Agent runtime for edge agents
Agent‑Infra released AIO Sandbox, an all‑in‑one runtime that bundles browser, shell, and a shared filesystem for AI agents — a toolkit aimed at accelerating on‑device and edge agent development and testing. The runtime lowers engineering friction for building agentic workflows that need local I/O and system access. (marktechpost.com)
agent‑infra’s sandbox repo shows 3.8k stars and 311 forks on GitHub, with active commits as recently as “last week” indicating frequent upstream changes. (github.com) The project publishes a ready‑to‑run Docker image and documents a one‑line quick start using docker run with --security-opt seccomp=unconfined and port 8080 exposed for the web consoles. (github.com) AIO Sandbox exposes developer services including a VSCode Server and Jupyter notebooks and ships language runtimes surfaced via APIs for programmatic execution. (sandbox.agent-infra.com) The browser surface inside AIO Sandbox is Chromium‑based for automation and scraping flows, enabling headless and full‑VNC browser interactions for agents. (ubos.tech) AIO Sandbox includes a pre‑configured MCP Hub that aggregates tool endpoints so agents can call Browser, File, and Shell tools via a single Model Context Protocol endpoint. (sandbox.agent-infra.com) (marktechpost.com) The project maintains a formal evaluation framework and dataset and published a v1.0.0.x lineage (latest visible release notes reference v1.0.0.152 from November 12, 2025) with options to disable services on demand. (github.com) Official examples and SDK docs show Python and JavaScript clients (SDK updates for API v1.4.0 are in the repo) and include integrations such as Playwright browser automation, OpenAI connector examples, and docker‑compose/Kubernetes deployment snippets with resource requests and limits. (github.com 1) (github.com 2)
