Launched this week
Meshia
Your autonomous research lab
17 followers
Your autonomous research lab
17 followers
Meshia brings agents, your team’s knowledge, and compute into one shared workspace. Manage experiments, build on teammates’ work, and recursively improve toward your goals. Find compute from 17+ providers worldwide, combine GPUs and TPUs, and keep your files and context as your research grows.
Free Options
Launch Team / Built With
LoomHit record, hand off to AI
Promoted
Hi guys, I’m Luke!
Meshia is a research lab I’ve originally built for myself to enable massive agent orchestration and use resources anywhere to research and keep iterating for discovery. It packages together a bunch of things I used to do for AI Labs I worked at.
We’ve went from 0 to $100k in revenue in the last 4 weeks, and I’m excited to more publicly share the platform with researchers.
Getting a bunch of agents to work together is really hard (just like getting a bunch of people working together). Meshia broadly tackles context management, communication, trust, and direction, allowing 1000s of agents to work together efficiently. Frontier models automatically audit each other’s work, and the workplace is designed to surface only the most important live information.
The best research usually requires additional compute resources (for training, inference and simulation). Meshia is able to fuse together resources from different providers into the same workplane, provisioning avalible compute geographically close to each other to reduce latency. You can have GPUs on AWS running simulations and feeding directly into TPUs on GCP for training.
Meshia works well at scale, but I’ve found the environment also became my personal go-to for smaller scale experiments I’m curious about. Having a structured research process built in, combined with the cooperative intelliegence from different kinds of models makes the ideas Meshia comes up with much more unique compared to any one model alone.
Everyone can try out Meshia for free with enough credits to run a small fine-tuning run on a couple GPUs. You can connect your existing ChatGPT, Claude or Gemini subscriptions right in the workspace, and control research projects via MCP (but did a lot of work on the workspace itself so feel free to check that out!)
This is early and I'm a one man team so expect some bugs, but in return, I’ll personally be available 24/7 to help fix problems and take your research to the next level! I hope the experience feels magical! Let me know what you think + any feedback you have!
———
Wanted to highlight some of the things researchers in SF have been using Meshia for:
(1) AI lab exploring more efficient distributed training architectures
(2) Training speech to speech models from scratch at a startup building voice interfaces
(3) Lab machine control at UCSF
——
Happy discovering ❤️🚀!
luke@meshia.io
Mixing GPUs and TPUs from 17 providers under one workspace is what would actually make a side experiment affordable, instead of me keeping three cloud bills just to compare two training runs.