THE AI COMPANY, CREWSPACE.
An open-source multi-agent AI workforce platform, built from scratch — a CEO agent that decomposes goals, delegates across specialist agents, proposes new hires when the org is missing a skill, and coordinates everything through approval workflows a human actually controls. Stark built a suit; this builds the whole company.
AN ORG CHART THAT EXECUTES ITSELF
Most "multi-agent" demos are a group chat with extra steps. CrewSpace asks the harder question: what if a company's entire operating structure — hierarchy, delegation, hiring, approvals — was software?
You hand the platform a goal. A CEO agent decomposes it into domain subtasks, routes them to specialist agents, and when the org lacks a capability, it proposes a new hire — a fresh specialist spun up with its own role, memory, and workspace permissions. Like JARVIS running Stark Industries: autonomous where it's safe, asking the human where it matters.
"Anyone can spawn agents. CrewSpace runs a company."
- FILE SNAPSHOT
- Founder-led — HITL approval workflows
- Per-agent memory — persistent across sessions
- A2A + group chat — built-in comms layer
- Workspaces — shared files, scoped permissions
- Terminal ops — integrated & reviewable
- Open source — full build on GitHub
FROM ONE GOAL TO A WORKING COMPANY
Goal Intake
You give the company one objective. The CEO agent breaks it into domain subtasks — engineering, research, content, ops — each with a scope and an owner.
Delegation
Subtasks route to specialist agents by domain. Each specialist works inside its own workspace with scoped file permissions — no agent touches what it shouldn't.
The Hiring Loop
Missing a capability? The CEO proposes a new hire. You approve, and a new specialist joins the org with its own role, memory, and permissions.
Coordination
Agents collaborate over agent-to-agent and group chat. Consequential actions flow through approval workflows, so a human signs off before anything irreversible happens.
Delivery
Work lands on real-time task boards; the founder dashboard tracks every agent, task, and conversation — and the interactive 3D office lets you literally walk the floor.
EVERYTHING A COMPANY NEEDS, AGENT-NATIVE
Persistent Agent Memory
Every agent keeps its own persistent memory across sessions — decisions, context, and learnings survive restarts instead of resetting to zero.
Comms Layer
Agent-to-agent messaging and group chat with humans in the room. Conversations are first-class objects, not hidden prompt chains.
Shared Workspaces
Agents collaborate on shared files with per-agent permissions — read, write, or hands-off — so the intern agent can't touch prod configs.
Integrated Terminal Ops
Agents run real terminal operations inside the platform — visible and reviewable, not hidden behind an opaque API call.
Real-Time Task Boards
Every subtask, owner, and status on a live board, with a founder dashboard over the whole org — the standup that writes itself.
Interactive 3D Office
A rendered 3D workspace where you watch your AI org at work — because dashboards are better with a skyline.
HIERARCHY, WIRED FOR AUTONOMY
Founder Dashboard
Human Oversight
3D Office
Live Org View
Task Boards
Real-Time Status
CEO Agent
Goal Decomposition · Delegation · Hiring Proposals
Specialist Agents
Domain Execution
Hiring Loop
HITL Approval
Group Chat
A2A Coordination
Memory Graph
Persistent
Workspaces
Scoped Files
Terminal Ops
Reviewable
Approval Engine
Human Gates