Skip to main content
loomloom is an AI work compiler and runtime. It enables developers and AI agents to define the inputs, logic, capabilities, and expected outputs required to produce an outcome, then compile that work into reusable AI systems. AI work can range from a prompt or skill to a multi-step system containing instructions, tools, workflows, and generated artifacts. loomloom provides the toolchain to version, execute, maintain, and distribute it like software.

AI work IR

Define typed inputs, steps, dependencies, models, and outputs in a reusable template.

SkillBots

Package a compiled template version as a deployable, modular AI system.

Execution

Validate, estimate, confirm, run, monitor, and retrieve observable results.

​
How loomloom works

loomloom applies a compiler-style pipeline to AI work:

​
Components

​
Current public workflows

  • Run official templates maintained by the selected execution environment.
  • Create immutable private template versions with TemplateSpec.
  • Use workbooks by default or explicit JSON/JSONL input for programmatic integrations.
  • Monitor runs and retrieve result rows, result workbooks, and generated artifacts.
  • Publish a proven private version as a SkillBot or call an approved SkillBot from the Market.
  • Install a private template or Market SkillBot as a local Agent Skill wrapper for Codex, Claude Code, or OpenClaw.
loomloom is currently in beta. Implemented behavior is documented here; architecture and design details may continue to evolve through public releases.
The CLI can be used directly. Agent integrations add guided planning, input collection, and confirmation around the same deterministic commands.

​
Start here

  1. Install loomloom.
  2. Configure your credentials.
  3. Complete the Quick Start or browse the main workflows.
  4. Review Security Notes before using production data.