Bevel

Bevel is a vendor-agnostic control plane that lets companies define their AI agents, context, skills, tools, and permissions as files stored in their own Git repository. Instead of locking agent configuration inside a single vendor's console, Bevel treats everything as version-controlled artifacts your team owns. Agents connect over MCP (or UTCP) and read only what they're allowed to — whether they're running on a desktop tool like Claude Code or as background server-side processes.

The core idea is that four things fully specify an agent: what it knows (context with provenance), how it works (procedures written in Markdown), what it may do (tool manifests with vault-stored secrets and file-level access rules), and who it acts as (per-agent credentials, not shared service accounts). All of these live in plain Markdown and YAML files inside your company's repository. Branches, diffs, and change requests are handled by Git itself — no custom governance system to learn. The platform also supports agent identity and auditable actions.

Bevel is aimed at engineering and platform teams in enterprises that want to run AI agents across multiple runtimes (Claude Code, Cursor, ChatGPT, custom background agents, etc.) without being tied to a single provider. It's also useful for companies that need rigorous governance, compliance, and audit trails around agent behavior. Real-world use cases include building RFI agents, tender management systems, GTM campaign agents that unify Salesforce and ad platform data, and market intelligence agents that bridge external and internal signals. The company also offers consulting services to help build these agents on top of the platform.