Your Modal functions fail quietly. Polylane notices, investigates, and tells you why.
Modal runs your functions, sandboxes, and GPU jobs on demand. Polylane maps the whole workspace and watches it against its own baselines.
Join the waitlistEvery supported resource.
The Modal resource types that show up in your graph.
Environment App Function Class Sandbox Image Volume Secret Queue Dict Compute that vanishes, history that doesn't.
Agents map the workspace, watch its signals, and investigate what breaks.
The workspace in one graph
Environments, apps, functions, classes, images, sandboxes, volumes, secrets, queues, and dicts, each with its edges, deploy provenance, and history, next to everything else you run.
Endpoints linked to their callers
Polylane matches modal.run URLs in env vars across your other clouds, so a service calling a Modal web endpoint is a real edge in the graph, not tribal knowledge.
The failure signal Modal does have
Modal reports no error rate, and code that raises still exits as a success. Agents watch stderr volume across your recently active apps instead, judged against how each app normally behaves.
Spend, watched by the hour
Metered spend across the workspace, summed per hour and judged against normal. A looping function or a container that never exits shows up long before the invoice does.
Cold starts and backlogs
Sandbox startup time and queue depth, judged against the workspace's own recent history. When producers outpace consumers, you find out before the backlog does the telling.
A token pair the agent never sees
Modal connects with a workspace token: an ID and a secret from your Modal settings. Both are encrypted before they're stored and the agent never sees them. Code changes land as pull requests.
Questions.
Is Polylane an AI SRE for Modal?
It's what most people mean by the term: agents that keep your apps, functions, sandboxes and queues in one context graph, watch them, investigate issues and fix what they can, with a human approving anything that changes production. We call it self-operating software rather than an AI SRE: the point is that the app looks after itself, not that a bot joins the on-call rotation.
How does Polylane connect to my Modal workspace?
With a Modal API token. Create one in your Modal workspace settings, then paste the token ID and token secret. Both are encrypted before they're stored, and the agent never sees them.
What Modal resources does Polylane sync?
Environments, apps, functions, classes, images, sandboxes, volumes, secrets, queues, and dicts, each with its edges and history in the graph. The full list is on this page.
Can Polylane change my Modal apps?
Agents investigate and explain. Code changes only ever land as pull requests on the repo behind the app, and any write action pauses for your approval first, with the exact request and reason on screen.
How does detection work without alert rules?
Agents watch the signals Modal actually exposes, judged against the workspace's own baselines: stderr log volume across active apps, metered spend per hour, sandbox startup time, and queue depth. Modal doesn't expose error rates, GPU utilization, or per-cron metrics, so Polylane doesn't pretend to read them.
Does it work with my observability provider?
Yes. Datadog, Sentry, Honeycomb, Axiom, and Grafana Cloud join the graph as first-class sources, and their alerts get triaged by an agent the moment they fire.
What does it cost?
Pricing is public and self-serve: a free plan, paid plans from $80 a month, and pay-as-you-go overage past the included credits. Every plan and limit is published at polylane.com/pricing.