HomeProjectsArticlesPK BoardAbout
Explore Projects
🏗️

Sandbox & Execution

Secure execution environments, code sandboxes, and runtime isolation tools for AI agents

55 projects

DeerFlow

81.3k · Python
Active A+

An open-source long-horizon SuperAgent harness by ByteDance that researches, codes, and creates with sandboxes, memories, tools, skills, subagents and message gateway for complex tasks.

super-agentmulti-agentresearch +4
  • · Super Agent harness - orchestrates sub-agents, memory and sandboxes for complex long-horizon tasks from minutes to hours
  • · Extensible skills system - modular skill-based architecture for extending agent capabilities with custom skills
  • · Sandboxed code execution - built-in secure sandbox for agents to safely execute code and file operations in isolation
View full details → View on GitHub

Daytona

71.8k · Unknown
Normal A

Daytona provides secure development-environment infrastructure for coding agents and automation workflows, serving as a runtime base for remote execution tasks.

dev-environmentsandboxcoding +1
  • · Sandboxes - fully isolated compute units with dedicated kernel, filesystem, network stack, and allocated vCPU/RAM/disk
  • · Sub-90ms startup - sandboxes spin up in under 90ms from code to execution, supporting Python/TypeScript/JavaScript
  • · Multi-language SDK - Python, TypeScript, Go SDKs plus REST API and CLI for programmatic control
View full details → View on GitHub

CUA

22.1k · HTML
Active A

CUA provides open-source infrastructure for Computer-Use Agents, including sandboxes, SDKs, and benchmarks to train and evaluate AI agents that control full desktops (macOS, Linux, Windows).

computer-usedesktop-automationsandbox +2
  • · Cua Driver: background computer-use agent without stealing cursor focus
  • · Cross-platform sandboxes: unified API for macOS, Linux, Windows, Android
  • · CuaBot: multi-agent computer-use sandbox CLI with native window display
View full details → View on GitHub

Context Mode

20.3k · TypeScript
Active A

Context Mode is a context window optimization tool for AI coding agents that sandboxes tool output for 98% context reduction across 12 major platforms.

context-optimizationmcpclaude-code +2
  • · Sandboxes tool output, reducing 315KB raw data to 5.4KB — a 98% context reduction
  • · SQLite + FTS5 session continuity that recovers key context via BM25 search after conversation compaction
  • · Think in Code paradigm: LLM writes analysis scripts instead of processing raw data, saving 100x context
View full details → View on GitHub

keploy

18.4k · Go
Active A

Keploy is an open-source sandbox platform for API and E2E testing that records and replays real traffic inside an isolated environment to generate test cases.

sandboxtestingapi-mock +2
  • · Traffic record-and-replay - capture real API calls including requests, responses, and dependencies as runnable test cases
  • · Dependency mocking - intercept databases, HTTP calls, and Kafka topics and replace them with deterministic mocks
  • · Time freeze and clock control - freeze system time inside the test sandbox for time-sensitive logic
View full details → View on GitHub

OpenSandbox

14.9k · Go
Active A

OpenSandbox is an open-source, secure, fast, and extensible sandbox runtime for AI agents, developed by Alibaba.

sandboxai-infrastructurekubernetes +2
  • · Multi-language SDKs: Python, Java/Kotlin, JavaScript/TypeScript, C#/.NET, and Go with unified sandbox APIs
  • · Docker and Kubernetes runtimes: built-in lifecycle management for both local development and large-scale distributed scheduling
  • · Strong isolation: supports gVisor, Kata Containers, and Firecracker microVM secure container runtimes
View full details → View on GitHub

E2B

13.7k · Python
Active A+

E2B provides secure cloud sandboxes for AI agents, supporting code execution, file operations, and isolated compute as an execution layer for coding and automation workflows.

sandboxcode-executionsecurity +1
  • · Secure Cloud Sandboxes: Provides isolated cloud execution environment for AI agents
  • · Code Execution: Supports JavaScript/Python SDK to run AI-generated code
  • · File Operations: Perform file read/write and management within sandboxes
View full details → View on GitHub

CubeSandbox

11.7k · Go
Active A

A high-performance, secure sandbox service for AI agents by Tencent Cloud, built on RustVMM and KVM with hardware-level isolation, sub-60ms cold start, <5MB memory overhead, and E2B SDK compatibility.

sandboxcode-executionrust +2
  • · Blazing-fast cold start: sub-60ms end-to-end sandbox creation via resource pool pre-provisioning and snapshot cloning
  • · Ultra-low memory overhead: less than 5MB per instance, enabling thousands of sandboxes on a single machine
  • · True kernel-level isolation: each agent runs with its own dedicated Guest OS kernel based on RustVMM and KVM
View full details → View on GitHub

Omnigent

9.6k · Python
Active B

Open-source AI agent framework and meta-orchestrator that unifies Claude Code, Codex, Cursor, Pi, and custom agents with policy enforcement, sandboxing, and real-time collaboration.

agent-frameworkmulti-agentsandbox +2
  • · Multi-agent orchestration — Unifies Claude Code, Codex, Cursor, Pi, and custom agents with seamless framework switching
  • · Policy and sandbox governance — Centrally defined and enforced access policies, resource limits, and sandbox boundaries
  • · Real-time collaboration — Multi-user real-time collaborative debugging of agent tasks from any device
View full details → View on GitHub

Databend

9.4k · Rust
Active B

A Data Agent Ready Warehouse unifying Analytics, Search, AI, and Python Sandbox in one system. Runs on your S3 with built-in vector search, full-text search, and Python execution for AI-powered data analysis.

databasevector-databaserag +2
  • · Enterprise data warehouse built in Rust — analytics, vector search, full-text search, and auto schema evolution unified in one engine
  • · Agent-ready architecture with Sandbox UDF for running Python agent logic, SQL orchestration, and transactions
  • · Git-like data branching for safe experimentation on production data snapshots
View full details → View on GitHub

Sandcastle

7.8k · TypeScript
Normal B

A TypeScript tool for orchestrating sandboxed coding agents with secure execution environments powered by sandcastle.run.

coding-agentsandboxtypescript +1
  • · Sandboxed agent orchestration — safely execute AI coding agents in isolated sandboxes via sandcastle.run()
  • · Multiple sandbox providers — built-in Docker, Podman, Vercel support, custom sandbox creation
  • · Provider-agnostic design — swappable agent providers, built-in Claude Code support
View full details → View on GitHub

Steel Browser

7.6k · TypeScript
Active A+

Steel Browser is an open-source browser sandbox purpose-built for AI agents and applications. It provides a full browser API with session management, proxy integration, and built-in anti-detection, enabling web automation without infrastructure headaches.

browser-automationbrowser-sandboxanti-detection +3
  • · Full Browser Control — Puppeteer and CDP-based complete control over Chrome instances with Puppeteer, Playwright, or Selenium support
  • · Session Management — maintains browser state, cookies, and local storage across requests for persistent browsing
  • · Anti-Detection — built-in stealth plugins and fingerprint management to avoid bot detection
View full details → View on GitHub

Agent Governance Toolkit

6.2k · Python
Active A

Microsoft's AI Agent Governance Toolkit providing policy enforcement, zero-trust identity, execution sandboxing, and reliability engineering for autonomous AI agents. Covers 10/10 OWASP Agentic Top 10.

securityevaluationpython +2
  • · Deterministic policy enforcement engine: intercepts and evaluates all operations before model output reaches tool calls, with YAML policy files defining allow/deny/approval-required rules — denied actions are structurally impossible
  • · Zero-trust identity and audit trail: assigns unique DIDs to each agent, generates tamper-evident audit records for every operation including active policy, request content, and decision rationale
  • · Multi-language SDK support: Python, TypeScript, .NET, Rust, and Go SDKs with unified policy evaluation APIs covering major development stacks
View full details → View on GitHub

Agent Sandbox

5.8k · Python
Active A

All-in-One Sandbox for AI Agents that combines Browser, Shell, File, MCP and VSCode Server in a single Docker container, providing a secure isolated execution environment for agents.

automationdevopsdocker +2
  • · All-in-one sandbox combining Browser, Shell, File, VSCode Server, Jupyter, and MCP in one container
  • · Unified filesystem enabling seamless data flow between browser downloads and shell operations
  • · Full browser automation via VNC, CDP, and MCP interfaces
View full details → View on GitHub

Sandbox Runtime

5.1k · TypeScript
Active A

An experimental lightweight isolated runtime from Anthropic for executing agent tasks in a sandboxed environment.

sandboxruntimecode-execution
  • · OS-level sandboxing without containers using native primitives (sandbox-exec on macOS, bubblewrap on Linux)
  • · Filesystem isolation with deny-then-allow read patterns and allow-only write patterns
  • · Network isolation via HTTP and SOCKS5 proxy-based domain filtering with allowlists/denylists
View full details → View on GitHub

Magicrew

5.0k · TypeScript
Active A

Open-source all-in-one AI productivity platform combining a generalist AI agent, workflow engine, instant messaging, and online documents

productivityworkflow-engineno-code +4
  • · Enterprise-grade AI Agent platform with sandbox isolation and human-in-the-loop approval
  • · Results as deliverables: AI output rendered directly into PPTs, dashboards, reports, Excel
  • · Unified data hub replacing scattered employee accounts with institutional knowledge persistence
View on GitHub

JupyterLite

4.9k · TypeScript
Active A

A Wasm-powered JupyterLab that runs entirely in the browser, providing a zero-install Python execution environment for AI agents.

jupyterwebassemblypython +2
  • · Browser-native runtime - Runs entirely in the browser via Pyodide + JupyterLab with no backend required
  • · Offline execution - Executes Python in a sandboxed WebAssembly runtime with no server
  • · Embeddable Lab - Drop into any web app as an iframe to deliver an interactive notebook UI
View on GitHub

(24 / 55)

Related Articles

sandbox-executionmicrovmfirecracker

Sandboxing Code Execution in AI Agents: From Docker to microVMs, a Decision Matrix

A side-by-side comparison of five sandbox technologies, weighing latency, security, and ops cost.

AI Agent沙箱代码执行

Sandboxing AI Agents: Isolation Strategies for Safe Code Execution

Comparing container, WebAssembly, and process-level isolation approaches, with practical code for safely executing agent-generated code.

Web Scraping反爬Browser Fingerprint

Anti-Bot and Browser Fingerprinting: Modern Web Scraping Strategies

A systematic deep dive into modern web scraping anti-bot defenses: HTTP layer (curl_cffi for TLS fingerprint), browser layer (Playwright plus stealth), behavior layer (human behavior simulation), covering Cloudflare bypass, proxy IP rotation, Scrapling selector self-healing, and legal compliance boundaries.

Browser AgentWeb 自动化Playwright

Browser Agents in Practice: Architecture and Pitfalls of AI-Controlled Browsers

Breaking down three abstraction layers for browser automation—from raw Playwright to structured extraction—with production patterns, runnable code, and common pitfalls.

上下文工程长上下文RAG

Context Engineering: Context Decay and Recovery in Long-Conversation Agents

Long-conversation agents fail at context management, not model capability. A systematic comparison of sliding window, retrieval injection, and layered compression strategies with practical decay diagnosis and recovery patterns.

DifyAgent 工作流RAG

Dify in Practice: Full-Stack Low-Code Platform from RAG to Agent Workflows

Dify (145k Stars, $30M Pre-A) is the benchmark for open-source LLM app platforms. From Docker deployment to RAG pipelines, agent orchestration and MCP integration — this article takes you end-to-end.