hezo-ai/hezo

What it solves

Hezo provides a secure, self-hosted environment to manage teams of AI agents organized like a corporate hierarchy. It eliminates the need to manage multiple terminal tabs or disparate tools by providing a centralized web app for coordinating goals, projects, and budgets across various AI roles (e.g., CEO, Captain, Engineer).

How it works

Hezo runs as a self-contained binary that leverages Docker-compatible runtimes to sandbox every project in its own container. It uses a "meta-harness" to standardize tools and memory across different model providers (Anthropic, OpenAI, Google, xAI, etc.) by running them through their first-party CLI runtimes. To ensure security, it employs an egress proxy that swaps secret placeholders for real credentials only when requests are sent to authorized hosts, preventing agents from leaking sensitive keys.

Who it’s for

It is designed for users who want to run autonomous AI agent teams on their own hardware or cloud VMs, maintaining full control over their data, model spend, and security credentials.

Highlights

• Organizational Structure: Uses a hierarchy of CEOs, Coaches, and Captains to manage worker agents and project progress.
• Secure Secret Management: Agents use placeholders instead of real keys; the egress proxy prevents credential leakage.
• Model Agnostic: Supports a wide range of providers via first-party CLIs, including local models via Ollama and LM Studio.
• Sandboxed Execution: Every project runs in its own isolated container to protect the host system.
• Budgetary Control: Includes budget caps and tracking for both token spend and container runtime hours.
• External Integration: Connects to GitHub for verified commits and integrates with Slack, Telegram, and Discord for communication.