Exactly. The difference we’re exploring is collaboration instead of pure routing: multiple specialized LLMs sharing the same context and memory, reasoning together in real time.

Specialization becomes a strength once models stop working in isolation.

Trump Threatens to Invoke Insurrection Act in Response to Minnesota Protests

https://www.wsj.com/politics/policy/trump-minnesota-insurrection-act-98e11761
1•perihelions•48s ago•0 comments

Filetypes and Airports

https://airports.swtch.tel/
1•firloop•1m ago•0 comments

Can Wikipedia Survive?

https://www.ft.com/content/513761bb-3b6c-4b32-9931-a34f01047558
1•alephnerd•1m ago•0 comments

Snowtree: Review-Driven Safe AI Coding

https://www.bohutang.me/2026/01/10/snowtree-review-driven-safe-ai-coding/
1•xiaohanyu•1m ago•0 comments

Radxa AICore Ax-M1M.2 AI Acceleration Module

https://radxa.com/products/aicore/ax-m1/
1•yawniek•2m ago•0 comments

Vibe – Claude Skill to let Claude Code read screen automatically

https://github.com/Blurjp/vibe
1•blurjp•3m ago•1 comments

A tool to see what changed in your code (beyond the diff)

https://gitlab.com/hilly.bopper/nuky
1•nuky•3m ago•0 comments

Upcoming Changes to Spotify Premium Subscriptions

https://newsroom.spotify.com/2026-01-15/premium-pricing-update/
1•soheilpro•4m ago•0 comments

Can Others Explain My Work Without Me?

https://www.olafalders.com/2026/01/15/can-others-explain-my-work-without-me/
1•oalders•7m ago•0 comments

Investigating the methodological foundation of lesion network mapping

https://www.nature.com/articles/s41593-025-02196-7
1•carschno•7m ago•1 comments

Forcing TypeScript to Be Exhaustive

https://www.carlos-menezes.com/posts/exhaustive-checking-typescript
1•carlos-menezes•7m ago•0 comments

Show HN: Azurite – Triage Linear and GitHub notifications like a deck of cards

https://azurite-labs.vercel.app/
2•QaysHaji•7m ago•0 comments

Why there are no job posting when looked on portal

1•hwagh511•9m ago•0 comments

Self Sanitizing Door Handle

https://www.jamesdysonaward.org/en-US/2019/project/self-sanitizing-door-handle/
1•rendaw•10m ago•0 comments

The Curse of Genius

https://www.economist.com/1843/2019/04/29/the-curse-of-genius
2•andsoitis•11m ago•0 comments

Browser DevTools MCP – AI agents that can test and debug their own web code

https://medium.com/@serkan_ozal/empowering-ai-to-test-and-debug-its-own-code-introducing-browser-...
1•sozal•13m ago•1 comments

ICE agents do not have 'absolute immunity' from state prosecution

https://reason.com/2026/01/13/no-ice-agents-do-not-have-absolute-immunity-from-state-prosecution/
5•treetalker•13m ago•0 comments

Prediction Markets Turned Life into a Dystopian Gambling Experiment

https://www.theringer.com/2026/01/14/tech/prediction-markets-betting-explained-meaning-polymarket...
4•cainxinth•15m ago•0 comments

Nvidia Reportedly Ends GeForce RTX 5070 Ti Production, RTX 5060 Ti 16 GB Next

https://www.techpowerup.com/345224/nvidia-reportedly-ends-geforce-rtx-5070-ti-production-rtx-5060...
7•ndiddy•16m ago•1 comments

Show HN: I vibecoded a language server for web assembly

https://github.com/EmNudge/wat-lsp
1•emnudge•17m ago•0 comments

Chess Programming Wiki

https://www.chessprogramming.org/Main_Page
1•Antibabelic•18m ago•1 comments

'It's AI blackface': account hailed as Aboriginal Steve Irwin is AI character

https://www.theguardian.com/australia-news/2026/jan/15/aboriginal-steve-irwin-ai-character-create...
1•joejohnson•18m ago•0 comments

The Death of the Cool

https://chrismarino.substack.com/p/the-death-of-the-cool
2•jger15•19m ago•0 comments

How stagnant is CPU technology?

https://lemire.me/blog/2026/01/14/how-stagnant-is-cpu-technology/
1•ibobev•20m ago•0 comments

Poking holes into bytecode with peephole optimisations

https://xnacly.me/posts/2026/purple-garden-first-optimisations/
1•ibobev•21m ago•0 comments

SIMD Programming in Pure Rust

https://kerkour.com/introduction-rust-simd
4•randomint64•22m ago•0 comments

Show HN: MarkView – Markdown viewer with folder navigation and bookmarks

https://getmarkview.com/
1•dangtung•22m ago•1 comments

Go Home, Windows EXE, You're Drunk

https://gpfault.net/posts/drunk-exe.html
2•ibobev•22m ago•0 comments

Forgejo v14.0 Is Available

https://forgejo.org/2026-01-release-v14-0/
4•todsacerdoti•23m ago•0 comments

Why Senior Engineers Let Bad Projects Fail

https://lalitm.com/post/why-senior-engineers-let-bad-projects-fail/
2•lalitmaganti•23m ago•0 comments