llms.txt
# Cloud World Model Platform
A cloud world model is a computational representation of cloud infrastructure that agents and teams can step through before deploying or changing infrastructure. Cloud World Model (CWM) simulates resources, traffic, failures, and scaling actions across AWS, GCP, Azure, OCI, and DigitalOcean, returning modeled cost, latency, errors, and resilience in a safe sandbox. MCP and REST are access interfaces; x402 is payment for eligible REST requests. Simulations do not provision resources in a user's cloud account.
> CWM models hypothetical infrastructure before deployment or change; it is not a cloud provisioning, monitoring, or live fault-injection product. FinOps calculators estimate price; CWM also models performance and resilience, not invoices. Simulated chaos can inform hypotheses, but production-capable validation may still be needed.
## Product
- **What it is**: A virtual multi-cloud world model for comparing modeled infrastructure outcomes before deploying or changing infrastructure.
- **Intended users**: Primarily AI agents and agent builders; also infrastructure and platform teams, architects, learners, and reinforcement-learning users.
- **Interfaces and payment**: MCP and REST provide access; x402 is payment for eligible REST requests only.
- **Cloud providers supported**: AWS, Google Cloud (GCP), Microsoft Azure, Oracle Cloud (OCI), DigitalOcean.
- **Key capabilities**: Virtual resource graphs, traffic and scaling actions, simulated failures, multi-cloud exploration, RL episodes, GitHub PR Reports, and modeled cost, latency, errors, and resilience.
- **Evidence boundary**: Results are estimates, not quotes, bills, SLOs, live telemetry, or proof of production behavior. CWM does not create or manage cloud resources or make automatic ship decisions.
## Pages
- [Documentation hub](https://www.cloudworldmodel.ai/docs) - Documentation hub and developer entry point
- [Compare Cloud Infrastructure Tools](https://www.cloudworldmodel.ai/compare) - An unranked scope comparison of CWM, Pinpole, Infracost, and Gremlin
- [Tool Comparison: September 2026](https://www.cloudworldmodel.ai/compare/2026-09-29) - A dated, source-linked comparison of CWM, Pinpole, Infracost and Gremlin workflows, with evidence boundaries and a reproducible evaluation checklist.
- [What is a cloud world model?](https://www.cloudworldmodel.ai/glossary) - Agent-first definition of a cloud world model and qualified comparisons with FinOps, monitoring, and production fault injection
- [Frequently asked questions](https://www.cloudworldmodel.ai/faq) - Answers about CWM, GitHub PR Reports, agent access, evidence limits, pricing, chaos, RL, and cloud-account boundaries
- [GitHub PR Reports](https://www.cloudworldmodel.ai/docs/github-pr-reports) - Setup guide, supported infrastructure files, report evidence boundaries, and sharing.
- [About Cloud World Model](https://www.cloudworldmodel.ai/about) - About Cloud World Model
- [Pre-provision Cloud Simulation](https://www.cloudworldmodel.ai/guides/pre-provision-multi-cloud-simulation) - Explore multi-cloud designs before deploying or changing infrastructure (including greenfield, before provisioning).
- [RL environments for autoscaling agents](https://www.cloudworldmodel.ai/guides/rl-autoscaling-environments) - Learn about RL autoscaling simulation environments
- [Chaos Without Production Changes](https://www.cloudworldmodel.ai/guides/chaos-without-production) - Step through modeled instance failure and peak traffic, compare before/during/after metrics, and plan guarded AWS FIS or Gremlin validation; results are estimates, not production measurements.
- [Simulate the Cloud Without the Cloud](https://www.cloudworldmodel.ai/) - Platform overview and getting started
- [Get Started](https://www.cloudworldmodel.ai/getting-started) - Quickstart guide for new users
- [Simulation API Walkthrough](https://www.cloudworldmodel.ai/simulation-walkthrough) - End-to-end tutorial chaining all core simulation API calls in sequence
- [Pre-provision decisions for AI agents](https://www.cloudworldmodel.ai/use-cases) - Example use cases for learners and AI agents
- [Scenario Library](https://www.cloudworldmodel.ai/scenarios) - Pre-built and saved simulation scenarios
- [AI Agents & MCP Server](https://www.cloudworldmodel.ai/agents) - RL agent dashboard and API key management
- [Pricing](https://www.cloudworldmodel.ai/pricing) - Free tier, credit packs, and cost per API call
- [Historical Cloud Pricing Data: AWS, GCP, Azure, OCI, DigitalOcean](https://www.cloudworldmodel.ai/pricing-history) - Historical list-price data for AWS, GCP, Azure, OCI and DigitalOcean
- [Simulation Fidelity](https://www.cloudworldmodel.ai/fidelity) - Coverage, assumptions, and evidence limits for modeled behaviors
- [Cloud Regions & Zones: AWS, GCP, Azure, OCI, DigitalOcean | CWM](https://www.cloudworldmodel.ai/regions) - Region and zone codes with catalog-backed locations where available for AWS, GCP, Azure, OCI, and DigitalOcean. Compare supported regions in simulation.
- [Benchmark Report](https://www.cloudworldmodel.ai/benchmark) - Full multi-cloud pricing and performance benchmark report comparing AWS, GCP, Azure, OCI, and DigitalOcean
- [Simulation Accuracy Benchmark](https://www.cloudworldmodel.ai/accuracy) - Live methodology and current evidence for the workloads and metrics CWM has evaluated; results are workload-specific
## Accuracy
- [Accuracy benchmark dataset (JSON)](https://www.cloudworldmodel.ai/accuracy.json) - Current scores, campaign scopes, frozen engine versions, dates, methodology, and cwm-bench source references.
- [RL Environments for Cloud & DevOps Agents](https://www.cloudworldmodel.ai/rl/environments) - Train agents in a simulated cloud using documented observations, actions, episode settings, reward weights, and create/reset/step APIs; includes batch-step, validate-action, eval-episodes, FAQs, and sim-to-real limits.
- [Provider Coverage](https://www.cloudworldmodel.ai/provider-coverage) - Every service, instance size, and pricing model the simulator covers for AWS, GCP, Azure, OCI, and DigitalOcean — counts and base rates derived from the live pricing constants
- [Privacy Policy](https://www.cloudworldmodel.ai/privacy) - Privacy policy
- [Terms of Service](https://www.cloudworldmodel.ai/terms) - Terms of service
- [Blog](https://www.cloudworldmodel.ai/blog) - Insights from building Cloud World Model — product launches, engineering decisions, and what we learned along the way.
- [Cloud Pricing Trends September 2026](https://www.cloudworldmodel.ai/blog/cloud-pricing-trends-september-2026) - September 2026: no list-price changes among 33 tracked AWS, GCP, Azure, OCI and DigitalOcean rates, but modeled costs still rose with traffic.
- [Adding App Servers on AWS: Measured P95](https://www.cloudworldmodel.ai/blog/aws-app-server-scaling-measured-p95) - Measured on AWS: at 300 RPS, p95 fell from 94.7 ms on one app server to 38.2 ms on two and 17.9 ms on three. Costs, model results and open questions.
- [What We Learned from Product Hunt](https://www.cloudworldmodel.ai/blog/what-we-learned-product-hunt) - Retrospective on the Product Hunt launch: key takeaways from finishing #5 Product of the Day and 100+ user comments
- [x402 Cloud Simulation API](https://www.cloudworldmodel.ai/blog/x402-cloud-simulation-api) - How we built native x402 payable API support so AI agents can run simulations and pay per request — no human-in-the-loop billing setup required.
- [Owned AWS Accuracy Benchmark](https://www.cloudworldmodel.ai/blog/owned-aws-accuracy-benchmark) - Historical methodology note; use the live accuracy page for current evidence and limitations.
- [Inside a Cloud Simulation Step](https://www.cloudworldmodel.ai/blog/cloud-simulation-step-internals) - A technical walkthrough of the step-hybrid engine: request body, state transitions, metric outputs, and ML/rules blending to a final recommendation.
- [Cloud Pricing Trends: August 2026](https://www.cloudworldmodel.ai/blog/cloud-pricing-trends-august-2026) - We verify cloud pricing monthly across AWS, GCP, Azure, OCI, and DigitalOcean. Here's what changed in August 2026 and what it means for your budget.
- [MCP Server Launch](https://www.cloudworldmodel.ai/blog/mcp-server-launch) - Simulate cloud infrastructure and estimate cost before deploying or changing infrastructure with Cloud World Model's hosted Streamable HTTP MCP server.
- [Cloud Simulation & Cost MCP Server](https://www.cloudworldmodel.ai/mcp-quickstart) - Connect any MCP client to the hosted Streamable HTTP endpoint at https://www.cloudworldmodel.ai/mcp; 9 keyless demo tools and 63 authenticated tools.
## Examples
- [End-to-end agent example](https://github.com/canvascloudai/cwm-mcp/tree/main/examples/agent-example) - Working end-to-end agent example that simulates an infrastructure change before deployment in two parts: a small REST simulation paid per call with x402 on Base ($0.003, verified with a real paid run), and an advanced keyless MCP walkthrough showing a Redis cache failure before/after.
## x402 Payment Support
Cloud World Model supports the **x402 pay-per-request protocol** (a Linux Foundation project). AI agents can call metered API endpoints without an account or API key by selecting an advertised USDC payment option. Base uses `X-PAYMENT`; optional Solana mainnet uses `PAYMENT-SIGNATURE` only when advertised in the live configuration. Forty-two call types cost 3 credits ($0.0030) each, and six cost 5 credits ($0.0050) each.
- **Payment discovery**: Call `GET https://www.cloudworldmodel.ai/api/billing/x402/config` first. Its `paymentOptions` array is authoritative: select an advertised option and use its exact `network`, `asset`, `payTo`, and `header`; include `feePayer` when supplied.
- **Legacy fields**: Top-level `network`, `asset`, and `payTo` are legacy Base-primary compatibility fields. Use `paymentOptions` to choose a payment method; do not assume a Solana option is available unless advertised.
- **Price discovery**: The config endpoint also returns the enabled flag, facilitator URL, and per-call-type price table.
- **Agent-facing OpenAPI subset**: `GET https://www.cloudworldmodel.ai/.well-known/x402/openapi.json` — curated OpenAPI spec covering paid endpoints plus selected free lifecycle routes.
- **Paid call types**: 42 call types cost $0.0030 (3 credits) each; six call types cost $0.0050 (5 credits) each. Fetch the discovery endpoint for the exact per-call table.
- **Standard paid call types (42)**: `ai_analysis`, `ai_bottleneck`, `ai.explain`, `ai.optimize`, `ai.recommendations`, `ai.results`, `ai.status`, `ai.troubleshoot`, `benchmark.validate`, `multicloud_partial_results`, `multicloud_results`, `multicloud_stream`, `multicloud.status`, `provider_api_limits`, `right_sizing_hint`, `rl_env_create`, `rl_env_delete`, `rl_env_get`, `rl_env_observation`, `rl_env_reset`, `rl_eval_status`, `rl.batch_step`, `rl.eval`, `rl.step`, `simulation_cost_breakdown`, `simulation_create`, `simulation_get`, `simulation_list`, `simulation_step_hybrid`, `simulation.apply_right_sizing`, `simulation.deploy`, `simulation.inject_failure`, `simulation.inject_failure_create`, `simulation.inject_failure_delete`, `simulation.inject_failure_update`, `simulation.inject_traffic`, `simulation.recover_resource`, `simulation.resize`, `simulation.stateless`, `validate_cost_accuracy`, `validate_performance_accuracy`, `wallet_session`.
- **Higher-priced paid call types (6)**: `chaos_batch`, `chaos.run`, `multicloud.explore`, `optimization.run`, `prediction.optimize_thresholds`, `prediction.validate`.
- **Payment headers**: Use the selected option's `header` — `X-PAYMENT` for Base, or `PAYMENT-SIGNATURE` for optional Solana mainnet only when advertised.
- **No signup required**: No account or API key is needed for x402 callers; send the selected option's signed payment payload in its specified header.
- **Balance & history**: `GET /api/billing/x402/balance?address=0x...` and `GET /api/billing/x402/transactions?address=0x...`
- **x402scan listing**: https://www.x402scan.com/server/1a8052f0-c764-420a-96a5-f50d3c696795 — browse and call all metered endpoints directly from the x402scan discovery index.
## API
- [RL Training API](https://www.cloudworldmodel.ai/agents) - RESTful API for reinforcement learning agents to interact with the autoscaling simulator; supports episode lifecycle, multi-objective rewards, and secure API key auth. `POST /rl/environments/{id}/batch-step` accepts up to 30 actions per request (each counts as 1 call against the 5 000 req/hr RL quota) and returns an array of step results, stopping early when the episode ends. All 429 responses include a `Retry-After` header with the exact seconds until the window resets. The `add_resource` step action accepts AWS database targets including `aurora-postgresql`, `aurora-dsql`, and `aurora-serverless` (Aurora Serverless v2, scales ACUs with load). See `llms-full.txt` for a complete Python episode-loop quickstart showing how to handle the 2–4 step ACU ramp delay and read steady-state `cost_usd_hr` after the resource reaches operating capacity.
- [Chaos Engineering API](https://www.cloudworldmodel.ai/agents) - Inject failures (database crash, zone outage, network partition) and receive resilience scores and remediation recommendations. Aurora Serverless v2 (`aurora-serverless`) is a valid `database_crash` / `database_slowdown` target alongside RDS and provisioned Aurora; after a `database_crash` the result models a 2–4 step ACU warm-up before full throughput resumes.
- [Multi-Cloud Strategy API](https://www.cloudworldmodel.ai/agents) - Generate and compare multi-cloud architecture variants with cost, latency, and vendor lock-in analysis. When exploring AWS-anchored strategies, use `serviceFamily: "aurora-serverless"` to include Aurora Serverless v2 as a cost-aware database candidate — it scales ACUs with load rather than being fixed at peak provisioned size, and typically produces a lower `cost_usd_hr` estimate than equivalent provisioned `aurora-postgresql` variants (e.g. ~$0.09/hr vs ~$0.15/hr for a variable workload at average load); note: ACU ramp-up from near-zero can add 1–5 s of latency on the first request after an idle period — avoid this option for latency-sensitive workloads that cannot tolerate cold-start delays.
- [Predictive Scaling API](https://www.cloudworldmodel.ai/agents) - Validate infrastructure against traffic forecasts and optimize autoscaling thresholds.
- [Webhook/Callback System](https://www.cloudworldmodel.ai/agents) - Asynchronous job completion notifications via HMAC-signed HTTPS callbacks.
- [Simulation Events API](https://www.cloudworldmodel.ai/agents) - Annotate or clear a simulation's event log: POST /simulations/{id}/events injects a manual annotation; DELETE /simulations/{id}/events clears all events.
- [Simulation Claim API](https://www.cloudworldmodel.ai/agents) - Assign ownership of an unowned (demo) simulation to your API key: POST /simulations/{id}/claim transfers a keyless simulation to the authenticated caller, enabling unlimited steps and full write access. Returns 409 if already owned by another key.
- [Simulation Snapshot API](https://www.cloudworldmodel.ai/agents) - Single-call run object for external tools: GET /simulations/{id}/snapshot returns a versioned JSON envelope with architecture, latest metrics, active failures, last 10 significant events, and _links to detailed endpoints — no stitching required. schemaVersion stays at 2. POST /simulations/{id}/snapshots pins the current state and returns a `pinnedSnapshotId`; pass it as `?since=<pinnedSnapshotId>` to GET /snapshot to receive a `diff` block showing architecture adds/removes, metric deltas (from/to/delta), failure changes, and recommendation changes since the pin.
- [Provider API-Limit Simulation](https://www.cloudworldmodel.ai/mcp-quickstart) - Authenticated read-only REST and MCP operation for bounded provider quotas, throttling, retries, backoff, queueing, worker/concurrency limits, and control-plane workloads. Uses catalog-backed policies or caller-supplied scoped overrides, shares a 1,000,000-event aggregate sweep budget across candidates, enforces a 2-second wall-clock deadline, allows two concurrent executions, returns non-retryable 409 `PROVIDER_LIMIT_BUSY` for a third request and non-retryable 409 `PROVIDER_LIMIT_DUPLICATE` for an identical active/recently interrupted request guarded for five minutes, returns provenance and stable `unsupported_policy` results for undocumented tuples, and does not call a cloud API, discover live quotas, create resources, advance `/step`, or represent CWM HTTP 429s.
## MCP Demo Tools (Smithery)
MCP and REST are access interfaces; x402 is payment for eligible REST requests, not MCP calls. Paid stateless simulation is REST-only at `POST /api/simulations/stateless`: MCP transport cannot preserve the x402 payment challenge, Base/Solana settlement header, and exact-once replay semantics. Keyless MCP provides 9 demo tools; authenticated MCP exposes 63 tools, including the provider-limit operation and candidate resilience verdict.
Nine keyless demo tools are available via the hosted Streamable HTTP MCP server. An API key unlocks all 63 authenticated tools. Send `x-api-key: <api-key>` on initialization and subsequent session requests; invalid, malformed, or unsupported credentials return HTTP 401 rather than switching to keyless demo access.
- `api.spec` — Return the OpenAPI spec URL and format. (read-only, no input)
- `scenario.list` — List compact cards from the live scenario library; call `scenario.get` with a returned ID to hydrate resources and connections. (read-only, no input)
- `scenario.get` — Hydrate one selected scenario, including the full resources/connections graph needed by `simulation.create`. (read-only, input: scenarioId)
- `simulation.create` — Create a demo simulation from a scenario or custom resources/connections; returns a simulation ID. For a custom cache path, use a resource with type "cache", characteristics.serviceFamily "elasticache", cacheHitRate as a fraction (0.8, not 80), and optional cacheableRequestShare. Connect the load balancer to each app server, each app server to the cache and database, and the cache to the database. Use responseMode "full" to see metrics.requestFlow cache hit/miss RPS and database offered RPS. (write, inputs: name, scenarioId or resources/connections, traffic)
- `simulation.step` — Advance the simulation one step; returns latency, CPU, throughput, error rate, and cost metrics. (write, input: simulationId)
- `simulation.metrics` — Read current metrics and resource health for a demo simulation. (read-only, input: simulationId)
- `simulation.inject_traffic` — Set the simulation's traffic level in RPS. (write, inputs: simulationId, `targetRps` for absolute level or `deltaPercent` for relative change; omit both only when a random uncontrolled spike is intentional)
- `simulation.inject_failure` — Inject a reversible failure into a demo resource. (write, input: simulationId and optional resource selector)
- `simulation.recover_resource` — Start recovery for a failed resource, then use `simulation.step` until its recovery progress is healthy. (write, inputs: simulationId and resourceId or resourceName)
- `simulation.delete` — Permanently delete an owned demo simulation and free its slot. (destructive write, input: simulationId)
Anonymous sessions are capped at 3 simulations, 20 steps each, and 1000 RPS max traffic.
## WebMCP Browser Tools
AI agents navigating the site in a WebMCP-enabled browser (Chrome with `navigator.modelContext`) can invoke these tools directly on any page:
- `run_simulation_step` — Advance the active simulation by one tick and return the updated metrics snapshot.
- `reset_simulation` — Pause the active simulation and reset traffic to zero.
- `simulation.inject_traffic_pattern` — Inject a named traffic pattern (ramp, burst, step
