Infragate
Infrastructure for agentic AI
Infragate builds the open-source tooling and cloud services that make agent capabilities manageable: what an agent is allowed to do, which servers it reaches, and what it actually did. This is where we write that down as we ship it.
What we build
Three products are out today. A capabilities registry and agent telemetry are still being built.
-
CAPA
An open-source CLI for the capabilities an agent gets: skills, tools, MCP servers, and sub-agents, declared in one file per repo.
-
Lanyard
A hosted MCP gateway. One URL in front of the servers you pick, each person signed in as themselves, one revocable key per agent.
-
ShareCube
Where agent output becomes team input: artifacts with a URL, versions, and comments anchored to the exact sentence.
-
Built on MCP
Everything speaks the Model Context Protocol, so the tools you wire up are not locked to one client or one vendor.
Why this way
An agent with every tool will eventually use one it should not, and a config copied onto five laptops grows five copies of every secret. Those are not prompt problems. They are infrastructure problems, and they get solved in a file you can read and a gateway you can revoke.
We build on the Model Context Protocol because an open standard is the only version of this that outlives whichever client is popular this year. More on the reasoning in about this blog.
Follow along
Infragate is under active development. Follow the blog for build notes and deep dives, or explore our open-source work on GitHub.