AI agents

Understand what makes an AI code execution sandbox secure: isolation models, egress controls, secrets handling, and the right questions to ask any provider.

Answers to common questions about AI agent sandbox isolation, egress controls, file access, session state, secrets, audit logs, compliance, pricing, and self-hosting.

A buyer's guide to AI agent sandbox pricing: per-session fees, compute tiers, storage, egress, package caching, idle time, and self-hosted cost models.

A security evaluation guide covering which events AI agent sandbox audit logs must capture, retention policies, log integrity, and how to surface logs for incident response.

A practical security guide for teams enabling AI agents to install packages in sandboxes: allowlists, version pinning, registry mirrors, egress controls, and audit logging.

Requirements checklist for AI-generated code sandboxes: isolation, lifecycle API, concurrency, observability, resource limits, and backend integration.

A practical evaluation checklist for AI developers comparing sandbox providers: API surface, SDK compatibility, session lifecycle, packages, network, and pricing.

Evaluate Firecracker microVMs for AI agent sandboxes, including isolation, lifecycle overhead, egress, packages, secrets, and audit controls.

Learn how RL agent sandboxes support repeatable trials with isolated state, resets, snapshots, telemetry, limits, and review boundaries.

Run MCP servers in an isolated MCP sandbox to scope filesystem, secrets, and network. Learn when MCP server isolation shifts the agent trust boundary.

Learn how to design an AI data analyst that runs Python, inspects CSV files, creates charts, and controls package access in a sandbox.

Harbor Novita Agent Sandbox support is visible on Harbor main. Learn the release boundary before using it in agent evaluations.