What is Workloom?
Workloom is an organization-level AI workflow package manager and workspace desired-state platform.
It owns canonical definitions of skills, workflows, agents, prompts, MCP configs, hooks, instructions, and provider-specific renderers, then materializes those into local repositories, monorepos, folders, or unique-key-bound workspaces through the wl CLI.
The problem
AI coding tools — Claude Code, Codex, GitHub Copilot, Gemini — each read configuration from different files in different formats at different paths. When your organization has 10, 50, or 200 repositories, you end up with:
- Copy-pasted configs that diverge immediately
- No canonical source for skills, agents, or prompts
- No distribution mechanism for updates
- No visibility into what is installed, what is stale, what is missing
- No CI enforcement that AI configs match the team standard
The solution
Workloom treats AI workflow assets the way npm treats packages:
- Define canonical packages of skills, agents, prompts, hooks, and configs
- Publish immutable, versioned, content-addressed package versions to a registry
- Bind workspaces (repos, monorepos, folders) to selected packages
- Render provider-specific projections — Claude gets
.claude/, Codex getsAGENTS.md, Copilot gets.github/copilot-instructions.md - Sync deterministic render plans into local workspaces with ownership tracking
- Detect drift and enforce desired state in CI with
wl sync --check
The mental model
npm package manager
+ Git object model (content-addressed, sha256)
+ Terraform plan/apply (desired-state, dry-run, drift)
+ Provider-specific renderers
+ CI drift checking
What Workloom is not
Workloom is not generic file sync, not Dropbox for repos, not rsync over SSH. The files it manages are not arbitrary — they are canonical AI configuration assets and provider-specific generated projections.
Current status
Workloom is in active development. The current release includes:
- 9 core packages (types, manifest, storage, auth, registry, workspace, renderer-core, renderer-claude, renderer-codex)
wlCLI with login, bind, pull, push, and status commands- Fastify API server with auth, registry, and workspace endpoints
- Next.js web app for browsing packages and managing workspaces
- Claude and Codex provider renderers
- Content-addressed storage with SHA-256 verification
- In-memory stores for development (database backends coming in a future phase)
See What's Coming for the roadmap.