Sandboxes & Runtimes

10 agents tracked · Open-source-first, with limited public GitHub comparators · Ranked by trust score · Updated 2026-08-29 06:32 UTC

HVTracker independently evaluates 10 sandboxes & runtimes using daily signals from GitHub, package registries, and security databases. The category is open-source-first, with a limited set of public GitHub-hosted proprietary or source-available comparators when they are important ecosystem reference points. Each agent is scored on activity, adoption, transparency, safety, and identity. The top-ranked sandboxes & runtime is BoxLite with a trust score of 85.2/100 (Grade A). Other leading projects include E2B and Daytona.

10
Agents
57
Avg Trust
135.5k
Total Stars
1
Grade A
# Agent Trust Stars Language
1 BoxLite A Listed The micro-VM for AI agents — light enough to embed on your laptop, elastic enough to power an agentic cloud. 85.2 2.3k Rust
2 E2B B Listed Open-source, secure environment with real-world tools for enterprise-grade agents. 78.4 13.6k Python
3 Daytona B Listed Daytona is a Secure and Elastic Infrastructure for Running AI-Generated Code 67.3 71.9k TypeScript
4 Fence C Listed Lightweight, container-free sandbox for running commands with network and filesystem restrictions 59.8 940 Go
5 Container Use C Listed Development environments for coding agents. Enable multiple agents to work safely and independently with your preferred 59.8 4.0k Go
6 Agent Infra Sandbox C Listed All-in-One Sandbox for AI Agents that combines Browser, Shell, File, MCP and VSCode Server in a single Docker container. 53.6 5.8k Python
7 OpenSandbox D Listed Secure, Fast, and Extensible Sandbox runtime for AI agents. 47.2 14.8k Go
8 Cube Sandbox D Listed Instant, Concurrent, Secure & Lightweight Sandbox for AI Agents. 46.2 11.4k Go
9 OpenShell D Listed OpenShell is the safe, private runtime for autonomous AI agents. 42.1 8.4k Rust
10 Zeroboot D Listed Sub-millisecond VM sandboxes for AI agents via copy-on-write forking 34.8 2.4k Rust

Grade = trust band: A ≥ 80, B ≥ 65, C ≥ 50, D < 50 — methodology