h5i-dev/h5i

What it solves

AI agents often struggle with secure and resource-efficient web browsing. Traditional tools like Playwright or Puppeteer are resource-heavy and lack built-in auditing and strict network controls. h5i provides a lightweight, secure, and auditable browser environment specifically designed for AI agents to interact with the web without compromising the host system.

How it works

h5i is written in pure Rust and avoids Chromium or V8 to reduce memory usage and increase read speeds. It operates through a series of "sessions" where agents can open pages, take snapshots, click elements, and extract structured data.

To ensure security, it offers multiple levels of confinement (sandboxing) ranging from simple process isolation to full hardware-isolated microVMs. It also implements policy-controlled network access, allowing users to define exactly which domains an agent can reach. Every action and network request is recorded in a reviewable audit log (receipt), ensuring that the agent's behavior is fully transparent.

Who it’s for

Developers building AI agents that need to browse the web, extract data, or perform actions on websites while maintaining strict security, auditing, and low resource overhead.

Highlights

• Pure Rust Implementation: Eliminates the dependency on Chromium/V8 for significantly lower peak memory and faster reads.
• Configurable Sandboxing: Five tiers of isolation (workspace, process, supervised, container, microVM) to protect the host system.
• Auditable Sessions: Detailed records of all network requests, including refusals, and session verbs for full transparency.
• Policy-Controlled Access: Fine-grained control over network egress, filesystem access, and resource limits via configuration files.
• Agent-Ready Interface: Built-in commands for snapshots, delta-reads, and structured data extraction tailored for LLM consumption.