Agentic Applications
Compute Infrastructure reinvented for long running agents, high throughput data workflows and good developer experience
Kubernetes, Serverless Platforms were built for running web applications, not agents which runs for 10x longer than web applications, queries more data sources, and are stateful between the steps.
We built a cluster scheduler from the ground up which can run up to 100000 functions concurrently. To build a multi-tenant serverless platform for running python applications.
Queues are irrelevant in Tensorlake, API calls made to agents and workflows are automatically queued in S3 while the cluster scales up to process them.
Every function runs in it’s own sandbox automatically, there is no need to use specialized sandbox apis to run LLM generated code.
It tracks call graphs of applications and workflows to restart from where they crash to provide automatic reliability.
Functions in applications can read and produce upto 5GB of output. Temporal and Vercel Workflows can handle upto a few MB of inputs and outputs.
How it works
Do more with Agentic Applications
Run Agent harnesses, code sandboxes and tools on the same infrastructure
Wake up functions every minute, month or year later.
Re-run failed requests in bulk without losing state of successful functions.
Stream progress from functions to users in real time
Complete control on the images software is run
Control concurrency for single function, and throughput of inflight request
Build more with serverless data applications:
Process pay stubs, tax forms and bank statements. Verify income automatically.
Ingest claim documents, pull key fields, route to the right workflow. Handle thousands.
Read through agreements, flag key clauses, extract terms.
