What Modal does
Modal is a serverless compute platform for deploying models, parallel data jobs, GPU training, notebooks, web endpoints, and isolated agent sandboxes from code.
Modal replaces much of the cluster and container orchestration around compute-heavy Python. Functions can request CPUs, memory, GPUs, images, secrets, volumes, queues, schedules, web endpoints, or sandboxes. Scale-to-zero makes it attractive for uneven inference, media processing, fine-tuning, evaluation, and parallel data work, while notebooks and endpoints cover interactive and production use.
Billing is based on requested or consumed resources, generally measured by the second, plus any subscription or storage charges. A payment method is required for continued use, and exact GPU, CPU, memory, disk, sandbox, and regional rates change by resource. Teams should test cold starts, concurrency, idle behavior, egress, retries, and peak cost with a representative workload rather than extrapolating from one function call.
Modal documents encrypted transport and storage, secrets, restricted environments, RBAC on Team and Enterprise, SOC 2 Type II controls, and prompt deletion after processing for covered services. Public endpoints and tunnels still need application authentication. Sandboxes reduce isolation risk but do not make arbitrary code safe; enforce network, time, CPU, memory, filesystem, secret, and output limits, and require human approval before consequential actions.
How Modal works
Developers describe an application, container image, resources, secrets, storage, and functions in Python, then authenticate and run or deploy it with the Modal client. Modal builds the environment, schedules isolated containers across available cloud capacity, scales concurrent work, and exposes jobs or web endpoints. Inputs can arrive as function arguments, files, queues, or HTTP requests; outputs return to the caller or approved storage. Operators must constrain resources, validate generated code and model output, and monitor failures and spend.
How to set up Modal
Choose a bounded workload
Start with one inference or batch job, expected inputs, latency target, accelerator, concurrency, output, and monthly cost ceiling.
Install and authenticate safely
Create separate development and production workspaces or environments, install the official client, and keep tokens out of source control.
Define a reproducible image
Pin system packages, Python dependencies, model revisions, resource requests, timeouts, volumes, and secrets in code.
Load-test failure paths
Measure cold starts, queueing, retries, cancellation, malformed inputs, GPU availability, output validation, and cost under realistic concurrency.
Deploy with controls
Add endpoint authentication, budgets, least privilege, logs, alerts, rollback, and manual approval around generated code or high-impact actions.
Modal FAQs
Does Modal require Kubernetes?
No. Modal manages scheduling and container infrastructure; developers define resources and environments in code.
How is Modal priced?
Compute is usage-based by resource and time, with plan and storage differences. Check the live pricing page for the chosen GPU and region.
Can Modal scale to zero?
Yes, many functions and endpoints can scale down when idle, although configuration and cold-start tradeoffs matter.
Can Modal run untrusted agent code?
Modal offers isolated Sandboxes, but customers must still restrict credentials, network access, resources, duration, and allowed outputs.
Is JavaScript supported?
Python is the primary authoring language; JavaScript, TypeScript, and Go clients can invoke functions and manage supported resources.
Listing reviewed 2026-07-30. Product details and pricing can change; verify important terms on the provider's website.
Related Coding AI tools
Related AI guides
Reviews
Tell the community what you made, what worked, and what you wish you knew before starting.