1# Azure VM Recommender
2 
3Recommend Azure VM sizes, VM Scale Sets (VMSS), and configurations by analyzing workload type, performance requirements, scaling needs, and budget. No Azure subscription required — data comes from public Microsoft documentation and the unauthenticated Retail Prices API.
4 
5## When to Use This Skill
6 
7- User asks which Azure VM or VMSS to choose for a workload
8- User wants to compare VM families, sizes, or pricing tiers
9- User asks about trade-offs (cost vs performance, single VM vs scale set, orchestration modes)
10- User needs a cost estimate without an Azure subscription
11- User asks "Needs autoscaling?" or wants to decide between a single VM and a scale set
12 
13## Workflow
14 
15> Use reference files for initial filtering. Then **verify with live documentation** via `web_fetch` before final recommendations. If `web_fetch` fails, fall back to the reference files and surface the staleness warning from [web-fetch-policy.md](references/web-fetch-policy.md).
16 
17### Step 1: Gather Requirements
18 
19Ask the user (infer when possible):
20 
21| Requirement | Examples |
22|---|---|
23| Workload type | Web server, relational DB, ML training, batch, dev/test |
24| vCPU / RAM needs | "4 cores, 16 GB" or "lightweight" / "heavy" |
25| GPU needed? | Yes → GPU families; No → general / compute / memory |
26| Storage needs | High IOPS, large temp disk, premium SSD |
27| Budget priority | Cost-sensitive, performance-first, balanced |
28| OS | Linux or Windows (affects pricing) |
29| Region | Affects availability and price |
30| Instance count | Single, fixed count, or variable |
31| Scaling needs | None, manual, autoscale (metrics / schedule) |
32| Availability needs | Best-effort, fault-domain, cross-zone HA |
33| Load balancing | None, Azure Load Balancer (L4), Application Gateway (L7) |
34 
35### Step 2: Determine VM vs VMSS
36 
37Review [VMSS Guide](../../references/vmss-guide.md). Decision shortcut — start by asking **Needs autoscaling?** then walk the table:
38 
39| Signal | Pick |
40|---|---|
41| Autoscale on CPU, memory, or schedule | **VMSS** |
42| Stateless web/API tier behind a load balancer | **VMSS** |
43| Batch / parallel processing across many nodes | **VMSS** |
44| Mixed VM sizes in one group | **VMSS (Flexible)** |
45| Single long-lived server (jumpbox, AD DC) | **VM** |
46| Unique per-instance config | **VM** |
47| Stateful, tightly-coupled cluster | **VM** (or VMSS case-by-case) |
48 
49If recommending VMSS, verify with `web_fetch` per [web-fetch-policy.md](references/web-fetch-policy.md). When in doubt, default to a single **VM**.
50 
51### Step 3: Select VM Family
52 
53Review [VM Family Guide](../../references/vm-families.md) and pick 2–3 candidate families. Verify each candidate's specs with `web_fetch` against:
54 
55```
56https://learn.microsoft.com/en-us/azure/virtual-machines/sizes/<family-category>/<series-name>
57```
58 
59For Spot eligibility, also fetch `https://learn.microsoft.com/en-us/azure/virtual-machine-scale-sets/use-spot`. If any fetch fails, follow [web-fetch-policy.md](references/web-fetch-policy.md). Same SKUs apply to single VMs and VMSS.
60 
61### Step 4: Look Up Pricing
62 
63Query the Azure Retail Prices API per [Retail Prices API Guide](../../references/retail-prices-api.md).
64 
65> **VMSS:** no extra charge — pricing is per-VM. Multiply per-instance price × expected count. For autoscale, estimate at both `min` and `max`.
66 
67### Step 5: Validate Quota Availability
68 
69> **GATE — do not present recommendations until quota is validated.**
70 
71If the user has a subscription + region, review and run the checks from [VM Quota Validation Guide](../../references/vm-quotas.md). Without a subscription, note quota must be checked before deployment.
72 
73| Outcome | Action |
74|---|---|
75| ✅ Sufficient | Proceed to Step 6 |
76| ⚠️ Near limit (>80%) | Proceed but warn; suggest quota increase |
77| ❌ Insufficient | Request increase, swap family, or try another region |
78 
79Include a "Quota Status" column (✅/⚠️/❌) in the table.
80 
81### Step 6: Present Recommendations
82 
83Provide **2–3 options** with trade-offs:
84 
85| Column | Purpose |
86|---|---|
87| Hosting Model | VM or VMSS (with orchestration mode if VMSS) |
88| VM Size | ARM SKU name (e.g., `Standard_D4s_v5`) |
89| vCPUs / RAM | Core specs |
90| Instance Count | `1` for VM; `min–max` for VMSS with autoscale |
91| Estimated $/hr | Per-instance pay-as-you-go |
92| Why | Workload fit |
93| Trade-off | What the user gives up |
94 
95Always explain *why* a family fits and the Trade-off (cost vs cores, burstable vs dedicated, VM simplicity vs VMSS scale).
96 
97For VMSS, also mention orchestration mode (default **Flexible**), autoscale strategy (metric / schedule / both), and load balancer type.
98 
99### Step 7: Offer Next Steps
100 
101- Compare reservation / savings plan pricing (`priceType eq 'Reservation'` in the API)
102- Suggest [Azure Pricing Calculator](https://azure.microsoft.com/pricing/calculator/) for full estimates
103- For VMSS: [autoscale best practices](https://learn.microsoft.com/azure/azure-monitor/autoscale/autoscale-best-practices), [VMSS networking](https://learn.microsoft.com/azure/virtual-machine-scale-sets/virtual-machine-scale-sets-networking)
104 
105### Step 8: Hand Off to VM Creator (Optional)
106 
107If the user wants to **actually provision** what was recommended, hand off to [vm-creator](../vm-creator/vm-creator.md). See [handoff-to-creator.md](references/handoff-to-creator.md) for the required Plan Card render and routing rules.
108 
109## Error Handling
110 
111| Scenario | Action |
112|---|---|
113| API returns empty results | Broaden filters — check `armRegionName`, `serviceName`, `armSkuName` spelling |
114| User unsure of workload type | Ask clarifying questions; default to General Purpose D-series |
115| Region not specified | Use `eastus` as default; note prices vary by region |
116| Unclear if VM or VMSS needed | Ask about scaling + instance count; default to single VM if still unsure |
117| User asks VMSS pricing directly | Same VM pricing API; VMSS has no extra charge — multiply by instance count |
118 
119## References
120 
121- [VM Family Guide](../../references/vm-families.md) — family-to-workload mapping
122- [Retail Prices API Guide](../../references/retail-prices-api.md) — query patterns, filters
123- [VMSS Guide](../../references/vmss-guide.md) — when to use VMSS, orchestration, autoscale
124- [VM Quota Validation Guide](../../references/vm-quotas.md) — vCPU checks, CLI commands
125- [web-fetch-policy.md](references/web-fetch-policy.md) — fail-safe behavior for live docs lookups
126- [handoff-to-creator.md](references/handoff-to-creator.md) — Step 8 hand-off rules
127- [vm-creator](../vm-creator/vm-creator.md) — provision the recommended SKU
128