Compose AI fromreusable parts.
A single catalog of agents, skills, and commands. Browse, install, and chain them into autonomous workers — then deploy without touching infrastructure.
Everything builds from these three.
Each primitive is small, scoped, and self-describing. Combine them the way you combine functions — predictable, testable, reusable.
Command in. Agent out.
Start with a command
A reusable prompt frames the task and tone.
Equip with skills
Skills hand the agent the tools it is allowed to use.
Run as an agent
The agent reasons, acts, and calls those skills in a loop.
Skills carry their own permissions.
No skill runs with more access than it declares. Every entry in the Hub ships with an explicit allowed-tools list and a signed trust level, so you always know what a block can touch before you install it.
// declared permissions name: "github.search" trustLevel: "verified" allowedTools: - web_search - github_read userInvocable: true
Build with blocks,not boilerplate.
Browse the catalog, install a skill, and run your first agent in minutes. No GPUs to manage, no credit card to start.
Also available from the CLI: llmtune hub install