L↻Loop Labs 🧪Strategy system
The engine

What lets the lab ship its own work?

A scoped agent environment connects Slack intent to the right project, evidence, approvals, and deployment path—without pretending autonomy removes accountability.

07 / Ship Lab

One agent. Two scoped businesses. One place to ship.

The goal is not a clever chatbot. It is a managed operating environment where Slack opens the right project, the agent remembers the business, and software moves from request to reviewed release.

Operating model

One core agent, separate hats with hard walls.

Run one Hermes gateway, then route each approved Slack channel into an isolated project profile, workspace, repository, knowledge base, tool policy, and secret set. Marketing cannot see engineering secrets. Engineering cannot publish or spend without the right approval.

How a Slack request reaches the right lab

The channel is the door, not the authorization system. The router enforces identity, project scope, tools, and approval rules.

Slack doorsPeople speak in a project channel and keep each thread attached to one mission.#ship-marketing
#ship-engineering
Project routerChecks the Slack user and channel ID, then selects the matching Hermes profile and policy.channel → profile → policy
Isolated workspaces
Marketing Control RoomRails app · business memory · VP Marketing hat
Product Engineering LabProduct repo · tests · pull requests · deploys
Hosting comparison

Three viable homes. Different trade-offs.

Mac mini and Hetzner behave like durable computers. Railway behaves like a managed application platform. That difference matters more than the logo.

Closest to today

Mac mini

Maximum familiarity and control on hardware you already own.

Setup speed
Agent freedom
Reliability
Managed ops
Rails fit
Best atFull shell, local MCPs, persistent repos, browser automation, and odd integrations.
Watch outHome power and internet, remote recovery, disk failure, residential exposure, and manual backups.
Use it forStaging, private experiments, and a recovery node once the main agent moves.
Cost shape: sunk hardware + power + internet
Recommended core

Hetzner Cloud VM

A remote Linux computer with persistent disk and full administrative control.

Setup speed
Agent freedom
Reliability
Managed ops
Rails fit
Best atAlways-on Hermes, multiple repos, Docker services, workers, Postgres, SSH, and unrestricted MCP integrations.
Watch outYou own Linux updates, firewalling, monitoring, snapshots, encrypted off-site backups, and incident response.
Use it forThe core agent, Slack gateway, project router, private services, and the first product pilot.
Cost shape: roughly €12+/mo shared VM; backups and AI usage extra
Managed alternative

Railway

The smoothest route to managed Rails services, databases, deploys, logs, and domains.

Setup speed
Agent freedom
Reliability
Managed ops
Rails fit
Best atDeploying the Rails apps and Postgres with managed environments, logs, domains, volumes, backups, CLI, and MCP.
Watch outIt is a container platform, not a general workstation. Persistent repos need deliberate volume design, restarts are normal, and usage is metered.
Use it forPublic Rails products after the pilot. Railway’s Slack agent manages Railway infrastructure; it does not replace the custom Loop Labs agent.
Cost shape: $20 Pro floor for a team + metered CPU, RAM, storage, and egress

Scores are planning judgments for this specific use case, not vendor benchmarks. Prices are a July 17, 2026 snapshot and can change.

What the first version actually contains

Core

Agent control plane

Hermes gateway, a small Slack-to-project router, two isolated Hermes profiles/workspaces, scheduler, audit log, shared Postgres, encrypted backups, monitoring, and scoped service accounts.

agent@looplabs…
Project 01 · VP Marketing

Marketing Control Room

A Rails app for positioning, offers, audiences, campaigns, content, spend, experiments, KPIs, decisions, and weekly recommendations. The agent may analyze and draft; publishing or spending starts behind approval.

#ship-marketing
Project 02 · Engineering

One shippable product

One product repository with a test suite, issue queue, preview environment, pull-request flow, deployment runbook, error tracking, and a narrow product metric. The agent branches and tests; humans approve merge and production.

#ship-engineering

Non-negotiable guardrails

Separate identity

Give the agent its own mailbox and service accounts. Never lend it a founder login. Begin with outbound email in draft-only mode.

Project isolation

Map exact Slack channel IDs to exact workspaces, repos, knowledge, MCPs, and secrets. Deny everything else by default.

Human gates

Require approval for production deploys, merges, payments, mass email, publishing, deleting data, and changing access.

Recovery first

Nightly encrypted off-site backups, restore drills, health checks, spend limits, logs, and a documented kill switch before autonomy expands.

Smallest credible pilot

Provision the boxHetzner VM, Ubuntu, non-root admin, firewall, Tailscale, Docker, automatic security updates, snapshots, and off-site backup.
Install the brainHermes gateway, Slack app, project router, two profiles, scoped MCPs, mailbox, secret store, logging, and approval defaults.
Build both lanesCreate the Marketing Control Room Rails skeleton and one small engineering product with tests, preview, and pull-request delivery.
Earn autonomyRun supervised for two weeks. Expand permissions only after clean audit logs, successful restores, and repeatable releases.
Hermes Agent docsHetzner CloudRailway for AgentsRailway VolumesRailway Rails guideRailway pricing