ByteBulletin

[tooling] · · 2 min read

Hexis: A Git-Backed Control Plane for Managing AI Agent Skills and Permissions

Bevel Software releases Hexis, an open-source tool that treats AI agent skills and tools as version-controlled files with granular access control and bidirectional review capabilities.

By ByteBulletin Editors · Editorial Team

[tooling]

As AI coding agents become more integral to development workflows, managing the skills, tools, and context they rely on has become a significant operational challenge. Bevel Software has introduced Hexis, an open-source control plane designed to centralize the management of AI agent capabilities. Unlike traditional Model Context Protocol (MCP) gateways that are often unidirectional and lack fine-grained access controls, Hexis positions itself as a bidirectional management layer where skills and tools are treated as reviewable files within a Git repository.

The core philosophy of Hexis is that the audit trail is the storage layer. By storing every skill, tool manual, and permission as a file in a Git repository that the organization owns, the platform leverages Git's native capabilities for history, blame, and reversion. This approach ensures that changes to agent capabilities are tracked with the same rigor as code changes. On protected branches, changes require approval from file owners before they become available to the team's agents, creating a robust review process that extends to non-technical stakeholders who can propose changes or browse available tools.

Hexis is designed to be MCP-native and self-hosted, allowing it to integrate seamlessly with existing AI agents such as Claude, Cline, and Cursor. The platform enforces per-file access control, ensuring that agents can only access the skills and context relevant to the user's role. This granular permission model addresses a common pain point in enterprise AI deployments, where broad access to tools can lead to security risks or unintended actions. By keeping credentials out of the agent's hands and resolving permissions per file, Hexis aims to provide a secure and auditable environment for AI agent operations.

For developers looking to deploy Hexis, the platform requires Docker with Compose and an empty Git repository on any host such as GitHub, GitLab, or Bitbucket. The deployment process is streamlined, with the application seeding the repository with a starter template on first run. Bevel Software also offers a managed hosting option for teams that prefer not to handle infrastructure, providing hosting, upgrades, backups, and SSO integration. The project is licensed under Apache-2.0, encouraging community contribution and enterprise adoption.

SHARE

← All stories