Show HN: Maestro – A Framework to Orchestrate and Ground Competing AI Models
4 comments
·May 27, 2025Yusefmosiah
I’m building something similar. https://github.com/YusefMosiah/Choir.chat — if you email me at yusef@choir.chat I can invite you to the iOS TestFlight alpha. Happy to talk about in more detail as well.
Getting the UX to work well enough is a major challenge. I’m redesigning that currently, as I got negative feedback from early testers on my initial experimental UX. There’s a balance to be struck between giving users a low latency response, giving the models time to work together and call tools, and not overloading the user with too much information.
snappyleads
I also tried something similar - I called it https://supergo.ai .. However - I went with the approach to conclude with a final output.
defqon1
hi all, thanks for the attention
feel free to ask me anything you like. while at face value it seems to be a simple prompt aggregator and optimizer it goes far beyond that. consider it a meta-architecture for future synthetic intelligence and self-improving digital minds
peter_d_sherman
I think it's a great idea!
Even if other people/groups are doing something like this, either now and/or in the future, I don't think there ever would be or could be enough software tools which act as the intermediary between different AI's and perform different value-added functions relative to the AI's so interacted with.
The original author has named a few of these ideas so far, i.e., "The system compares their outputs and synthesizes them", "The feedback loop learns not only from right/wrong outputs, but from what kind of disagreements lead to deeper truth", etc., etc. -- but there are (or should!) exist many many more!
Point is, if there is value to be created in creating software abstractions on top of other software abstractions (for example, container orchestration software, which acts as a software abstraction on top of other software abstractions, i.e., containers A.K.A. "virtual(ized) machines"/"virtual software environments") -- then there's value to be created in creating software abstractions on top of multiple different AI's!
To the original author: I think you've got a good set of ideas going on, and I wish you much success with your ideas and software, both now and in the future!
(Now, I must temper all that I say (in the new age/era of AI) with a generalized reminder to all of the creators/implementors of AI, of Joel Spolsky's excellent essay on "Leaky Abstractions" -- https://www.joelonsoftware.com/2002/11/11/the-law-of-leaky-a... -- but I'd also like to temper that idea with the idea that all software is based on abstraction, and software abstractions can add a lot value (Microsoft Word anyone?) while leaking (failing to do what the user wants) very little, or potentially not at all... so there's a fine line between too much and too little abstraction, relative to what users may desire from their software and AI's...)
Anyway, generally speaking, I think it would be great to have some software which performs different value added tasks -- with multiple AI's!
ive spent the past few months designing a framework for orchestrating multiple large language models in parallel — not to choose the “best,” but to let them argue, mix their outputs, and preserve dissent structurally.
It’s called Maestro heres the whitepaper https://github.com/d3fq0n1/maestro-orchestrator (Narrative version here: https://defqon1.substack.com/p/maestro-a-framework-for-coher...)
Core ideas:
Prompts are dispatched to multiple LLMs (e.g., GPT-4, Claude, open-source models)
The system compares their outputs and synthesizes them
It never resolves into a single voice — it ends with a 66% rule: 2 votes for a primary output, 1 dissent preserved
Human critics and analog verifiers can be triggered for physical-world confirmation (when claims demand grounding)
The feedback loop learns not only from right/wrong outputs, but from what kind of disagreements lead to deeper truth
Maestro isn’t a product or API — it’s a proposal for an open, civic layer of synthetic intelligence. It’s designed for epistemic integrity and resistance to centralized control.
Would love thoughts, critiques, or collaborators.