Environments forAI agents.
Launch in one click. Every outcome verified. Cost falls run after run.
Free tier · no card · live in about a minute
OKXMagicPathNebiusGaiaEVM CapitalHysolwin Green EnergyPPAI InnovationsEnviroEdgeNVIDIAOKXMagicPathNebiusGaiaEVM CapitalHysolwin Green EnergyPPAI InnovationsEnviroEdgeNVIDIAOKXMagicPathNebiusGaiaEVM CapitalHysolwin Green EnergyPPAI InnovationsEnviroEdgeThe unit is the work.
Most AI pilots can't show a bottom-line effect. XO turns agent work into units of work you can check, price, and repeat.
Read the researchA definition of done
You say what finished and correct looks like before the agent starts.
A checkable result
The environment snapshots state before and after. Comparing the two is the same check a manager makes.
One owner
Scoped small enough for a single agent to be accountable: one name against the result, like a contractor.
A record it can't fake
Before, after, and every action in between. The agent never grades its own homework.
The quirq.
Tokens meter what AI consumes. The quirq meters what it delivers, including the review time your token bill never sees.
Whitepaper · Draft v3
The quirq: a unit of measurement for the business impact of an agentic workforce
A quirq is minted, never self-reported: an owner budgets an outcome, the environment snapshots before and after, and verification mints delivered work.
Suraj Sharma · XO Labs · 2026
Impact,
measured.
Your AI has an electricity bill. XO gives it a P&L.
Budget minus spend, tracked for every unit of work, every session. The gap is yours to widen.
The whitepaper's worked ledger: efficiency 3.1× → 5.6× in a quarter. The token bill alone read +83%. Illustrative.
Agents, humans, tools.
No barriers.
Agents change but the environment compounds. Swap in a cheaper model tomorrow, keep every saving you've earned.
Create. Set up. Share.
Create your first project in about a minute.
Sign up free
Isolated
by default.
Every environment is sandboxed, encrypted, budget-capped, and fully auditable.
Environment isolation
Each agent runs in its own sandboxed environment.
Encrypted memory
Data encrypted at rest and in transit.
Full audit trails
The environment keeps the record, not the agent: tamper-evident, hash-chained.
Permission boundaries
Principle of least privilege by design.
Pay per workspace.
Never per seat.
Tokens price the machinery. Units of work price the result. A workspace is where both happen, and it's the only thing you pay for.
Free
Start at zero, upgrade when the work sticks.
Free forever
- 1 workspace to start
- 10 workspaces free for researchers
- Full SDK + usage dashboards
Starter
10 workspaces to put agents to work.
14-day free trial
- 10 workspaces
- All four templates: Cowork, OpenClaw, Hermes, Claude Code
- Community support
Pro
30 workspaces for teams shipping with agents.
- 30 workspaces
- Custom templates
- MCP integrations
- Spend tracking + agent comparison
Max
500 workspaces. White-label. For fleets.
- 500 workspaces included
- White-label + SSO / SAML
- Full audit trails per environment
- Email + Slack support
A workspace where your agent does real work: runtime, memory, tools, and a record of everything that happens. A desk, not a chat window. (Environment and workspace mean the same thing; pricing is per workspace.)
Runtime plus memory, tools, a record, and a definition of done. That bundle is what makes AI work checkable, priceable, and repeatable. The research behind it lives in our docs.
The output meter for AI. You budget an outcome, the environment snapshots the world before and after, verifies completion, and mints quirqs of delivered work. A quirq is a dollar of verified work, never self-reported, comparable to payroll and vendor spend. The full calculus is in the whitepaper.
With a ledger, not a token bill. XO meters every unit of work all-in (inference, compute, APIs, human intervention) against what it delivered, so you report cost per verified outcome and its trend, the same way you'd report any other investment.
Cowork, OpenClaw, Hermes, and Claude Code today, or bring your own. Agents are separate from projects: set one up once and point it at any project, or put several agents on the same one.
Per workspace, never per person, with unlimited seats on every plan. Current tiers and trials are in the pricing table above.
Yes. Bring your own model with no markup, and connect the tools you already use (WhatsApp, Slack, GitHub, Linear, ClickUp) through MCP.
Every environment is sandboxed, encrypted, budget-capped, and fully auditable. The environment keeps the record, not the agent, so the agent never grades its own homework.
Yes. Blank Canvas gives you a full dev workspace: start from scratch or import from GitHub, pick your runtime, watch it run. The docs cover setup end to end.