Compile English specs into 22 MB neural functions that run locally
22 MB compiled neural functions run locally — no API calls after compilation.

Map any Python function to 1000 VMs in <1 second, automatic env sync.
Data scientists, ML engineers, biotech/geospatial researchers doing batch computing
Ray · Dask · Modal
Given: a function, and a list of inputs; `remote_parallel_map` runs that function on every input in the list in parallel, each in a separate vm in the cloud, then returns a list of results. It can run any Python function on 1,000 vm's in less than 1 second, works with any hardware, and any custom container. Remote tracebacks/stdout appear locally, and your python env is automatically (very quickly) cloned on remote machines.
Cloud-beginners use Burla to turn simple scripts into pipelines that handle many terabytes of data, or perform difficult computation, without help from other engineering teams.
We have users in Biotech, Geospatial, and Data-Science/ML building pipelines with 20+ `remote_parallel_map` calls in them to fan different parts of the program across many machines, then combine results. Traditionally this required help from more complex software, other engineering teams, or both.
We created Burla so that anyone, even complete beginners, can create programs that handle terabytes of data, perform significant computation, or run in the cloud in the background for days with an easy way to monitor them (dashboard is open-source too).
22 MB compiled neural functions run locally — no API calls after compilation.
PyPI package with zero description — no README, no docs, no idea what it does.
91 reaction templates plus process engineering in pure Python, eliminating RDKit's conda nightmare.
Running 15 concurrent agents without burning through API limits faster than CrewAI or AutoGen.
Win98 nostalgia meets functional remote browser, CORS-free angle is clever.
First library to resolve Python function singularities algebraically without symbolic expressions.