Coiled review

Coiled is a hosted cloud-compute service with a Python package and web interface. It provisions virtual machines in your AWS, GCP or Azure account, then installs your code environment on those machines. Your code can run on the machines as a Dask cluster or as batch jobs; Coiled says batch jobs can run non-Python code too.

In a typical stack, Coiled sits between a Python application or data workflow and the cloud provider’s compute layer. You request hardware from Python, send work to it, and use the web UI to debug and analyze jobs. The service handles provisioning and cleanup rather than requiring your team to manage machines directly. Its pitch is less infrastructure work without moving compute into a vendor-owned cloud by default. For people starting out, Coiled also offers a quickstart that can run in its cloud account.

The named integration surface is Coiled’s Python package, installed with pip install coiled. The homepage shows a Python API that creates a coiled.Cluster, sets worker counts, regions and VM types, and submits functions through a Dask client. Coiled’s quickstart is invoked with coiled quickstart; the homepage also shows shell-based batch jobs configured through comments such as #COILED memory 32GB and environment variables.

Environment copying is intended to run code remotely without requiring Docker. The homepage names Dask, dataframes, Xarray and Prefect as tools on the platform, and says batch jobs can run code such as Fortran. A web UI provides job debugging and analysis. Python is the named API language; confirm SDK support for other languages and any API or CLI details in Coiled’s documentation before designing a non-Python integration. The homepage does not establish a REST API or webhook workflow.

The Free plan costs $0 per month and includes one seat, one workspace and $25 of monthly usage, with package sync or Docker. That is useful for experiments, but a team needing more seats or usage must move to a paid plan. Basic costs $100 per month for two seats, one workspace and $100 of monthly usage. Professional costs $500 per month for six seats, three workspaces and $500 of monthly usage, plus Custom Networking and Private PyPI, Spot and ARM support.

Basic and Professional can buy additional usage at $0.05 per CPU-hour. GPU usage is billed at the listed hourly hardware rates, from $0.15 per hour for Nvidia one thread to $1.00 per hour for A100. Cloud-provider compute is billed separately by AWS, GCP or Azure, so Coiled’s plan price is not the full infrastructure bill. Enterprise pricing is custom and includes usage volume discounts, SSO, a custom S3 bucket and AMI, and a Field Engineer. Monthly prices are listed; annual billing is not published.

Coiled runs its managed service in your cloud account, but that is not the same as self-hosting Coiled’s control plane. The vendor doesn’t publish the software licence or whether the service can be self-hosted. It says compute runs in your AWS, GCP or Azure environment, giving teams control over their data and cloud costs.

The Python-facing workflow uses a Coiled package and service to provision machines, copy environments and manage jobs. That creates some dependency on Coiled’s orchestration, even though the compute is in your account. The vendor doesn’t publish a data-export format or migration guide. Before adopting it, ask how to retrieve job metadata and logs, and what steps are needed to replace Coiled’s provisioning and cleanup with your own cloud tooling.

Coiled’s approach removes some machine-management work, but production teams still need to budget for cloud-provider compute separately from the plan. Additional CPU usage on Basic and Professional is priced at $0.05 per CPU-hour, and GPU usage has separate hourly rates. Model a full workload bill with your provider’s compute charges included, and confirm how usage is measured for your workloads.

The vendor doesn’t publish API rate limits or uptime terms on the pages reviewed. Ask about service availability, support response commitments, scaling ceilings and recovery behavior before making Coiled a dependency for scheduled pipelines. The vendor says machines are cleaned up when work is done and that compute runs in your account. Confirm how cleanup behaves after failures, and review network access, identity controls, data handling and any compliance requirements with Coiled. Professional includes Custom Networking and Private PyPI; Enterprise adds SSO and custom S3 bucket and AMI options.