Guilherme Salazar

At Kong, Salazar worked on enterprise access controls, plugins, gRPC, service mesh and release infrastructure. Context Mesh connected OpenAPI parsing and code generation with a Python MCP runtime. Earlier work included Elasticsearch streaming at Appbase and Lua in the NetBSD kernel. His current role concerns governance and security for AI traffic.[2]

Expose the context that reaches a coding model

Lunar separates a small Rust host from Lua configuration. The host owns the terminal, model transport and four tools for reading, writing, editing and running commands. Rules and related-repository maps enter as inspectable context; the code itself is loaded when needed. Conversations persist as local missions, and compaction is an explicit user action. The harness deliberately leaves workflow choices to its operator, with execution intended for trusted environments.[3]

Keep a service token outside the agent runtime

Salazar’s Runta tutorial walks through creating a runtime, storing a GitHub token as a service-side secret and binding it to an allowed API request. The runtime can make an authorized request without receiving the raw token. The worked setup connects sandboxing with narrowly scoped network access and checks the path with a sample agent task. Commands are documented against Runta CLI 0.1.15.[4]