Reduce token use by up to 98% with Heroku’s code execution sandboxes. Securely run agent-generated code in isolated one-off dynos.

Heroku offers insights into cloud application development, deployment, and management, providing documentation and best practices for deploying and scaling applications on the Heroku platform. By exploring Heroku's curated content, developers can learn about Heroku's containerized deployment model, continuous delivery workflows, and add-on ecosystem for extending application functionality. Whether you're deploying web applications, APIs, or background workers, Heroku offers resources to streamline your development workflow and focus on building great software.

Heroku

Heroku launched code execution sandboxes that allow AI agents to run untrusted code securely in isolated one-off dynos. This approach can reduce token consumption by up to 98% by letting agents write scripts that orchestrate operations instead of forcing all data through the model's context window. The sandboxes are available as built-in tools in Heroku's Managed Inference and Agents API or as deployable open-source MCP servers for use with Agentforce, Claude Desktop, or Cursor. Implementations exist for Python, Ruby, Node, and Go.

Code Execution Sandbox for Agents on Heroku

How agents improve with code execution tools

Deploying your own code execution MCP server