Last updated: 2026-10-19
Godela, a Y Combinator Spring 2025 company, trains a physics-constrained AI model on an engineer's own simulation data and answers what-if questions in seconds. Its homepage demo reports 0.8 second inference on a rack cooling dataset of 2,400 simulations. Access is by pilot request, and no plan list is published.
About Godela
Godela is an AI physics engine from a San Francisco company in Y Combinator's Spring 2025 batch. It is aimed at engineers who wait days or weeks for simulation runs and physical prototype tests, and it is pitched as a faster, cheaper stand-in for both. The two founders named in the YC launch post, Cinnamon Sipper and Abhijit Pamarty, describe themselves as former engineers at Apple, Google and Intel who are turning Stanford research into a product.
The homepage describes a three-step loop. In Train, you upload simulation data and Godela fits a physics-constrained model to the behaviour of your system. In Explore, you ask what-if questions in plain language, such as raising airflow in one zone, and get a prediction without queuing a new solver job. In Optimize, the model searches for a better configuration. The YC post adds that CAD files and experimental results can be uploaded beside simulation data, and says the aim is to give existing simulations an adaptive home, although the one-line pitch talks about replacing them.
Five application areas are listed: data center thermals, electronics cooling, aerodynamics, electromagnetics and structures, plus a build-your-own route for other physical systems. The worked example on the homepage is a rack cooling study with 2,400 simulations and 12 parameters, answered in 0.8 seconds. It is a scripted demo, not a published benchmark.
On HokAI it has no direct peer, so the nearest neighbours are science-model products. Atmo applies machine learning to weather forecasting, Tamarind Bio serves published biology models through a web app and API, and Gemini for Science targets research workflows. A team that would rather build its own surrogate model needs GPU hosting such as Modal and a model hub such as Hugging Face, which is the do-it-yourself route Godela is meant to shorten.
Access is by pilot request through a booking link. The pages read for this entry show no pricing, documentation, security page or customer case studies, so accuracy against a conventional solver, supported file formats and data handling stay unanswered until a pilot. Treat the phrase simulation-quality as the vendor's own claim.
Pricing
No public pricing. Access is through a pilot request on the vendor's site, and the pages read list no plans, rates or free tier.
Key Features
- Training on your own simulation data: Fits a physics-constrained model to a system's behaviour from uploaded simulation runs, per the homepage.
- Plain-language what-if queries: Takes questions such as changing airflow in one zone and returns a prediction instead of queuing a fresh solver job.
- Configuration search: Looks for a better design and ranks candidates, as the homepage cooling demo shows.
- Mixed engineering inputs: Accepts CAD geometry and experimental measurements alongside solver output, according to the YC launch post.
- Five listed physics domains: Lists thermal, flow, electromagnetic and structural domains, plus a build-your-own option for other systems.
Pros
- Trains on a customer own simulation data, so predictions reflect their geometry and operating conditions rather than a generic model.
- Thermal, flow, electromagnetic and structural problems are listed under one product instead of one tool per domain.
- The founders come from hardware engineering at Apple, Google and Intel per the YC launch post, and the company is Y Combinator backed.
Cons
- No public pricing, documentation or trial: access starts with a pilot request.
- Accuracy against conventional solvers is a vendor claim, and no error bound or validation study appears on the pages read.
- The homepage workflow starts from uploaded simulation data, so teams with no solver history may need another starting point.
- It is an early-stage company with a team of 10 on the YC directory, so support depth is unproven.
Frequently Asked Questions
Does Godela publish a price?
No. The pages read for this entry list no plans, rates or free tier, and the only call to action is a pilot request through a booking link. Cost is settled in conversation with the vendor.
Can an engineer try Godela without a sales call?
Not from what the site shows. There is no self-serve signup, sandbox or public documentation, so the first step is requesting a pilot.
What on HokAI is closest to Godela?
Nothing on HokAI is a direct match for engineering simulation. Atmo applies machine learning to weather forecasting and Tamarind Bio serves science models through an API, so they are the nearest in kind. Teams building their own surrogate models can host them on Modal and start from Hugging Face.
How does Godela differ from running a conventional solver?
The vendor says it learns your system from existing simulation data and then answers new questions in seconds instead of days. That is a speed claim for a trained surrogate model, so accuracy outside the training data is the thing to test in a pilot.
What data does Godela need to get started?
The homepage starts from simulation data from your own system, and the YC launch post adds CAD files and experimental results. File formats and size limits are not published, so ask before preparing data.
Top Alternatives
- Atmo: Pick Atmo if the physical problem is weather forecasting; choose Godela for thermal, flow, electromagnetic or structural design questions.
- Tamarind Bio: Pick Tamarind Bio for running published biology models; choose Godela when you need a model trained on your own engineering simulations.
- Hugging Face: Pick Hugging Face if your team will build and host its own surrogate model; choose Godela if you want that training handled for you.
- Modal: Pick Modal when you have ML engineers who want serverless GPUs for a custom physics model; choose Godela for a packaged workflow.