CoCalc feature
CoCalc vs AI Agent Sandboxes
Choose the execution model that matches the work.
Compare CoCalc's persistent shared projects with API-first AI agent sandboxes for collaboration, review, durable project work, and deployment.
Compare CoCalc's persistent shared projects with API-first agent sandboxes, then choose based on who needs to use and review the environment.
Shared project or agent sandbox?
Agent sandboxes are built to launch isolated execution environments from an API or SDK. Some support persistent files, snapshots, pause and resume, or full virtual machines. CoCalc is built around a persistent shared project that people and agents open and use together.
Choose an agent sandbox when
- You are embedding isolated code execution inside an agent product
- Programmatic lifecycle, concurrency, and per-run environments are the main requirements
- People will review results somewhere outside the execution environment
Choose CoCalc when
- People and agents work directly in the same notebooks, files, documents, terminals, and services
- Collaborators need live work, review context, history, and recovery in one place
- The work must persist across sessions, collaborators, reviews, and handoffs
- You need a hosted service, a local or single-VM installation, or a customer-operated private deployment path
Check the operating boundary
Use the product and compute guides to confirm what is available for your deployment, account, and workload. If your design requires automatic fleets of per-run environments, GPU autoscaling, or exclusive GPU capacity, evaluate infrastructure built for those requirements.