Add Windows (PowerShell) support

- scripts/setup.ps1, scripts/pull-models.ps1, run.ps1 — PowerShell equivalents
  of the bash setup/pull/run scripts
- README + docs/DEPLOY.md: Windows quickstart and Task Scheduler autostart

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

README.md

@@ -61,6 +61,13 @@ git clone <your-fork-url> llm-router && cd llm-router
 ./run.sh
 ```
 
+On **Windows** (PowerShell), use the `.ps1` equivalents:
+```powershell
+powershell -ExecutionPolicy Bypass -File .\scripts\setup.ps1
+powershell -ExecutionPolicy Bypass -File .\scripts\pull-models.ps1
+powershell -ExecutionPolicy Bypass -File .\run.ps1
+```
+
 Then point any OpenAI client at `http://localhost:8080/v1` with `model: "auto"`:
 
 ```bash

docs/DEPLOY.md

@@ -27,6 +27,27 @@ launchctl kickstart -k gui/$(id -u)/com.llm-router
 systemctl --user restart llm-router
 ```
 
+## Windows
+
+```powershell
+powershell -ExecutionPolicy Bypass -File .\scripts\setup.ps1        # venv + LiteLLM
+powershell -ExecutionPolicy Bypass -File .\scripts\pull-models.ps1  # pull models
+powershell -ExecutionPolicy Bypass -File .\run.ps1                  # LiteLLM + router
+```
+`run.ps1` starts LiteLLM in the background, waits for it, then runs the router in
+the foreground; closing it stops both.
+
+**Run at login (Task Scheduler):**
+```powershell
+$action  = New-ScheduledTaskAction -Execute "powershell.exe" `
+  -Argument "-ExecutionPolicy Bypass -WindowStyle Hidden -File `"$PWD\run.ps1`""
+$trigger = New-ScheduledTaskTrigger -AtLogOn
+Register-ScheduledTask -TaskName "llm-router" -Action $action -Trigger $trigger `
+  -Settings (New-ScheduledTaskSettingsSet -RestartCount 3 -RestartInterval (New-TimeSpan -Minutes 1))
+```
+Requires **Python 3.12** (`winget install Python.Python.3.12`) and Ollama for
+Windows. The router (stdlib Python) and LiteLLM run the same as on macOS/Linux.
+
 ## Networking
 
 - The router binds `ROUTER_HOST` (default `0.0.0.0` = reachable on the LAN).

run.ps1

@@ -0,0 +1,39 @@
+# Serve on Windows: LiteLLM (:4000) + router (:8080). Windows equivalent of run.sh.
+#   Run:  powershell -ExecutionPolicy Bypass -File .\run.ps1
+$ErrorActionPreference = "Stop"
+Set-Location $PSScriptRoot
+
+# load keys from files into env (kept out of git)
+if (Test-Path ".apikey")         { $env:ROUTER_API_KEY        = (Get-Content ".apikey" -Raw).Trim() }
+if (Test-Path ".uncensored_key") { $env:ROUTER_UNCENSORED_KEY = (Get-Content ".uncensored_key" -Raw).Trim() }
+if (-not $env:OLLAMA_KEEP_ALIVE)  { $env:OLLAMA_KEEP_ALIVE = "30m" }
+
+# warm the small classifier so ambiguous routing has no cold start (best effort)
+try {
+    Invoke-RestMethod -Method Post -Uri "http://127.0.0.1:11434/api/generate" -ContentType "application/json" `
+        -TimeoutSec 10 -Body '{"model":"qwen3:8b","prompt":"ok","stream":false,"keep_alive":-1}' | Out-Null
+} catch {}
+
+# LiteLLM backend (localhost only)
+Write-Host ">> starting LiteLLM on 127.0.0.1:4000 ..."
+$litellm = Start-Process -FilePath ".\.venv\Scripts\litellm.exe" `
+    -ArgumentList "--config", "litellm.config.yaml", "--host", "127.0.0.1", "--port", "4000" `
+    -RedirectStandardOutput "litellm.log" -RedirectStandardError "litellm.err.log" -NoNewWindow -PassThru
+$litellm.Id | Set-Content ".litellm.pid"
+
+for ($i = 0; $i -lt 60; $i++) {
+    try { Invoke-WebRequest "http://127.0.0.1:4000/health/liveliness" -TimeoutSec 2 -UseBasicParsing | Out-Null; break }
+    catch { Start-Sleep -Seconds 2 }
+}
+
+# Router (network-facing). Trap tears down LiteLLM on exit.
+$env:ROUTER_UPSTREAM = "http://127.0.0.1:4000/v1"
+if (-not $env:ROUTER_HOST) { $env:ROUTER_HOST = "0.0.0.0" }
+Write-Host ">> router on http://$($env:ROUTER_HOST):8080/v1   (use model 'auto')"
+try {
+    & ".\.venv\Scripts\python.exe" router.py
+}
+finally {
+    if ($litellm -and -not $litellm.HasExited) { Stop-Process -Id $litellm.Id -Force -ErrorAction SilentlyContinue }
+    Remove-Item ".router.pid" -ErrorAction SilentlyContinue
+}

scripts/pull-models.ps1

@@ -0,0 +1,27 @@
+# Pull a model fleet into Ollama on Windows. Edit $models to taste.
+# Names must line up with the routing map in router.py (see docs/ROUTING.md).
+#   Run:  powershell -ExecutionPolicy Bypass -File .\scripts\pull-models.ps1
+$models = @(
+    # general / small
+    "qwen3:8b", "qwen3:14b",
+    # reasoning / agentic
+    "qwen3.6:35b-a3b", "gpt-oss:20b", "glm-4.7-flash",
+    # coding
+    "qwen3-coder:30b",
+    # vision / OCR
+    "qwen3-vl:8b", "qwen3-vl:30b-a3b-instruct",
+    # multimodal (vision + audio) — gemma4:e4b is the default ROUTER_AUDIO_MODEL
+    "gemma4:e4b", "gemma4:12b", "gemma4:26b"
+    # example HF GGUF:  "hf.co/USER/REPO:Q4_K_M"
+)
+
+$ok = @(); $fail = @(); $i = 0
+foreach ($m in $models) {
+    $i++
+    Write-Host "=== [$i/$($models.Count)] pulling $m ==="
+    ollama pull $m
+    if ($LASTEXITCODE -eq 0) { $ok += $m } else { Write-Host "!! failed: $m"; $fail += $m }
+}
+Write-Host ""
+Write-Host "done. $($ok.Count)/$($models.Count) succeeded."
+if ($fail.Count) { Write-Host "failed: $($fail -join ', ')" }

scripts/setup.ps1

@@ -0,0 +1,37 @@
+# One-time setup on Windows: create the Python 3.12 venv and install LiteLLM + Pillow.
+# The router itself is stdlib-only; this is for the LiteLLM backend.
+#   Run:  powershell -ExecutionPolicy Bypass -File .\scripts\setup.ps1
+$ErrorActionPreference = "Stop"
+Set-Location (Split-Path $PSScriptRoot -Parent)   # repo root
+
+function Find-Py312 {
+    if (Get-Command py -ErrorAction SilentlyContinue) {
+        try { if ((& py -3.12 --version 2>&1) -match "3\.12") { return @("py", "-3.12") } } catch {}
+    }
+    foreach ($c in @("python3.12", "python")) {
+        if (Get-Command $c -ErrorAction SilentlyContinue) {
+            try { if ((& $c --version 2>&1) -match "3\.12") { return @($c) } } catch {}
+        }
+    }
+    return $null
+}
+
+$py = Find-Py312
+if (-not $py) {
+    Write-Error "Python 3.12 not found. Install it:  winget install Python.Python.3.12   (or python.org)"
+    exit 1
+}
+$exe = $py[0]
+$baseArgs = if ($py.Count -gt 1) { $py[1..($py.Count - 1)] } else { @() }
+Write-Host ">> using Python 3.12 via '$($py -join ' ')'"
+
+if (-not (Test-Path ".venv")) { & $exe @baseArgs -m venv .venv }
+$vpy = ".\.venv\Scripts\python.exe"
+& $vpy -m pip install -q --upgrade pip
+& $vpy -m pip install -q -r requirements.txt
+& $vpy -c "import litellm, PIL; print('>> deps OK (litellm, pillow)')"
+
+Write-Host ""
+Write-Host "Next:"
+Write-Host "  .\scripts\pull-models.ps1    # pull a model fleet (edit the list first)"
+Write-Host "  .\run.ps1                    # start LiteLLM + router"