Your own agent control plane. On your own machine.
Powerloom Home is a self-hosted, single-user agent governance control plane. Alfred — the meta-agent — runs locally and drives provisioning through natural language. Claude Code connects as a first-class client. No AWS. No multi-tenant auth. No invite codes.
Open source under ELv2. Built on the same core as the enterprise edition at powerloom.tech.
What you get
Two containers come up on docker compose up: a Postgres 16 + pgvector store, and a FastAPI control plane. First-boot bootstrap creates your single-user org, provisions Alfred, mints a 365-day Personal Access Token, and writes a ready-to-paste Claude Code config to your host filesystem.
- Alfred. The meta-agent. Runs on
claude-opus-4-6by default. Deploys other agents via tool calls, reads context from your prompt, drives Powerloom primitives without a UI. - Docker-local MCP spawner. Alfred can deploy MCP server containers on your machine (echo, files, postgres, slack) with the same image-build path as the enterprise edition's Terraform deploys. No AWS required.
- Claude Code as client. The
powerloom-mcpstdio server exposes 12 tools for session coordination, the project tracker, agent listing, and invocation. Ready to paste into CC's.mcp.jsonafter the firstup. - Single-tenant. No multi-user RBAC, no billing, no super-admin surface. One OrgAdmin. One Anthropic key. Your data, your process.
Quick start
Full walkthrough at the setup guide. The short version:
git clone https://github.com/shanerlevy-debug/powerloom-home.git
cd powerloom-home
cp .env.home.example .env.home
# edit: POWERLOOM_HOME_EMAIL + ANTHROPIC_API_KEY
docker compose -f docker-compose.home.yml --env-file .env.home up -d --build
python scratch/home_init.py --write
python scratch/home_init.py --byok-anthropic sk-ant-api03-...
# paste home_data/.mcp.json into ~/.claude/mcp.json
# open Claude Code, paste home_data/cc-prompt.md as first message
# talk to AlfredBuilt on the same core as the enterprise edition
Powerloom Home and Powerloom Enterprise share the same control-plane kernel: identity, RBAC, agent lifecycle, MCP deployment, reconciler, runtime drivers, memory, and the Alfred meta-agent. What differs:
| Home | Enterprise | |
|---|---|---|
| Deployment | Your machine via Docker | Hosted SaaS |
| Tenancy | Single user | Multi-tenant org hierarchy |
| RBAC | OrgAdmin only | Full role bindings, groups, deny-override |
| MCP provisioning | Local Docker containers | AWS ECS + ALB + Terraform |
| SaaS templates | echo / files / postgres / slack | 10+ incl. github, notion, jira, linear, etc. |
| Audit | Console + DB | Hash-chained, SOC 2 ready, SIEM export |
| Billing | Free | Tiered |
| License | ELv2 (open source) | Commercial |
| Support | GitHub issues + community | SLA-backed |
If your use case outgrows the home edition — multiple users, audit compliance, managed infrastructure — the enterprise edition is at powerloom.tech.
Who this is for
- Developers building with agents. You want a governed control plane for your agent work, not a collection of scripts. Local-first, hackable, open.
- Claude Code power users. You've already got CC in your workflow. Adding Alfred + MCP templates makes CC materially more useful for agent orchestration.
- Privacy-conscious operators. You don't want your agent traffic going through a third party's control plane. Home mode runs entirely on localhost; only Alfred's invocations touch Anthropic.
- Contributors and hackers. ELv2 lets you modify, self-host, and fork. The enterprise features you'd never see in a permissive-license repo are here, minus the SaaS-specific bits.
Next steps
Get Powerloom Home
Release zips, setup PDF, GitHub repo.
Read →5-minute setup
Walkthrough from git clone to first Alfred conversation.
Read →MCP templates
What each template does. What config it needs. What's on the roadmap.
Read →Contribute + connect
GitHub issues, discussions, contributing guide.
Read →