Caution is backed by Y Combinator. Learn more.

Secure hosting

Take entire classes of attacks off the table. Deploy sensitive software in minutes with a familiar Git workflow, on your infrastructureinfra or ours.Take entire classes of attacks off the table. Deploy in minutes with a familiar Git workflow, on your infra or ours.

Trusted by

Protect workloads. Verify what runs.

Encrypt data in use

Protect sensitive data during computation inside trusted execution environments (TEEs).

Detect deployment tampering

Verify production matches reviewed source and build inputs through reproducible builds.

Deploy in minutes

Deploy to confidential computing in minutes with Git, without building your own platform.

How it works

Deploy in minutes

From source to a verifiable enclave in minutes. Watch this video for a full platform walkthrough.

1

Initialize

Run caution init to create a caution.hcl and other data required for the application.

$ caution init
Initializing the app...
2

Deploy

Push with git push caution main. Caution builds a reproducible enclave image and provisions the TEE.

$ git push caution main
Building enclave image...
Deployed to enclave
3

Verify

Run caution verify to rebuild the image, compare hashes, and confirm exactly what the enclave is running.

$ caution verify
Collecting source code...
Rebuilding image...
Comparing hashes...
Attestation verification PASSED

Verify the layers confidential compute leaves opaque

Regular cloud

Trust the operator. Production stays a black box.

Confidential compute

Trust the hardware. The software stack still stays opaque.

  • Application
  • Runtime
  • OS / Kernel
  • Hardware / TEE
  • Compiler / Toolchain

Verifiable compute

Don't trust. Verify every layer yourself, down to the compiler.

  • Application
  • Runtime
  • OS / Kernel
  • Compiler / Toolchain
Use cases

How teams use Caution

For sensitive workloads where integrity failures would be costly—from custody and signing to confidential AI.

One workflow for secure enclaves

Caution runs across AWS Nitro today, with Intel TDX, AMD SEV-SNP, and TPM 2.0 attestations coming in 2026.

FAQs

Frequently asked questions

Caution is a secure hosting platform built on verifiable compute. You deploy applications through a familiar developer workflow, while Caution handles infrastructure setup, enclave builds, and deployment.

It protects data in use by isolating workloads from the surrounding host. Anyone with access to the source code and build inputs can also independently verify what is running in production.

If an attacker modifies your software during building or deployment, you can detect the change and block the modified software from receiving protected data or secrets.

The platform is fully open source.

Explore how Caution works.

Caution is suited to software that handles sensitive data or critical business logic, where a compromise would have serious consequences. Examples include:

Verifiable compute is a way to run software with cryptographic evidence that it matches reviewed source code, build inputs, and configuration.

It builds on confidential computing, where workloads run in isolated environments called secure enclaves or trusted execution environments (TEEs) that protect data while it is being processed. But isolation alone does not tell you whether the application inside is the one you intended to run.

Verifiable compute lets you verify this cryptographically by combining reproducible builds with hardware attestation. Anyone with access to the source and build inputs can rebuild the software and compare its measurements with signed evidence from the running enclave. This gives you cryptographic evidence that the deployed software matches the code you reviewed, helping you detect changes an attacker may have introduced during building or deployment.

Learn how verifiability and reproducible builds connect source code to a running deployment.

No. You can deploy an existing containerized application without rewriting it for Caution. Some changes may be needed to its build, deployment configuration, or how clients connect.

Independent verification requires a reproducible build. Your chosen encryption setup determines whether client integration is needed.

Our AI coding agent skills can help prepare your application, configure deployments, and troubleshoot issues.

Caution offers bring your own compute (BYOC) and fully managed hosting. Because Caution is open source, teams can also self-host it to operate deployments independently.

Compare the options in the deployment models docs.

Book 20 minutes with one of our engineers. We’ll discuss your goals, answer technical questions, and help you get ready to deploy with Caution.

Deploy confidential compute in minutes. Prove exactly what is running in production, from source code to live enclave.