Pre-provision decisions

    Decide What to Ship Before You Provision

    First for AI agents and agent builders; also for infrastructure and platform teams and cloud architects. Compare estimated cost/hr, latency, error rate, and resilience before provisioning real cloud resources.

    Estimated cost/hr
    Latency
    Error rate
    Resilience

    Use Case 1

    AI agents that act on infra

    Give agents a callable world model (MCP, REST, x402) to compare infrastructure actions using estimated cost/hr, latency, error rate, and resilience — without provisioning real cloud resources.

    Which AI Agents Benefit Most

    Autoscaling RL Agents
    Best fit
    Agents that learn when to scale compute up or down by training on thousands of simulated traffic episodes. The Gym-compatible step/reset loop maps directly to standard RL frameworks like Stable-Baselines3 or RLlib.
    FinOps & Cost Optimization Agents
    Best fit
    Agents that reason over multi-cloud pricing, utilization data, and SLA constraints to recommend or automatically apply the cheapest viable architecture. The multi-cloud explore API returns scored strategies with full cost breakdowns.
    Resilience & Chaos Agents
    Best fit
    Agents that systematically probe an architecture for single points of failure by iterating over failure scenarios and measuring resilience scores. Useful for red-team automation and compliance validation pipelines.
    Capacity Planning Agents
    Best fit
    Agents that ingest traffic forecasts (from monitoring, analytics, or demand models) and validate whether current infrastructure will hold. They can also tune autoscaling thresholds automatically based on predicted headroom.
    Multi-Cloud Routing Agents
    Good fit
    Agents that continuously evaluate provider cost and latency signals to decide how to split traffic across AWS, GCP, Azure, and DigitalOcean. The simulator lets them test routing strategies without real provider commitments.
    LLM Infrastructure Advisors
    Good fit
    LLM-based agents that interpret simulation metrics, event logs, and AI coaching output to produce natural-language recommendations for human operators. The AI insights and optimization endpoints provide the structured context they need.
    Connect via MCP
    Without this platform

    Exploring infrastructure actions against live resources can incur cloud costs and affect real user traffic.

    With this platform

    Use simulated traffic patterns to compare actions and estimated outcomes. Check model assumptions and validate the resulting recommendations against your own workload before rollout.

    Explore agent actions without provisioning cloud resources

    Key Endpoints

    POST /api/simulations
    POST /api/keys
    POST /api/rl/environments
    POST /api/rl/environments/{id}/step
    POST /api/rl/environments/{id}/reset
    GET /api/rl/environments/{id}/observation

    Quick Start

    # Requires Bash, curl and jq. Set CWM_BASE_URL to your CWM origin and CWM_API_KEY to a key with read/write scopes and available credits. These are authenticated, metered API calls, not the no-key demo.
    # Creates an owned simulation and a cloned RL environment, then takes one scale-out step (not a full training loop).
    set -euo pipefail
    : "${CWM_BASE_URL:?Set CWM_BASE_URL}"
    : "${CWM_API_KEY:?Set CWM_API_KEY}"
    CWM_BASE_URL="${CWM_BASE_URL%/}"
    request() {
    curl --fail-with-body -sS --max-time 60 "$CWM_BASE_URL$2" \
    -X "$1" -H "Authorization: Bearer $CWM_API_KEY" \
    -H "Content-Type: application/json" "${@:3}"
    }
    sim=$(request POST "/api/simulations" --data "{
    \"name\": \"use-case-lab\",
    \"traffic\": 100,
    \"resources\": [
    {
    \"id\": \"web\",
    \"type\": \"compute\",
    \"name\": \"Web server\",
    \"provider\": \"aws\",
    \"location\": {
    \"regionKey\": \"us-east-1\",
    \"zoneKey\": \"us-east-1a\"
    },
    \"characteristics\": {
    \"size\": \"t3.medium\",
    \"autoscaling\": false
    }
    }
    ],
    \"connections\": []
    }")
    SIM_ID=$(jq -er '.id' <<< "$sim")
    env=$(request POST "/api/rl/environments" --data "{
    \"simulationId\": \"$SIM_ID\",
    \"episodeConfig\": {
    \"maxSteps\": 1000,
    \"initialTraffic\": 100,
    \"enableFailures\": false
    }
    }")
    ENV_ID=$(jq -er '.environment.id' <<< "$env")
    result=$(request POST "/api/rl/environments/$ENV_ID/step" --data "{
    \"action\": {
    \"type\": \"scale_out\",
    \"parameters\": {
    \"instanceCount\": 1
    }
    }
    }")
    jq . <<< "$result"

    Use Case 2

    Infrastructure and platform teams

    Review proposed architecture and scaling changes before rollout, with modeled cost, latency, errors, and resilience evidence in the places you already work (PR, ticket, Slack).

    Without this platform

    Discovering under-provisioning during a real launch or promotion, scrambling to scale reactively, and absorbing the revenue impact of a degraded user experience.

    With this platform

    Run what-if scenarios against any traffic shape in seconds. Get specific SLA violation windows and recommended threshold values before a single line of config changes in production.

    Catch capacity gaps before launch day, not after

    Key Endpoints

    POST /api/predictions/validate
    POST /api/predictions/optimize-thresholds
    GET /api/predictions/jobs/{jobId}
    GET /api/predictions/jobs/{jobId}/results

    Quick Start

    # Requires Bash, curl and jq. Set CWM_BASE_URL to your CWM origin and CWM_API_KEY to a key with read/write scopes and available credits. These are authenticated, metered API calls, not the no-key demo.
    # Validates a traffic forecast. For threshold recommendations, use /api/predictions/optimize-thresholds with the same body instead.
    set -euo pipefail
    : "${CWM_BASE_URL:?Set CWM_BASE_URL}"
    : "${CWM_API_KEY:?Set CWM_API_KEY}"
    CWM_BASE_URL="${CWM_BASE_URL%/}"
    request() {
    curl --fail-with-body -sS --max-time 60 "$CWM_BASE_URL$2" \
    -X "$1" -H "Authorization: Bearer $CWM_API_KEY" \
    -H "Content-Type: application/json" "${@:3}"
    }
    sim=$(request POST "/api/simulations" --data "{
    \"name\": \"use-case-lab\",
    \"traffic\": 100,
    \"resources\": [
    {
    \"id\": \"web\",
    \"type\": \"compute\",
    \"name\": \"Web server\",
    \"provider\": \"aws\",
    \"location\": {
    \"regionKey\": \"us-east-1\",
    \"zoneKey\": \"us-east-1a\"
    },
    \"characteristics\": {
    \"size\": \"t3.medium\",
    \"autoscaling\": false
    }
    }
    ],
    \"connections\": []
    }")
    SIM_ID=$(jq -er '.id' <<< "$sim")
    job=$(request POST "/api/predictions/validate" --data "{
    \"simulationId\": \"$SIM_ID\",
    \"trafficForecast\": {
    \"name\": \"Launch ramp\",
    \"dataPoints\": [
    {
    \"timestamp\": 0,
    \"rps\": 100
    },
    {
    \"timestamp\": 300,
    \"rps\": 12000
    },
    {
    \"timestamp\": 3900,
    \"rps\": 12000
    }
    ]
    },
    \"testSteps\": 100
    }")
    JOB_ID=$(jq -er '.job.id' <<< "$job")
    completed=false
    for ((attempt=0; attempt<300; attempt++)); do
    status=$(request GET "/api/predictions/jobs/$JOB_ID")
    state=$(jq -er '.status' <<< "$status")
    case "$state" in
    completed) completed=true; break ;;
    failed|cancelled) echo "$status" >&2; exit 1 ;;
    esac
    sleep 2
    done
    if [[ "$completed" != true ]]; then echo "Polling timed out; job may still be running." >&2; exit 1; fi
    result=$(request GET "/api/predictions/jobs/$JOB_ID/results")
    jq . <<< "$result"

    Use Case 3

    Cost and right-sizing

    Run right-sizing on simulated workloads. Compare instance types, spot over-provisioning, and quantify savings before changing real infra.

    Without this platform

    Changing instance sizes in production before comparing cost savings with the throughput and capacity trade-offs.

    With this platform

    Run an owned simulation with representative traffic, then compare current instance sizes with proposed sizes, estimated hourly rates, savings percentages, and trade-off notes. Recommendations are modeled estimates; test peak traffic before applying a change.

    Compare estimated savings without provisioning cloud resources

    Key Endpoints

    GET /api/simulations/{id}
    POST /api/simulations/{id}/step
    GET /api/simulations/{id}/metrics
    GET /api/simulations/{id}/right-sizing-hint

    Quick Start

    # Requires curl and jq. Set CWM_BASE_URL to your CWM origin,
    # CWM_API_KEY to your key, and SIM_ID to an owned simulation.
    # First run that simulation with representative traffic and inspect its metrics.
    : "${CWM_BASE_URL:?Set CWM_BASE_URL}"
    : "${CWM_API_KEY:?Set CWM_API_KEY}"
    : "${SIM_ID:?Set SIM_ID}"
    # 1. Read current instance sizes (no resources changed)
    simulation=$(curl --fail-with-body -sS \
    "$CWM_BASE_URL/api/simulations/$SIM_ID" \
    -H "Authorization: Bearer $CWM_API_KEY") || exit 1
    # 2. Get modeled right-sizing recommendations
    hints=$(curl --fail-with-body -sS \
    "$CWM_BASE_URL/api/simulations/$SIM_ID/right-sizing-hint" \
    -H "Authorization: Bearer $CWM_API_KEY") || exit 1
    # 3. Compare current vs. proposed sizes and estimated savings
    jq --argjson sim "$simulation" '
    if .hasHint then
    .hints[] | . as $hint |
    ($sim.resources | map(select(.id == $hint.affectedResourceId)) | first) as $resource |
    {resource: .affectedResourceName,
    currentSize: $resource.characteristics.size,
    proposedSize: .recommendedSlug,
    estimatedHourlyRate: .hourlyRate,
    estimatedSavingsPct: .estimatedSavingsPct,
    reason, tradeOffNote}
    else {message: "No right-sizing recommendation for the current modeled state."}
    end' <<< "$hints"
    # Savings are estimates, not a bill. Review trade-offs and test peak traffic.
    # A cross_provider_alternative is a migration option, not a direct resize.

    Use Case 4

    Multi-cloud strategy

    Generate and score architecture variants across AWS, GCP, Azure, OCI, and DigitalOcean — cost, latency, and resilience side by side.

    Without this platform

    Running parallel production workloads on multiple providers for weeks, or relying on rough estimates that miss provider-specific pricing nuances and latency trade-offs.

    With this platform

    Evaluate every provider combination and traffic-split ratio in minutes. Identify potential, model-estimated savings for your workload while comparing detailed per-strategy cost, latency, and vendor lock-in scores.

    Model-estimated savings vs. the most expensive modeled single-provider candidate

    Key Endpoints

    POST /api/multi-cloud/explore
    GET /api/multi-cloud/jobs/{jobId}
    GET /api/multi-cloud/jobs/{jobId}/results

    Quick Start

    # Requires Bash, curl and jq. Set CWM_BASE_URL to your CWM origin and CWM_API_KEY to a key with read/write scopes and available credits. These are authenticated, metered API calls, not the no-key demo.
    # Explores an explicit workload profile; this endpoint does not consume a simulationId. Costs are modeled estimates.
    set -euo pipefail
    : "${CWM_BASE_URL:?Set CWM_BASE_URL}"
    : "${CWM_API_KEY:?Set CWM_API_KEY}"
    CWM_BASE_URL="${CWM_BASE_URL%/}"
    request() {
    curl --fail-with-body -sS --max-time 60 "$CWM_BASE_URL$2" \
    -X "$1" -H "Authorization: Bearer $CWM_API_KEY" \
    -H "Content-Type: application/json" "${@:3}"
    }
    job=$(request POST "/api/multi-cloud/explore" --data "{
    \"workloadProfile\": {
    \"computeInstances\": 8,
    \"databaseInstances\": 1,
    \"storageGB\": 100,
    \"trafficRPS\": 5000,
    \"latencyRequirementMs\": 100,
    \"primaryRegion\": \"us-east-1\",
    \"sourceProvider\": \"aws\"
    },
    \"optimizationWeights\": {
    \"cost\": 0.5,
    \"latency\": 0.3,
    \"vendorLockIn\": 0.2
    }
    }")
    JOB_ID=$(jq -er '.job.id' <<< "$job")
    completed=false
    for ((attempt=0; attempt<300; attempt++)); do
    status=$(request GET "/api/multi-cloud/jobs/$JOB_ID")
    state=$(jq -er '.status' <<< "$status")
    case "$state" in
    completed) completed=true; break ;;
    failed|cancelled) echo "$status" >&2; exit 1 ;;
    esac
    sleep 2
    done
    if [[ "$completed" != true ]]; then echo "Polling timed out; job may still be running." >&2; exit 1; fi
    result=$(request GET "/api/multi-cloud/jobs/$JOB_ID/results")
    jq . <<< "$result"

    Use Case 5

    Failure experiments

    Inject AZ outages, database failures, and latency spikes to evaluate cost, latency, errors, and resilience before rollout. Chaos is the method; the outcomes are the decision.

    Without this platform

    Expensive game days with real production risk, manual failure simulation, and no repeatable way to measure resilience scores across architecture changes.

    With this platform

    Inject supported failures into a virtual architecture. Review modeled resilience scores, vulnerabilities, and remediation recommendations alongside estimated cost, latency, and errors — without touching production.

    No production incidents — find weak points safely

    Key Endpoints

    GET /api/chaos/scenarios
    POST /api/chaos/run
    GET /api/chaos/jobs/{jobId}
    GET /api/chaos/jobs/{jobId}/results
    POST /api/chaos/batch
    GET /api/chaos/batch/{batchId}/results

    Quick Start

    # Requires Bash, curl and jq. Set CWM_BASE_URL to your CWM origin and CWM_API_KEY to a key with read/write scopes and available credits. These are authenticated, metered API calls, not the no-key demo.
    # Creates an owned single-region simulation and tests a modeled AZ outage. No real cloud resources are changed.
    set -euo pipefail
    : "${CWM_BASE_URL:?Set CWM_BASE_URL}"
    : "${CWM_API_KEY:?Set CWM_API_KEY}"
    CWM_BASE_URL="${CWM_BASE_URL%/}"
    request() {
    curl --fail-with-body -sS --max-time 60 "$CWM_BASE_URL$2" \
    -X "$1" -H "Authorization: Bearer $CWM_API_KEY" \
    -H "Content-Type: application/json" "${@:3}"
    }
    sim=$(request POST "/api/simulations" --data "{
    \"name\": \"use-case-lab\",
    \"traffic\": 100,
    \"resources\": [
    {
    \"id\": \"web\",
    \"type\": \"compute\",
    \"name\": \"Web server\",
    \"provider\": \"aws\",
    \"location\": {
    \"regionKey\": \"us-east-1\",
    \"zoneKey\": \"us-east-1a\"
    },
    \"characteristics\": {
    \"size\": \"t3.medium\",
    \"autoscaling\": false
    }
    }
    ],
    \"connections\": []
    }")
    SIM_ID=$(jq -er '.id' <<< "$sim")
    job=$(request POST "/api/chaos/run" --data "{
    \"simulationId\": \"$SIM_ID\",
    \"scenarioId\": \"zone_failure\",
    \"duration\": 300
    }")
    JOB_ID=$(jq -er '.job.id' <<< "$job")
    completed=false
    for ((attempt=0; attempt<300; attempt++)); do
    status=$(request GET "/api/chaos/jobs/$JOB_ID")
    state=$(jq -er '.status' <<< "$status")
    case "$state" in
    completed) completed=true; break ;;
    failed|cancelled) echo "$status" >&2; exit 1 ;;
    esac
    sleep 2
    done
    if [[ "$completed" != true ]]; then echo "Polling timed out; job may still be running." >&2; exit 1; fi
    result=$(request GET "/api/chaos/jobs/$JOB_ID/results")
    jq . <<< "$result"

    Ready to Dive In?

    Explore cloud architectures and stress-test designs in a safe simulator — no real resources provisioned.

    Start with 1,000 free credits every 30 days; no card required.