Skip to content

Infrastructure
for agentic
engineering.

Building with agents is fast. Deploying with LazyCloud is too.

One definition. Three ways to run it.

Develop locally

Step through your function with local inputs and your usual debugger.

api.pyBreakpoint
@app.endpoint(route="/count")
def count_words(text):
words = text.split()
return {"words": len(words)}
Variables Local
text"hello cloud"
words["hello", "cloud"]

Test in the cloud

Call the function on cloud compute and inspect its returned value.

count_wordsCompleted
0500 ms1 s
Prepare
Execute
Return
Resultdict
{"words": 2}

Deploy your app

Publish the function as an HTTPS endpoint for your application.

count_words / productionDeployed
Revision 07
Revision 08Current
POST/count
Handlercount_words
TransportHTTPS
200{"words": 2}

APIs

Ship an endpoint or a full ASGI app.

Deploy a Python function as an API, or bring your ASGI app with its routes and middleware. LazyCloud handles TLS, scaling, and idle shutdown.

In flight36
Utilization100%
Containers3
Request loadLast 60 seconds
/review18 active141 ms median
/generate12 active128 ms median
/status6 active105 ms median

Jobs and pipelines

Run the work behind your app.

Send tests, evals, data processing, and GPU jobs to the cloud. Connect tasks with dependencies and follow their logs and saved results.

Deployments / release_evals · productionReady · 0:00
Run progress0%current execution
Completed0/5dependency-aware
Elapsed0:0030s expected
Task timeline5 tasks · 0 running
checkoutfunction2 vCPU
testfunction8 vCPU
evaluatefunctionA100 · 40 GiB
security_scanfunctionL4 · 24 GiB
publishfunction1 vCPU

Background jobs and crons

Keep working after the request ends.

Submit background tasks or add a cron schedule to a function. Each run has retries, cancellation, and live logs.

Queued tasks27waiting
Processing4tasks running
Workers online2target 4 · max 8
Queue activitytasks left · workers right · last 60 seconds

Agent sandboxes

Run your agent's code in a sandbox.

Run code and tests in isolation. Control files, processes, ports, Docker, snapshots, and network access.

Processes
1 running
python agent.py · pid 12
Files
clean
/workspace · 24 files
Network
block-all
egress blocked · 0 attempts

Choose where your code runs.

Deploy without managing servers. LazyCloud provisions CPU capacity and scales down when idle.

LazyCloud

Deploy without managing servers. LazyCloud provisions CPU capacity and scales down when idle.

coding-agentSandbox
TaskAdd search to the API
Workspace
app/
api.py +8
search.py +32
tests/
test_search.py +24
Read filesapp/api.py
Edit sourceapp/search.py
Run testsuv run pytest
$ uv run pytest
tests/test_search.py ···

Your AWS account

Use CPU and GPU capacity in your AWS account. Manage deployments through LazyCloud.

product-apiEndpoints
Your AWS accountRequest activity
POST/search200
POST/chat200
GET/documents200

Your own machines

Connect your Linux servers, VMs, or GPU machines. Deploy through the same Python API.

model-traininggpu-server-01
Training loss0.24
Epoch06 / 10
0Training steps6,000
Checkpointepoch-05.safetensors

Export a typed package for your app.

Call deployed functions, endpoints, and ASGI routes from Python with autocomplete and type checking. ASGI route types come from OpenAPI.

Define

your app
review_app.pyPython
from lazycloud import App
from pydantic import BaseModel

app = App("review_app")

class Review(BaseModel):
    summary: str
    risks: list[str]

@app.endpoint(route="/review")
def review_patch(diff: str) -> Review:
    added = [line[1:] for line in diff.splitlines()
             if line.startswith("+") and not line.startswith("+++")]
    risks = ["Review added TODOs"] if any("TODO" in line for line in added) else []
    return Review(summary=f"{len(added)} added lines", risks=risks)

@app.function(retries=3)
def run_checks(commit_sha: str) -> dict[str, bool]:
    valid = len(commit_sha) == 40 and all(c in "0123456789abcdef" for c in commit_sha)
    return {"valid_commit_sha": valid}
$uv run lazycloud deploy review_app:app 2 workloads deployed

Generate

a typed package
$resolving
lazycloud_clients/review_apphashing manifest
Methods and modelsgenerating

Import

from another project
release.pypyproject.tomlyour codebase
1 
2 
3 
4 
5 
0 problemspython 3.12 · fully typed

Deploy your first workload.

Take the code you built locally and put it to work in the cloud. Deploy an API, launch a background job, or schedule its next run.