Cloud World Model
Frequently asked questions
Answers about CWM, GitHub PR Reports, modeled outcomes, agent access, accuracy limits, pricing, chaos, and RL.
What is Cloud World Model?
Cloud World Model (CWM) is a multi-cloud world-model sandbox for simulating cost, latency, errors, and resilience before teams deploy or change infrastructure. MCP and REST are ways to access CWM; x402 is a payment mechanism for eligible requests. Simulations use virtual resources and do not provision resources in a user's cloud account.
What do GitHub PR Reports show?
The GitHub App reads supported infrastructure files changed in a pull request, plus bounded support files needed for modeling, at the base and head commits; it does not read cloud credentials, live infrastructure, application code, provider APIs, or production traffic. The report compares both snapshots under the same modeled workload and summarizes Before, After, and Impact for hourly cost, p99 latency, peak CPU, and resilience. Coverage is classified as Exact, Inferred, or Unsupported; a detected construct is not necessarily fully modeled, and unsupported or unmodeled inputs are disclosed rather than represented as unchanged.
Who is Cloud World Model for?
CWM is intended primarily for AI agents and the people who build them. It is also designed for infrastructure and platform teams, cloud architects, learners, and reinforcement-learning users who want to explore modeled outcomes.
How can an AI agent access CWM?
Hosted MCP and REST provide agent access to simulation creation and steps, failure injection, metrics, multi-cloud exploration, and RL episodes. A keyless demo is available; authentication and payment requirements depend on the access path, and x402 payment is supported on eligible requests.
Open the MCP quick start →See pricing and authentication details →
What does “before deploying or changing infrastructure” mean?
CWM lets teams model a hypothetical resource graph before deploying new infrastructure or changing an existing system, including changes proposed in a pull request. Treat modeled results as planning inputs, then validate decisions with your own workloads, provider prices, and telemetry.
Which clouds can CWM compare?
CWM can model AWS, GCP, Azure, OCI, and DigitalOcean side by side as planning aids. Cross-cloud results are estimates, not provider invoices or service-level objectives (SLOs).
Is CWM a production chaos or fault-injection tool?
No. CWM uses chaos as a method inside a virtual simulation to test hypotheses about modeled failures and resilience. It does not replace AWS Fault Injection Service (FIS), Gremlin, or other production-capable experiments; use appropriately safeguarded real-world validation later.
Read the simulated-chaos guide →Review AWS FIS documentation ↗
How accurate are CWM's outputs?
CWM outputs are estimates, not quotes, bills, or SLOs, and their usefulness depends on workload inputs, model coverage, and stated assumptions. The live accuracy page explains the current methodology and evidence without implying that results transfer to every workload.
How do CWM pricing and credits work?
CWM API usage is metered in credits according to the operation, with a free allowance and optional credit packs; virtual simulations do not incur resource charges in your cloud account. Eligible requests can also use x402 pay-per-request. Check the pricing page for current allowances, pack prices, call rates, and access details.
How does CWM compare with Pinpole, Infracost, and Gremlin?
These products address related but different questions: Pinpole and CWM include pre-deployment and pre-change simulation workflows, Infracost focuses on infrastructure-as-code cost estimates, and Gremlin documents reliability experiments on real systems. This is a scope distinction, not a ranking or a claim of comparative accuracy.
What does Cloud World Model not do?
CWM does not provision or manage cloud infrastructure, replace cloud providers, or claim to be a perfect digital twin. It is not a monitoring or operations console and does not make automatic ship-or-don't-ship decisions for users.
Can reinforcement-learning policies transfer directly to production?
RL episodes run against a sandboxed simulation and its modeled observations, actions, and rewards. A learned policy may not transfer unchanged to production, so evaluate it against independent workloads and real telemetry before considering deployment.
Will a CWM simulation change my live cloud account?
No. CWM simulations operate on virtual resources and do not create, modify, or delete resources in your AWS, GCP, Azure, OCI, or DigitalOcean account. They do not call your cloud account to apply simulation actions.