Building the infrastructure layer for agentic AI
Open-source tooling and hosted services that help teams manage, share and observe what their agents can do.
AI agents need infrastructure, not just models
As AI agents move beyond demos into production, teams need reliable ways to package capabilities, host tool servers, distribute reusable skills, and understand what their agents are doing. Today, most of this is held together by hand.
We are building the stack that makes agentic AI manageable at scale. It starts with ShareCube, where agent output becomes team input; Lanyard, a hosted MCP gateway; and CAPA, an open-source capabilities package manager. A cloud registry and telemetry come next.
We build on the Model Context Protocol because open standards, not proprietary lock-in, will define how agents interact with the world.
The Infragate platform
ShareCube
Agents publish HTML and Markdown artifacts to a URL your team can open, edit and comment on sentence by sentence. No copy-pasting a wall of text out of a chat window.
AvailableLanyard
A hosted MCP gateway. Your agents get one URL and one revocable key; each person signs in to the servers behind it as themselves, so an agent acts as a person and never as a shared bot.
AvailableCAPA
Declare every skill, rule, MCP server and hook once in capabilities.yaml. CAPA writes them into Cursor, Claude Code, Codex, Copilot and 35 other agents in each one's native format.
AvailableCapabilities Registry
A cloud registry for the skills, plugins and tools CAPA installs. Versioned, searchable, with organization scopes and dependency resolution.
Coming soonAgent Telemetry
Distributed tracing for agent tool calls, usage analytics and cost attribution, anomaly detection, and a dashboard and API to read it all.
Coming soon
Follow our progress
ShareCube, Lanyard and CAPA are available today. Star the repo, follow us on LinkedIn, or get in touch.