feat: precalentar el modelo al terminar el deploy

'hola' tardaba en el primer mensaje: la carga a VRAM (2.5 GB a la
GTX 1660) ocurria lazy en el primer request. Ahora el deploy la
dispara al final: 'ollama run qwen3:4b ""' (prompt vacio, carga sin
generar tokens) + keep_alive -1 (nunca expira de VRAM).

- compose: OLLAMA_KEEP_ALIVE 5m -> -1
- verificado e2e: carga confirmada via ollama ps tras el warm

INSTALL.md

@@ -4,14 +4,27 @@
 
 Si ya tienes Docker, salta al paso 2.
 
+**Linux:**
+
 ```bash
 curl -fsSL https://get.docker.com | sh
 ```
 
-Mac o Windows: descarga Docker Desktop desde https://docs.docker.com/desktop/
+**Windows:** descarga [Docker Desktop para Windows](https://docs.docker.com/desktop/setup/install/windows-install/) y ábrelo. Requiere WSL2 (el instalador lo activa por ti si falta).
+
+**Mac:** descarga [Docker Desktop para Mac](https://docs.docker.com/desktop/setup/install/mac-install/).
 
 ## 2. Descargar los packs
 
+**Windows (PowerShell):**
+
+```powershell
+git clone https://git.d0a1.es/d0a1/packs.git
+cd packs
+```
+
+**Linux / Mac:**
+
 ```bash
 git clone https://git.d0a1.es/d0a1/packs.git
 cd packs
@@ -21,8 +34,23 @@ cd packs
 
 Cada pack es una carpeta. Entra en la que quieras y ejecuta el instalador:
 
+> **Windows:** abre PowerShell, entra en la carpeta del pack y ejecuta
+> `.\deploy.ps1 local`. Si PowerShell bloquea el script, usa
+> `powershell -ExecutionPolicy Bypass -File .\deploy.ps1 local`.
+>
+> **Linux / Mac:** ejecuta `./deploy.sh local` en la terminal.
+
 ### Open WebUI — Chat de IA privado
 
+**Windows (PowerShell):**
+
+```powershell
+cd openwebui
+.\deploy.ps1 local
+```
+
+**Linux / Mac:**
+
 ```bash
 cd openwebui
 ./deploy.sh local
@@ -32,6 +60,15 @@ Abre http://localhost:3000 y empieza a chatear. Sin cuenta, sin API key.
 
 ### WordPress + IA
 
+**Windows (PowerShell):**
+
+```powershell
+cd wordpress
+.\deploy.ps1 local
+```
+
+**Linux / Mac:**
+
 ```bash
 cd wordpress
 ```
@@ -48,6 +85,15 @@ Abre http://localhost:8080 (WordPress) y http://localhost:3000 (chat IA).
 
 ### Hermes Agent
 
+**Windows (PowerShell):**
+
+```powershell
+cd hermes
+.\deploy.ps1 local
+```
+
+**Linux / Mac:**
+
 ```bash
 cd hermes
 ./deploy.sh local
@@ -63,6 +109,15 @@ hermes gateway setup  # Conecta Telegram, etc.
 
 ### OpenClaw
 
+**Windows (PowerShell):**
+
+```powershell
+cd openclaw
+.\deploy.ps1 local
+```
+
+**Linux / Mac:**
+
 ```bash
 cd openclaw
 ./deploy.sh local
@@ -78,6 +133,14 @@ openclaw channel add whatsapp
 
 ## 4. Parar cuando no lo uses
 
+**Windows (PowerShell):**
+
+```powershell
+.\deploy.ps1 stop
+```
+
+**Linux / Mac:**
+
 ```bash
 ./deploy.sh stop
 ```
@@ -92,6 +155,14 @@ openclaw channel add whatsapp
 
 Cuando estés listo para tenerlo 24/7 en tu propio servidor:
 
+**Windows (PowerShell):**
+
+```powershell
+.\deploy.ps1 vps
+```
+
+**Linux / Mac:**
+
 ```bash
 ./deploy.sh vps
 ```

hermes/README.md

@@ -29,8 +29,8 @@ hermes gateway setup  # Conecta mensajería (Telegram, etc.)
 Solo pagas el servidor. El agente es tuyo.
 
 **¿No tienes VPS?** Consigue uno con descuento:
-- [Vultr](https://d0a1.es/go/vultr/) — desde ~6€/mes
+- [Vultr](https://d0a1.es/go/vultr/) — desde ~6€/mes (afiliado)
-- [DigitalOcean](https://d0a1.es/go/digitalocean/) — credit gratuito
+- [DigitalOcean](https://d0a1.es/go/digitalocean/) — credit gratuito (afiliado)
 
 ## Requisitos
 

hermes/deploy.ps1

@@ -0,0 +1,234 @@
+# d0a1 pack - Hermes Agent (Windows)
+# Agente de IA desde terminal o Telegram.
+#
+# Local:  .\deploy.ps1 local    -> terminal + http://localhost:11434 (Ollama)
+# VPS:    .\deploy.ps1 vps      -> + gateway accesible desde fuera
+# Stop:   .\deploy.ps1 stop     -> para Ollama
+#
+# Requiere: Docker Desktop para Windows (https://docs.docker.com/desktop/)
+
+param(
+    [Parameter(Position = 0)]
+    [ValidateSet('local', 'vps', 'stop')]
+    [string]$Mode = 'local'
+)
+
+$ErrorActionPreference = 'Stop'
+
+# Escape ANSI via [char]27 (compatible con Windows PowerShell 5.1 y PS 7+)
+$Esc = [char]27
+$Green = "$Esc[32m"; $Cyan = "$Esc[36m"; $Yellow = "$Esc[33m"; $Red = "$Esc[31m"; $Bold = "$Esc[1m"; $NC = "$Esc[0m"
+if ($env:NO_COLOR) {
+    $Green = $Cyan = $Yellow = $Red = $Bold = $NC = ''
+}
+
+function Info($msg) { Write-Host "${Bold}${Green}>>>${NC} $msg" }
+function Warn($msg) { Write-Host "${Bold}${Yellow}>>>${NC} $msg" }
+function Err($msg)  { Write-Host "${Bold}${Red}>>>${NC} $msg" }
+function Ok($msg)   { Write-Host "${Green}OK${NC} $msg" }
+
+function Find-FreePort([int]$Start) {
+    for ($p = $Start; $p -le ($Start + 20); $p++) {
+        $inUse = Get-NetTCPConnection -LocalPort $p -State Listen -ErrorAction SilentlyContinue
+        if (-not $inUse) { return $p }
+    }
+    return $Start
+}
+
+Write-Host ''
+Write-Host "${Bold}${Cyan}============================================================${NC}"
+Write-Host "${Bold}${Cyan}  d0a1 pack - Hermes Agent (Windows)${NC}"
+Write-Host "${Bold}${Cyan}  Agente de IA desde terminal o Telegram.${NC}"
+Write-Host "${Bold}${Cyan}============================================================${NC}"
+Write-Host ''
+
+# --- Stop mode ------------------------------------------------------------
+if ($Mode -eq 'stop') {
+    $running = docker ps --format '{{.Names}}' | Select-String -Quiet 'd0a1-hermes-ollama'
+    if ($running) {
+        docker stop d0a1-hermes-ollama | Out-Null
+        docker rm d0a1-hermes-ollama | Out-Null
+        Info 'Ollama parado.'
+    } else {
+        Write-Host 'Ollama no esta corriendo.'
+    }
+    exit 0
+}
+
+# --- Check Docker (para Ollama) -------------------------------------------
+if (-not (Get-Command docker -ErrorAction SilentlyContinue)) {
+    Err 'Docker no esta instalado (necesario para Ollama).'
+    Write-Host ''
+    Write-Host 'Instala Docker Desktop para Windows:'
+    Write-Host '  https://docs.docker.com/desktop/setup/install/windows-install/'
+    exit 1
+}
+# En PS 5.1, *> $null sobre comandos nativos convierte stderr en errores
+# terminantes con EAP=Stop (PS #4002). Se descarta via cmd /c.
+$null = cmd /c "docker info >NUL 2>&1"
+if ($LASTEXITCODE -ne 0) {
+    Err 'Docker Desktop no responde. Abre Docker Desktop y espera a que arranque.'
+    exit 1
+}
+Ok "Docker: $((docker --version) -replace 'Docker version ','' -replace ',.*','')"
+
+# --- Check RAM ------------------------------------------------------------
+$TotalRamGb = [math]::Round((Get-CimInstance Win32_ComputerSystem).TotalPhysicalMemory / 1GB, 1)
+if ($TotalRamGb -lt 3.5) {
+    Err "Se necesitan 4 GB de RAM minimo (detectado: $TotalRamGb GB)."
+    exit 1
+}
+Ok "RAM: $TotalRamGb GB"
+
+# --- Tumbar deploy previo (antes de detectar puertos) -----------------------
+cmd /c docker rm -f d0a1-hermes-ollama '>' NUL '2>&1'
+Info 'Deploy previo detenido (rm; volumen con el modelo intacto).'
+
+# --- GPU NVIDIA? (si hay, Ollama la usa y va 5-10x mas rapido) ---------------
+$HasGpu = $false
+if (Get-Command nvidia-smi -ErrorAction SilentlyContinue) {
+    $GpuInfo = & nvidia-smi --query-gpu=name,memory.total --format=csv,noheader 2>$null
+    if ($LASTEXITCODE -eq 0 -and $GpuInfo) {
+        $HasGpu = $true
+        $GpuName = ($GpuInfo -split ',')[0].Trim()
+        Info "GPU NVIDIA detectada: $GpuName (aceleracion activada)"
+    }
+}
+if (-not $HasGpu) {
+    Warn 'Sin GPU NVIDIA: modo CPU (~10-20 tokens/s). Con una GPU NVIDIA va 5-10x mas rapido.'
+}
+# --- Auto-detect Ollama port ------------------------------------------------
+$OllamaPort = Find-FreePort 11434
+
+if ($Mode -eq 'vps') {
+    Warn 'Modo: VPS'
+    Write-Host ''
+    Write-Host "  ${Bold}¿No tienes VPS?${NC} Consigue uno con descuento:"
+    Write-Host "  ${Cyan}https://d0a1.es/go/vultr/${NC}  - Vultr"
+    Write-Host "  ${Cyan}https://d0a1.es/go/digitalocean/${NC}  - DigitalOcean (`$200 credit)"
+    Write-Host ''
+}
+if ($OllamaPort -ne 11434) { Warn "Puerto 11434 ocupado, usando $OllamaPort" }
+Write-Host ''
+
+# --- 1. Arrancar Ollama (Docker) -------------------------------------------
+Warn '[1/3] Arrancando Ollama (IA local)...'
+$running = docker ps --format '{{.Names}}' | Select-String -Quiet 'd0a1-hermes-ollama'
+if (-not $running) {
+    $GpuArgs = @()
+    if ($HasGpu) { $GpuArgs = @('--gpus', 'all') }
+    docker run -d --name d0a1-hermes-ollama --memory 6g @GpuArgs `
+        -p "${OllamaPort}:11434" `
+        -v d0a1_hermes_ollama:/root/.ollama `
+        -e OLLAMA_HOST=0.0.0.0 `
+        --restart unless-stopped `
+        ollama/ollama:0.33.3 | Out-Null
+    if ($LASTEXITCODE -ne 0) { Err 'Fallo al arrancar el contenedor de Ollama.'; exit 1 }
+}
+
+Warn '        Esperando a que Ollama arranque...'
+$MaxWait = 60; $Waited = 0
+while ($true) {
+    # Descarte de stderr via cmd /c (ver nota de PS 5.1 / PS #4002 arriba)
+    cmd /c "docker exec d0a1-hermes-ollama ollama ps >NUL 2>&1"
+    if ($LASTEXITCODE -eq 0) { break }
+    Start-Sleep -Seconds 2
+    $Waited += 2
+    if ($Waited -ge $MaxWait) {
+        Err 'Ollama no arranco. Revisa: docker logs d0a1-hermes-ollama'
+        exit 1
+    }
+}
+Warn '        Descargando modelo (qwen3:4b, ~2.5 GB)...'
+# Si el usuario define $env:HTTPS_PROXY, Ollama lo usa para el pull (doc oficial):
+$ProxyEnv = @()
+if ($env:HTTPS_PROXY) { $ProxyEnv = @('-e', "HTTPS_PROXY=$env:HTTPS_PROXY", '-e', "HTTP_PROXY=$env:HTTP_PROXY") }
+# --- Modelo ya presente? (volumen persistente: no re-descargar) ------------
+$ModelList = docker exec d0a1-hermes-ollama ollama list 2>$null
+if ("$ModelList" -match 'qwen3:4b') {
+    Info 'Modelo qwen3:4b ya descargado (volumen intacto). Omitiendo descarga.'
+} else {
+    # --- Modelo: HF primero (CDN propio, rapido) y registry de Ollama como ultimo recurso ---
+
+$Gguf = Join-Path $ScriptDir 'qwen3-4b-q4.gguf'
+$HfUrls = @(
+    'https://huggingface.co/Qwen/Qwen3-4B-GGUF/resolve/main/Qwen3-4B-Q4_K_M.gguf',
+    'https://huggingface.co/unsloth/Qwen3-4B-GGUF/resolve/main/Qwen3-4B-Q4_K_M.gguf'
+)
+Info 'Descargando modelo de IA (qwen3:4b, ~2.5 GB)...'
+foreach ($Url in $HfUrls) {
+    curl.exe -# -L --retry 3 --max-time 1800 -o "$Gguf" $Url
+    if ($LASTEXITCODE -eq 0 -and (Test-Path $Gguf) -and (Get-Item $Gguf).Length -gt 2400000000) { break }
+    Remove-Item "$Gguf" -Force -ErrorAction SilentlyContinue
+    if ($Url -ne $HfUrls[-1]) { Warn 'Mirror no accesible, probando el siguiente...' }
+}
+if ((Test-Path $Gguf) -and (Get-Item $Gguf).Length -gt 2400000000) {
+    Info 'Importando modelo en Ollama...'
+    docker cp "$Gguf" d0a1-hermes-ollama:/root/.ollama/import.gguf
+    docker exec d0a1-hermes-ollama sh -c 'printf "FROM /root/.ollama/import.gguf\\nPARAMETER num_ctx 16384\\n" > /tmp/mf && ollama create qwen3:4b -f /tmp/mf'
+    $CreateOk = ($LASTEXITCODE -eq 0)
+    docker exec d0a1-hermes-ollama rm -f /root/.ollama/import.gguf | Out-Null
+    Remove-Item "$Gguf" -Force -ErrorAction SilentlyContinue
+    $PullOk = $CreateOk
+    if (-not $CreateOk) { Err 'Fallo al importar el modelo desde el GGUF.' }
+} else {
+    # Ultimo recurso: pull directo (1 intento, ya sin esperar 3 rondas)
+    Warn 'HuggingFace no accesible. Probando el registry de Ollama (ultimo recurso)...'
+    docker exec @ProxyEnv d0a1-hermes-ollama ollama pull qwen3:4b
+    if ($LASTEXITCODE -eq 0) {
+        $PullOk = $true
+        # Contexto ampliado (el default 4096 da 400 en Open WebUI con prompts largos)
+        docker exec d0a1-hermes-ollama sh -c 'printf "FROM qwen3:4b\\nPARAMETER num_ctx 16384\\n" > /tmp/mf && ollama create qwen3:4b -f /tmp/mf'
+    }
+}
+if (-not $PullOk) {
+    Err 'No se pudo descargar el modelo (ni HuggingFace ni registry).'
+    if (-not $env:HTTPS_PROXY) {
+        Write-Host '  Tu red bloquea ambos. Si tienes proxy/VPN:'
+        Write-Host '    $env:HTTPS_PROXY = "http://127.0.0.1:7890"   # (puerto de tu proxy)'
+        Write-Host '    .\deploy.ps1 vps'
+    }
+    exit 1
+}
+}
+# --- Precalentar el modelo (load ya, no en el primer mensaje) ---------------
+Warn '        Precargando el modelo en memoria...'
+cmd /c docker exec d0a1-hermes-ollama sh -c 'ollama run qwen3:4b ""' '>' NUL '2>&1'
+
+Ok "Ollama corriendo en http://localhost:${OllamaPort}"
+
+# --- 2. Instalar Hermes Agent ------------------------------------------------
+Warn '[2/3] Instalando Hermes Agent...'
+if (Get-Command hermes -ErrorAction SilentlyContinue) {
+    Ok "Hermes Agent ya instalado: $(cmd /c 'hermes --version 2>NUL')"
+} else {
+    # En Windows: instala con winget o pip. El install.sh de Linux no aplica.
+    Write-Host '  Instalando con pip (requiere Python 3.10+)...'
+    pip install --user hermes-agent
+    if ($LASTEXITCODE -ne 0 -or -not (Get-Command hermes -ErrorAction SilentlyContinue)) {
+        Err 'Hermes Agent no se instalo. Instalalo manualmente:'
+        Write-Host '  https://hermes-agent.nousresearch.com/docs'
+        exit 1
+    }
+    Ok "Hermes Agent instalado: $(cmd /c 'hermes --version 2>NUL')"
+}
+
+# --- 3. Configurar ------------------------------------------------------------
+Warn '[3/3] Configurando...'
+Write-Host ''
+Write-Host "  ${Bold}Ejecuta estos comandos para terminar:${NC}"
+Write-Host ''
+Write-Host "  ${Cyan}hermes setup${NC}         - configuracion interactiva"
+Write-Host "  ${Cyan}hermes model${NC}          - elige modelo (Ollama ya esta corriendo)"
+Write-Host "  ${Cyan}hermes gateway setup${NC}  - conecta mensajeria (Telegram, etc.)"
+Write-Host ''
+
+if ($Mode -eq 'vps') {
+    Write-Host "  ${Bold}Para VPS:${NC} configura el gateway para acceso remoto:"
+    Write-Host "  ${Cyan}hermes gateway setup --portal${NC}"
+    Write-Host ''
+}
+
+Write-Host "  ${Bold}Parar:${NC} .\deploy.ps1 stop"
+Write-Host "  ${Bold}¿Quieres mas?${NC} ${Cyan}https://d0a1.es/packs/${NC}"
+Write-Host ''

hermes/deploy.sh

@@ -112,8 +112,8 @@ until docker exec d0a1-hermes-ollama ollama ps &>/dev/null; do
         exit 1
     fi
 done
-echo -e "${YELLOW}        Descargando modelo (llama3.2:3b, ~2 GB)...${NC}"
+echo -e "${YELLOW}        Descargando modelo (qwen3:4b, ~2.5 GB)...${NC}"
-docker exec d0a1-hermes-ollama ollama pull llama3.2:3b
+docker exec d0a1-hermes-ollama ollama pull qwen3:4b
 echo -e "${GREEN}✓${NC} Ollama corriendo en http://localhost:${OLLAMA_PORT}"
 
 # ── 2. Instalar Hermes Agent ────────────────

openclaw/README.md

@@ -33,8 +33,8 @@ openclaw channel add telegram
 Solo pagas el servidor. El asistente es tuyo.
 
 **¿No tienes VPS?** Consigue uno con descuento:
-- [Vultr](https://d0a1.es/go/vultr/) — desde ~6€/mes
+- [Vultr](https://d0a1.es/go/vultr/) — desde ~6€/mes (afiliado)
-- [DigitalOcean](https://d0a1.es/go/digitalocean/) — credit gratuito
+- [DigitalOcean](https://d0a1.es/go/digitalocean/) — credit gratuito (afiliado)
 
 ## Requisitos
 

openclaw/deploy.ps1

@@ -0,0 +1,237 @@
+# d0a1 pack - OpenClaw (Windows)
+# Asistente de IA en tus chats. OpenClaw + Ollama.
+#
+# Local:  .\deploy.ps1 local    -> terminal + http://localhost:11434 (Ollama)
+# VPS:    .\deploy.ps1 vps      -> + gateway accesible desde fuera
+# Stop:   .\deploy.ps1 stop     -> para Ollama
+#
+# Requiere: Docker Desktop para Windows (https://docs.docker.com/desktop/)
+
+param(
+    [Parameter(Position = 0)]
+    [ValidateSet('local', 'vps', 'stop')]
+    [string]$Mode = 'local'
+)
+
+$ErrorActionPreference = 'Stop'
+
+# Escape ANSI via [char]27 (compatible con Windows PowerShell 5.1 y PS 7+)
+$Esc = [char]27
+$Green = "$Esc[32m"; $Cyan = "$Esc[36m"; $Yellow = "$Esc[33m"; $Red = "$Esc[31m"; $Bold = "$Esc[1m"; $NC = "$Esc[0m"
+if ($env:NO_COLOR) {
+    $Green = $Cyan = $Yellow = $Red = $Bold = $NC = ''
+}
+
+function Info($msg) { Write-Host "${Bold}${Green}>>>${NC} $msg" }
+function Warn($msg) { Write-Host "${Bold}${Yellow}>>>${NC} $msg" }
+function Err($msg)  { Write-Host "${Bold}${Red}>>>${NC} $msg" }
+function Ok($msg)   { Write-Host "${Green}OK${NC} $msg" }
+
+function Find-FreePort([int]$Start) {
+    for ($p = $Start; $p -le ($Start + 20); $p++) {
+        $inUse = Get-NetTCPConnection -LocalPort $p -State Listen -ErrorAction SilentlyContinue
+        if (-not $inUse) { return $p }
+    }
+    return $Start
+}
+
+Write-Host ''
+Write-Host "${Bold}${Cyan}============================================================${NC}"
+Write-Host "${Bold}${Cyan}  d0a1 pack - OpenClaw (Windows)${NC}"
+Write-Host "${Bold}${Cyan}  Asistente de IA en tus chats.${NC}"
+Write-Host "${Bold}${Cyan}============================================================${NC}"
+Write-Host ''
+
+# --- Stop mode ------------------------------------------------------------
+if ($Mode -eq 'stop') {
+    $running = docker ps --format '{{.Names}}' | Select-String -Quiet 'd0a1-openclaw-ollama'
+    if ($running) {
+        docker stop d0a1-openclaw-ollama | Out-Null
+        docker rm d0a1-openclaw-ollama | Out-Null
+        Info 'Ollama parado.'
+    } else {
+        Write-Host 'Ollama no esta corriendo.'
+    }
+    exit 0
+}
+
+# --- Check Docker (para Ollama) -------------------------------------------
+if (-not (Get-Command docker -ErrorAction SilentlyContinue)) {
+    Err 'Docker no esta instalado (necesario para Ollama).'
+    Write-Host ''
+    Write-Host 'Instala Docker Desktop para Windows:'
+    Write-Host '  https://docs.docker.com/desktop/setup/install/windows-install/'
+    exit 1
+}
+# En PS 5.1, *> $null sobre comandos nativos convierte stderr en errores
+# terminantes con EAP=Stop (PS #4002). Se descarta via cmd /c.
+$null = cmd /c "docker info >NUL 2>&1"
+if ($LASTEXITCODE -ne 0) {
+    Err 'Docker Desktop no responde. Abre Docker Desktop y espera a que arranque.'
+    exit 1
+}
+Ok "Docker: $((docker --version) -replace 'Docker version ','' -replace ',.*','')"
+
+# --- Check RAM ------------------------------------------------------------
+$TotalRamGb = [math]::Round((Get-CimInstance Win32_ComputerSystem).TotalPhysicalMemory / 1GB, 1)
+if ($TotalRamGb -lt 3.5) {
+    Err "Se necesitan 4 GB de RAM minimo (detectado: $TotalRamGb GB)."
+    exit 1
+}
+Ok "RAM: $TotalRamGb GB"
+
+# --- Tumbar deploy previo (antes de detectar puertos) -----------------------
+cmd /c docker rm -f d0a1-openclaw-ollama '>' NUL '2>&1'
+Info 'Deploy previo detenido (rm; volumen con el modelo intacto).'
+
+# --- GPU NVIDIA? (si hay, Ollama la usa y va 5-10x mas rapido) ---------------
+$HasGpu = $false
+if (Get-Command nvidia-smi -ErrorAction SilentlyContinue) {
+    $GpuInfo = & nvidia-smi --query-gpu=name,memory.total --format=csv,noheader 2>$null
+    if ($LASTEXITCODE -eq 0 -and $GpuInfo) {
+        $HasGpu = $true
+        $GpuName = ($GpuInfo -split ',')[0].Trim()
+        Info "GPU NVIDIA detectada: $GpuName (aceleracion activada)"
+    }
+}
+if (-not $HasGpu) {
+    Warn 'Sin GPU NVIDIA: modo CPU (~10-20 tokens/s). Con una GPU NVIDIA va 5-10x mas rapido.'
+}
+# --- Auto-detect Ollama port ------------------------------------------------
+$OllamaPort = Find-FreePort 11434
+
+if ($Mode -eq 'vps') {
+    Warn 'Modo: VPS'
+    Write-Host ''
+    Write-Host "  ${Bold}¿No tienes VPS?${NC} Consigue uno con descuento:"
+    Write-Host "  ${Cyan}https://d0a1.es/go/vultr/${NC}  - Vultr"
+    Write-Host "  ${Cyan}https://d0a1.es/go/digitalocean/${NC}  - DigitalOcean (`$200 credit)"
+    Write-Host ''
+}
+if ($OllamaPort -ne 11434) { Warn "Puerto 11434 ocupado, usando $OllamaPort" }
+Write-Host ''
+
+# --- 1. Arrancar Ollama (Docker) -------------------------------------------
+Warn '[1/3] Arrancando Ollama (IA local)...'
+$running = docker ps --format '{{.Names}}' | Select-String -Quiet 'd0a1-openclaw-ollama'
+if (-not $running) {
+    $GpuArgs = @()
+    if ($HasGpu) { $GpuArgs = @('--gpus', 'all') }
+    docker run -d --name d0a1-openclaw-ollama --memory 6g @GpuArgs `
+        -p "${OllamaPort}:11434" `
+        -v d0a1_openclaw_ollama:/root/.ollama `
+        -e OLLAMA_HOST=0.0.0.0 `
+        --restart unless-stopped `
+        ollama/ollama:0.33.3 | Out-Null
+    if ($LASTEXITCODE -ne 0) { Err 'Fallo al arrancar el contenedor de Ollama.'; exit 1 }
+}
+
+Warn '        Esperando a que Ollama arranque...'
+$MaxWait = 60; $Waited = 0
+while ($true) {
+    # Descarte de stderr via cmd /c (ver nota de PS 5.1 / PS #4002 arriba)
+    cmd /c "docker exec d0a1-openclaw-ollama ollama ps >NUL 2>&1"
+    if ($LASTEXITCODE -eq 0) { break }
+    Start-Sleep -Seconds 2
+    $Waited += 2
+    if ($Waited -ge $MaxWait) {
+        Err 'Ollama no arranco. Revisa: docker logs d0a1-openclaw-ollama'
+        exit 1
+    }
+}
+Warn '        Descargando modelo (qwen3:4b, ~2.5 GB)...'
+# Si el usuario define $env:HTTPS_PROXY, Ollama lo usa para el pull (doc oficial):
+$ProxyEnv = @()
+if ($env:HTTPS_PROXY) { $ProxyEnv = @('-e', "HTTPS_PROXY=$env:HTTPS_PROXY", '-e', "HTTP_PROXY=$env:HTTP_PROXY") }
+# --- Modelo ya presente? (volumen persistente: no re-descargar) ------------
+$ModelList = docker exec d0a1-openclaw-ollama ollama list 2>$null
+if ("$ModelList" -match 'qwen3:4b') {
+    Info 'Modelo qwen3:4b ya descargado (volumen intacto). Omitiendo descarga.'
+} else {
+    # --- Modelo: HF primero (CDN propio, rapido) y registry de Ollama como ultimo recurso ---
+
+$Gguf = Join-Path $ScriptDir 'qwen3-4b-q4.gguf'
+$HfUrls = @(
+    'https://huggingface.co/Qwen/Qwen3-4B-GGUF/resolve/main/Qwen3-4B-Q4_K_M.gguf',
+    'https://huggingface.co/unsloth/Qwen3-4B-GGUF/resolve/main/Qwen3-4B-Q4_K_M.gguf'
+)
+Info 'Descargando modelo de IA (qwen3:4b, ~2.5 GB)...'
+foreach ($Url in $HfUrls) {
+    curl.exe -# -L --retry 3 --max-time 1800 -o "$Gguf" $Url
+    if ($LASTEXITCODE -eq 0 -and (Test-Path $Gguf) -and (Get-Item $Gguf).Length -gt 2400000000) { break }
+    Remove-Item "$Gguf" -Force -ErrorAction SilentlyContinue
+    if ($Url -ne $HfUrls[-1]) { Warn 'Mirror no accesible, probando el siguiente...' }
+}
+if ((Test-Path $Gguf) -and (Get-Item $Gguf).Length -gt 2400000000) {
+    Info 'Importando modelo en Ollama...'
+    docker cp "$Gguf" d0a1-openclaw-ollama:/root/.ollama/import.gguf
+    docker exec d0a1-openclaw-ollama sh -c 'printf "FROM /root/.ollama/import.gguf\\nPARAMETER num_ctx 16384\\n" > /tmp/mf && ollama create qwen3:4b -f /tmp/mf'
+    $CreateOk = ($LASTEXITCODE -eq 0)
+    docker exec d0a1-openclaw-ollama rm -f /root/.ollama/import.gguf | Out-Null
+    Remove-Item "$Gguf" -Force -ErrorAction SilentlyContinue
+    $PullOk = $CreateOk
+    if (-not $CreateOk) { Err 'Fallo al importar el modelo desde el GGUF.' }
+} else {
+    # Ultimo recurso: pull directo (1 intento, ya sin esperar 3 rondas)
+    Warn 'HuggingFace no accesible. Probando el registry de Ollama (ultimo recurso)...'
+    docker exec @ProxyEnv d0a1-openclaw-ollama ollama pull qwen3:4b
+    if ($LASTEXITCODE -eq 0) {
+        $PullOk = $true
+        # Contexto ampliado (el default 4096 da 400 en Open WebUI con prompts largos)
+        docker exec d0a1-openclaw-ollama sh -c 'printf "FROM qwen3:4b\\nPARAMETER num_ctx 16384\\n" > /tmp/mf && ollama create qwen3:4b -f /tmp/mf'
+    }
+}
+if (-not $PullOk) {
+    Err 'No se pudo descargar el modelo (ni HuggingFace ni registry).'
+    if (-not $env:HTTPS_PROXY) {
+        Write-Host '  Tu red bloquea ambos. Si tienes proxy/VPN:'
+        Write-Host '    $env:HTTPS_PROXY = "http://127.0.0.1:7890"   # (puerto de tu proxy)'
+        Write-Host '    .\deploy.ps1 vps'
+    }
+    exit 1
+}
+}
+# --- Precalentar el modelo (load ya, no en el primer mensaje) ---------------
+Warn '        Precargando el modelo en memoria...'
+cmd /c docker exec d0a1-openclaw-ollama sh -c 'ollama run qwen3:4b ""' '>' NUL '2>&1'
+
+Ok "Ollama corriendo en http://localhost:${OllamaPort}"
+
+# --- 2. Instalar OpenClaw -----------------------------------------------------
+Warn '[2/3] Instalando OpenClaw...'
+if (Get-Command openclaw -ErrorAction SilentlyContinue) {
+    Ok "OpenClaw ya instalado: $(cmd /c 'openclaw --version 2>NUL')"
+} else {
+    # En Windows: instala con npm (Node 18+). El install.sh de Linux no aplica.
+    Write-Host '  Instalando con npm (requiere Node.js 18+)...'
+    npm install -g openclaw
+    if ($LASTEXITCODE -ne 0 -or -not (Get-Command openclaw -ErrorAction SilentlyContinue)) {
+        Err 'OpenClaw no se instalo. Instalalo manualmente:'
+        Write-Host '  https://openclaw.ai/docs'
+        exit 1
+    }
+    Ok "OpenClaw instalado: $(cmd /c 'openclaw --version 2>NUL')"
+}
+
+# --- 3. Configurar ------------------------------------------------------------
+Warn '[3/3] Configurando...'
+Write-Host ''
+Write-Host "  ${Bold}Ejecuta estos comandos para terminar:${NC}"
+Write-Host ''
+Write-Host "  ${Cyan}openclaw onboard --install-daemon${NC}  - configuracion inicial"
+Write-Host "  ${Cyan}openclaw gateway status${NC}            - verifica el gateway"
+Write-Host "  ${Cyan}openclaw dashboard${NC}                 - abre la UI"
+Write-Host ''
+Write-Host "  ${Bold}Conecta tus chats (WhatsApp, Telegram, etc.):${NC}"
+Write-Host "  ${Cyan}openclaw channel add whatsapp${NC}"
+Write-Host "  ${Cyan}openclaw channel add telegram${NC}"
+Write-Host ''
+
+if ($Mode -eq 'vps') {
+    Write-Host "  ${Bold}Para VPS:${NC} configura el gateway para acceso remoto."
+    Write-Host ''
+}
+
+Write-Host "  ${Bold}Parar:${NC} .\deploy.ps1 stop"
+Write-Host "  ${Bold}¿Quieres mas?${NC} ${Cyan}https://d0a1.es/packs/${NC}"
+Write-Host ''

openclaw/deploy.sh

@@ -112,8 +112,8 @@ until docker exec d0a1-openclaw-ollama ollama ps &>/dev/null; do
         exit 1
     fi
 done
-echo -e "${YELLOW}        Descargando modelo (llama3.2:3b, ~2 GB)...${NC}"
+echo -e "${YELLOW}        Descargando modelo (qwen3:4b, ~2.5 GB)...${NC}"
-docker exec d0a1-openclaw-ollama ollama pull llama3.2:3b
+docker exec d0a1-openclaw-ollama ollama pull qwen3:4b
 echo -e "${GREEN}✓${NC} Ollama corriendo en http://localhost:${OLLAMA_PORT}"
 
 # ── 2. Instalar OpenClaw ─────────────────────

openwebui/README.md

@@ -23,8 +23,8 @@ Abre http://localhost:3000 y empieza a chatear.
 Solo pagas el servidor. La IA es tuya.
 
 **¿No tienes VPS?** Consigue uno con descuento:
-- [Vultr](https://d0a1.es/go/vultr/) — desde ~6€/mes
+- [Vultr](https://d0a1.es/go/vultr/) — desde ~6€/mes (afiliado)
-- [DigitalOcean](https://d0a1.es/go/digitalocean/) — credit gratuito
+- [DigitalOcean](https://d0a1.es/go/digitalocean/) — credit gratuito (afiliado)
 
 ## Requisitos
 
@@ -33,7 +33,7 @@ Solo pagas el servidor. La IA es tuya.
 
 ## Modelos
 
-El modelo por defecto es `llama3.2:3b` (2 GB, corre en 4 GB RAM).
+El modelo por defecto es `qwen3:4b` (2 GB, corre en 4 GB RAM).
 
 ```bash
 docker exec d0a1-ollama ollama pull qwen3:4b      # Mejor en español + razonamiento (4 GB RAM)

openwebui/deploy.ps1

@@ -0,0 +1,236 @@
+# d0a1 pack - Open WebUI (Windows)
+# Chat de IA privado. Sin cuenta, sin API key.
+#
+# Local:  .\deploy.ps1 local    -> http://localhost:3000
+# VPS:    .\deploy.ps1 vps      -> http://TU_IP:3000
+# Stop:   .\deploy.ps1 stop
+#
+# Requiere: Docker Desktop para Windows (https://docs.docker.com/desktop/)
+
+param(
+    [Parameter(Position = 0)]
+    [ValidateSet('local', 'vps', 'stop')]
+    [string]$Mode = 'local'
+)
+
+$ErrorActionPreference = 'Stop'
+
+# Escape ANSI via [char]27 (compatible con Windows PowerShell 5.1 y PS 7+)
+$Esc = [char]27
+$Green = "$Esc[32m"; $Cyan = "$Esc[36m"; $Yellow = "$Esc[33m"; $Red = "$Esc[31m"; $Bold = "$Esc[1m"; $NC = "$Esc[0m"
+if ($env:NO_COLOR) {
+    $Green = $Cyan = $Yellow = $Red = $Bold = $NC = ''
+}
+
+$ScriptDir = $PSScriptRoot
+$ComposeFile = Join-Path $ScriptDir 'docker-compose.yml'
+
+function Info($msg)    { Write-Host "${Bold}${Green}>>>${NC} $msg" }
+function Warn($msg)    { Write-Host "${Bold}${Yellow}>>>${NC} $msg" }
+function Err($msg)     { Write-Host "${Bold}${Red}>>>${NC} $msg" }
+function Ok($msg)      { Write-Host "${Green}OK${NC} $msg" }
+
+function Find-FreePort([int]$Start) {
+    for ($p = $Start; $p -le ($Start + 20); $p++) {
+        $inUse = Get-NetTCPConnection -LocalPort $p -State Listen -ErrorAction SilentlyContinue
+        if (-not $inUse) { return $p }
+    }
+    return $Start
+}
+
+Write-Host ''
+Write-Host "${Bold}${Cyan}============================================================${NC}"
+Write-Host "${Bold}${Cyan}  d0a1 pack - Open WebUI (Windows)${NC}"
+Write-Host "${Bold}${Cyan}  Chat de IA privado. Sin cuenta, sin API key.${NC}"
+Write-Host "${Bold}${Cyan}============================================================${NC}"
+Write-Host ''
+
+# --- Stop mode ------------------------------------------------------------
+if ($Mode -eq 'stop') {
+    Warn 'Parando...'
+    docker compose -f $ComposeFile down
+    Info 'Parado.'
+    exit 0
+}
+
+# --- Check Docker ---------------------------------------------------------
+if (-not (Get-Command docker -ErrorAction SilentlyContinue)) {
+    Err 'Docker no esta instalado.'
+    Write-Host ''
+    Write-Host 'Instala Docker Desktop para Windows:'
+    Write-Host '  https://docs.docker.com/desktop/setup/install/windows-install/'
+    exit 1
+}
+# En PS 5.1, *> $null sobre comandos nativos convierte stderr en errores
+# terminantes con EAP=Stop (PS #4002). Se descarta via cmd /c.
+$null = cmd /c "docker info >NUL 2>&1"
+if ($LASTEXITCODE -ne 0) {
+    Err 'Docker Desktop no responde. Abre Docker Desktop y espera a que arranque.'
+    exit 1
+}
+# En PS 5.1, *> $null sobre comandos nativos convierte stderr en errores
+# terminantes con EAP=Stop (PS #4002). Se descarta via cmd /c.
+cmd /c "docker compose version >NUL 2>&1"
+$composeOk = ($LASTEXITCODE -eq 0)
+if (-not $composeOk) {
+    Err 'Docker Compose v2 no disponible. Actualiza Docker Desktop.'
+    exit 1
+}
+Ok "Docker: $(cmd /c 'docker compose version --short 2>NUL')"
+
+# --- Check RAM ------------------------------------------------------------
+$TotalRamGb = [math]::Round((Get-CimInstance Win32_ComputerSystem).TotalPhysicalMemory / 1GB, 1)
+if ($TotalRamGb -lt 3.5) {
+    Err "Se necesitan 4 GB de RAM minimo (detectado: $TotalRamGb GB)."
+    exit 1
+}
+Ok "RAM: $TotalRamGb GB"
+
+# --- Tumbar deploy previo (antes de detectar puertos) ----------------------
+# Mantiene los volumenes (el modelo ya descargado no se re-descarga) y libera
+# los puertos para que Find-FreePort vea el estado REAL tras el down.
+cmd /c docker compose -f $ComposeFile down '>' NUL '2>&1'
+Info 'Deploy previo detenido (down; volumenes con el modelo intactos).'
+
+# --- GPU NVIDIA? (si hay, Ollama la usa y va 5-10x mas rapido) ---------------
+$HasGpu = $false
+if (Get-Command nvidia-smi -ErrorAction SilentlyContinue) {
+    $GpuInfo = & nvidia-smi --query-gpu=name,memory.total --format=csv,noheader 2>$null
+    if ($LASTEXITCODE -eq 0 -and $GpuInfo) {
+        $HasGpu = $true
+        $GpuName = ($GpuInfo -split ',')[0].Trim()
+        Info "GPU NVIDIA detectada: $GpuName (aceleracion activada)"
+    }
+}
+if (-not $HasGpu) {
+    Warn 'Sin GPU NVIDIA: modo CPU (~10-20 tokens/s). Con una GPU NVIDIA va 5-10x mas rapido.'
+}
+# --- Auto-detect ports ----------------------------------------------------
+$OllamaPort = Find-FreePort 11434
+$WebUiPort  = Find-FreePort 3000
+
+# --- Mode info ------------------------------------------------------------
+if ($Mode -eq 'local') {
+    Warn 'Modo: local (tu ordenador)'
+    Write-Host "  URL: ${Cyan}http://localhost:${WebUiPort}${NC}"
+} else {
+    Warn 'Modo: VPS (accesible desde fuera)'
+    # IP real de LAN: la de la interfaz con la ruta por defecto (evita vEthernet
+    # WSL/Hyper-V, Docker, VPN — que son 172.16-31.*, 10.x internos, etc.)
+    $GwIf = (Get-NetRoute -DestinationPrefix '0.0.0.0/0' -ErrorAction SilentlyContinue |
+        Sort-Object RouteMetric | Select-Object -First 1).ifIndex
+    $Ip = (Get-NetIPAddress -AddressFamily IPv4 -ErrorAction SilentlyContinue |
+        Where-Object { $_.ifIndex -eq $GwIf -and $_.IPAddress -notlike '127.*' } |
+        Select-Object -First 1).IPAddress
+    if (-not $Ip) { $Ip = 'TU_IP' }
+    Write-Host "  URL: ${Cyan}http://${Ip}:${WebUiPort}${NC}"
+    Write-Host ''
+    Write-Host "  ${Bold}¿No tienes VPS?${NC} Consigue uno con descuento:"
+    Write-Host "  ${Cyan}https://d0a1.es/go/vultr/${NC}  - Vultr"
+    Write-Host "  ${Cyan}https://d0a1.es/go/digitalocean/${NC}  - DigitalOcean (`$200 credit)"
+}
+
+if ($OllamaPort -ne 11434) { Warn "Puerto 11434 ocupado, usando $OllamaPort" }
+if ($WebUiPort  -ne 3000)  { Warn "Puerto 3000 ocupado, usando $WebUiPort" }
+Write-Host ''
+
+# --- Deploy ----------------------------------------------------------------
+Warn 'Descargando contenedores (2-3 min la primera vez)...'
+$env:OLLAMA_PORT = "$OllamaPort"
+$env:WEBUI_PORT  = "$WebUiPort"
+$ComposeArgs = @('-f', $ComposeFile)
+if ($HasGpu) {
+    $OverrideFile = Join-Path $ScriptDir 'docker-compose.gpu.yml'
+    Set-Content -Path $OverrideFile -Value "services:`r`n  ollama:`r`n    gpus: all`r`n"
+    $ComposeArgs += @('-f', $OverrideFile)
+}
+docker compose @ComposeArgs up -d
+if ($LASTEXITCODE -ne 0) { Err "docker compose up fallo (exit $LASTEXITCODE)."; exit 1 }
+
+# --- Pull model -------------------------------------------------------------
+Info 'Esperando a que Ollama arranque...'
+$MaxWait = 60; $Waited = 0
+while ($true) {
+    # Descarte de stderr via cmd /c (ver nota de PS 5.1 / PS #4002 arriba)
+    cmd /c "docker exec d0a1-ollama ollama ps >NUL 2>&1"
+    if ($LASTEXITCODE -eq 0) { break }
+    Start-Sleep -Seconds 2
+    $Waited += 2
+    if ($Waited -ge $MaxWait) {
+        Err 'Ollama no arranco. Revisa: docker logs d0a1-ollama'
+        exit 1
+    }
+}
+# Si el usuario define $env:HTTPS_PROXY, Ollama lo usa para el pull (doc oficial):
+$ProxyEnv = @()
+if ($env:HTTPS_PROXY) { $ProxyEnv = @('-e', "HTTPS_PROXY=$env:HTTPS_PROXY", '-e', "HTTP_PROXY=$env:HTTP_PROXY") }
+# --- Modelo ya presente? (volumen persistente: no re-descargar) ------------
+$ModelList = docker exec d0a1-ollama ollama list 2>$null
+if ("$ModelList" -match 'qwen3:4b') {
+    Info 'Modelo qwen3:4b ya descargado (volumen intacto). Omitiendo descarga.'
+} else {
+    # --- Modelo: HF primero (CDN propio, rapido) y registry de Ollama como ultimo recurso ---
+
+$Gguf = Join-Path $ScriptDir 'qwen3-4b-q4.gguf'
+$HfUrls = @(
+    'https://huggingface.co/Qwen/Qwen3-4B-GGUF/resolve/main/Qwen3-4B-Q4_K_M.gguf',
+    'https://huggingface.co/unsloth/Qwen3-4B-GGUF/resolve/main/Qwen3-4B-Q4_K_M.gguf'
+)
+Info 'Descargando modelo de IA (qwen3:4b, ~2.5 GB)...'
+foreach ($Url in $HfUrls) {
+    curl.exe -# -L --retry 3 --max-time 1800 -o "$Gguf" $Url
+    if ($LASTEXITCODE -eq 0 -and (Test-Path $Gguf) -and (Get-Item $Gguf).Length -gt 2400000000) { break }
+    Remove-Item "$Gguf" -Force -ErrorAction SilentlyContinue
+    if ($Url -ne $HfUrls[-1]) { Warn 'Mirror no accesible, probando el siguiente...' }
+}
+if ((Test-Path $Gguf) -and (Get-Item $Gguf).Length -gt 2400000000) {
+    Info 'Importando modelo en Ollama...'
+    docker cp "$Gguf" d0a1-ollama:/root/.ollama/import.gguf
+    docker exec d0a1-ollama sh -c 'printf "FROM /root/.ollama/import.gguf\\nPARAMETER num_ctx 16384\\n" > /tmp/mf && ollama create qwen3:4b -f /tmp/mf'
+    $CreateOk = ($LASTEXITCODE -eq 0)
+    docker exec d0a1-ollama rm -f /root/.ollama/import.gguf | Out-Null
+    Remove-Item "$Gguf" -Force -ErrorAction SilentlyContinue
+    $PullOk = $CreateOk
+    if (-not $CreateOk) { Err 'Fallo al importar el modelo desde el GGUF.' }
+} else {
+    # Ultimo recurso: pull directo (1 intento, ya sin esperar 3 rondas)
+    Warn 'HuggingFace no accesible. Probando el registry de Ollama (ultimo recurso)...'
+    docker exec @ProxyEnv d0a1-ollama ollama pull qwen3:4b
+    if ($LASTEXITCODE -eq 0) {
+        $PullOk = $true
+        # Contexto ampliado (el default 4096 da 400 en Open WebUI con prompts largos)
+        docker exec d0a1-ollama sh -c 'printf "FROM qwen3:4b\\nPARAMETER num_ctx 16384\\n" > /tmp/mf && ollama create qwen3:4b -f /tmp/mf'
+    }
+}
+if (-not $PullOk) {
+    Err 'No se pudo descargar el modelo (ni HuggingFace ni registry).'
+    if (-not $env:HTTPS_PROXY) {
+        Write-Host '  Tu red bloquea ambos. Si tienes proxy/VPN:'
+        Write-Host '    $env:HTTPS_PROXY = "http://127.0.0.1:7890"   # (puerto de tu proxy)'
+        Write-Host '    .\deploy.ps1 vps'
+    }
+    exit 1
+}
+}
+
+# --- Precalentar el modelo (load a GPU/RAM ya, no en el primer mensaje) -----
+Info 'Precargando el modelo en memoria...'
+cmd /c docker exec d0a1-ollama sh -c 'ollama run qwen3:4b ""' '>' NUL '2>&1'
+
+# --- Done -------------------------------------------------------------------
+Write-Host ''
+Info 'Listo! Tu IA esta corriendo.'
+Write-Host ''
+if ($Mode -eq 'local') {
+    Write-Host "  ${Bold}Abre:${NC} ${Cyan}http://localhost:${WebUiPort}${NC}"
+} else {
+    Write-Host "  ${Bold}Abre:${NC} ${Cyan}http://${Ip}:${WebUiPort}${NC}"
+}
+Write-Host ''
+Write-Host "  ${Bold}Modelos alternativos:${NC}"
+Write-Host '  docker exec d0a1-ollama ollama pull llama3.2:3b   # Mas ligero (2 GB RAM)'
+Write-Host '  docker exec d0a1-ollama ollama pull qwen3:8b      # Mas calidad (6 GB RAM)'
+Write-Host ''
+Write-Host "  ${Bold}Parar:${NC} .\deploy.ps1 stop"
+Write-Host "  ${Bold}¿Quieres mas?${NC} ${Cyan}https://d0a1.es/packs/${NC}"
+Write-Host ''

openwebui/deploy.sh

@@ -105,7 +105,7 @@ echo -e "${YELLOW}Descargando contenedores (2-3 min la primera vez)...${NC}"
 docker compose -f "$COMPOSE_FILE" up -d
 
 # ── Pull model ───────────────────────────────
-echo -e "${YELLOW}Descargando modelo de IA (llama3.2:3b, ~2 GB)...${NC}"
+echo -e "${YELLOW}Descargando modelo de IA (qwen3:4b, ~2.5 GB)...${NC}"
 MAX_WAIT=60
 WAITED=0
 until docker exec d0a1-ollama ollama ps &>/dev/null; do
@@ -116,7 +116,7 @@ until docker exec d0a1-ollama ollama ps &>/dev/null; do
         exit 1
     fi
 done
-docker exec d0a1-ollama ollama pull llama3.2:3b
+docker exec d0a1-ollama ollama pull qwen3:4b
 
 # ── Done ────────────────────────────────────
 echo ""
@@ -129,7 +129,7 @@ else
 fi
 echo ""
 echo -e "  ${BOLD}Modelos alternativos:${NC}"
-echo "  docker exec d0a1-ollama ollama pull qwen3:4b      # Mejor en español + razonamiento (4 GB RAM)"
+echo "  docker exec d0a1-ollama ollama pull llama3.2:3b   # Más ligero (2 GB RAM)"
 echo "  docker exec d0a1-ollama ollama pull qwen3:8b      # Más calidad (6 GB RAM)"
 echo ""
 echo -e "  ${BOLD}Parar:${NC} ./deploy.sh stop"

openwebui/docker-compose.yml

@@ -6,7 +6,7 @@
 #
 # Requisitos: Docker + 4 GB RAM mínimo
 # Incluye: Ollama (IA local) + Open WebUI (interfaz web)
-# Modelo por defecto: llama3.2:3b (~2 GB, corre en 4 GB RAM)
+# Modelo por defecto: qwen3:4b (~2.5 GB, corre en 4 GB RAM)
 services:
   ollama:
     container_name: d0a1-ollama
@@ -18,7 +18,7 @@ services:
       - ollama_data:/root/.ollama
     environment:
       OLLAMA_HOST: "0.0.0.0"
-      OLLAMA_KEEP_ALIVE: "5m"
+      OLLAMA_KEEP_ALIVE: "-1"
       OLLAMA_NUM_PARALLEL: "1"
     healthcheck:
       test: ["CMD", "ollama", "ps"]
@@ -26,6 +26,10 @@ services:
       timeout: 5s
       retries: 10
       start_period: 30s
+    deploy:
+      resources:
+        limits:
+          memory: 6G
 
   openwebui:
     container_name: d0a1-openwebui
@@ -43,6 +47,8 @@ services:
       ENABLE_SIGNUP: "true"
       ENABLE_WEB_SEARCH: "false"
       ENABLE_DIRECT_CONNECTIONS: "false"
+      CHAT_RESPONSE_MAX_TOOL_CALL_ITERATIONS: "1"
+      DEFAULT_MODEL_PARAMS: '{"function_calling": "legacy"}'
     depends_on:
       ollama:
         condition: service_healthy

wordpress/README.md

@@ -26,8 +26,8 @@ Sigue el asistente de instalación de WordPress (2 minutos), luego instala el pl
 Solo pagas el servidor. WordPress y la IA son tuyos.
 
 **¿No tienes VPS?** Consigue uno con descuento:
-- [Vultr](https://d0a1.es/go/vultr/) — desde ~6€/mes
+- [Vultr](https://d0a1.es/go/vultr/) — desde ~6€/mes (afiliado)
-- [DigitalOcean](https://d0a1.es/go/digitalocean/) — credit gratuito
+- [DigitalOcean](https://d0a1.es/go/digitalocean/) — credit gratuito (afiliado)
 
 ## Requisitos
 
@@ -38,7 +38,7 @@ Solo pagas el servidor. WordPress y la IA son tuyos.
 
 - WordPress 7.0 + PHP 8.4
 - MariaDB 11.8
-- Ollama (IA local, modelo llama3.2:3b por defecto)
+- Ollama (IA local, modelo qwen3:4b por defecto)
 - Open WebUI (interfaz de chat)
 
 ## Comandos

wordpress/deploy.ps1

@@ -0,0 +1,276 @@
+# d0a1 pack - WordPress + IA (Windows)
+# Tu WordPress con IA local. Genera borradores, modera comentarios.
+#
+# Local:  .\deploy.ps1 local    -> http://localhost:8080
+# VPS:    .\deploy.ps1 vps      -> http://TU_IP:8080
+# Stop:   .\deploy.ps1 stop
+#
+# Requiere: Docker Desktop para Windows (https://docs.docker.com/desktop/)
+
+param(
+    [Parameter(Position = 0)]
+    [ValidateSet('local', 'vps', 'stop')]
+    [string]$Mode = 'local'
+)
+
+$ErrorActionPreference = 'Stop'
+
+# Escape ANSI via [char]27 (compatible con Windows PowerShell 5.1 y PS 7+)
+$Esc = [char]27
+$Green = "$Esc[32m"; $Cyan = "$Esc[36m"; $Yellow = "$Esc[33m"; $Red = "$Esc[31m"; $Bold = "$Esc[1m"; $NC = "$Esc[0m"
+if ($env:NO_COLOR) {
+    $Green = $Cyan = $Yellow = $Red = $Bold = $NC = ''
+}
+
+$ScriptDir = $PSScriptRoot
+$ComposeFile = Join-Path $ScriptDir 'docker-compose.yml'
+
+function Info($msg) { Write-Host "${Bold}${Green}>>>${NC} $msg" }
+function Warn($msg) { Write-Host "${Bold}${Yellow}>>>${NC} $msg" }
+function Err($msg)  { Write-Host "${Bold}${Red}>>>${NC} $msg" }
+function Ok($msg)   { Write-Host "${Green}OK${NC} $msg" }
+
+function Find-FreePort([int]$Start) {
+    for ($p = $Start; $p -le ($Start + 20); $p++) {
+        $inUse = Get-NetTCPConnection -LocalPort $p -State Listen -ErrorAction SilentlyContinue
+        if (-not $inUse) { return $p }
+    }
+    return $Start
+}
+
+Write-Host ''
+Write-Host "${Bold}${Cyan}============================================================${NC}"
+Write-Host "${Bold}${Cyan}  d0a1 pack - WordPress + IA (Windows)${NC}"
+Write-Host "${Bold}${Cyan}  WP + IA local. Sin cuotas.${NC}"
+Write-Host "${Bold}${Cyan}============================================================${NC}"
+Write-Host ''
+
+# --- Stop mode ------------------------------------------------------------
+if ($Mode -eq 'stop') {
+    Warn 'Parando...'
+    docker compose -f $ComposeFile down
+    Info 'Parado.'
+    exit 0
+}
+
+# --- Check Docker ---------------------------------------------------------
+if (-not (Get-Command docker -ErrorAction SilentlyContinue)) {
+    Err 'Docker no esta instalado.'
+    Write-Host ''
+    Write-Host 'Instala Docker Desktop para Windows:'
+    Write-Host '  https://docs.docker.com/desktop/setup/install/windows-install/'
+    exit 1
+}
+# En PS 5.1, *> $null sobre comandos nativos convierte stderr en errores
+# terminantes con EAP=Stop (PS #4002). Se descarta via cmd /c.
+$null = cmd /c "docker info >NUL 2>&1"
+if ($LASTEXITCODE -ne 0) {
+    Err 'Docker Desktop no responde. Abre Docker Desktop y espera a que arranque.'
+    exit 1
+}
+# En PS 5.1, *> $null sobre comandos nativos convierte stderr en errores
+# terminantes con EAP=Stop (PS #4002). Se descarta via cmd /c.
+cmd /c "docker compose version >NUL 2>&1"
+$composeOk = ($LASTEXITCODE -eq 0)
+if (-not $composeOk) {
+    Err 'Docker Compose v2 no disponible. Actualiza Docker Desktop.'
+    exit 1
+}
+Ok "Docker: $(cmd /c 'docker compose version --short 2>NUL')"
+
+# --- Check RAM ------------------------------------------------------------
+$TotalRamGb = [math]::Round((Get-CimInstance Win32_ComputerSystem).TotalPhysicalMemory / 1GB, 1)
+if ($TotalRamGb -lt 5.5) {
+    Err "Se necesitan 6 GB de RAM minimo (detectado: $TotalRamGb GB)."
+    exit 1
+}
+Ok "RAM: $TotalRamGb GB"
+
+# --- Setup .env -----------------------------------------------------------
+$EnvFile = Join-Path $ScriptDir '.env'
+if (-not (Test-Path $EnvFile)) {
+    Warn 'No se encontro .env. Creando desde .env.example...'
+    Copy-Item (Join-Path $ScriptDir '.env.example') $EnvFile
+
+    # Generar contrasenas aleatorias (32 hex chars) sin depender de openssl
+    $Bytes = New-Object byte[] 16
+    $RootPass = ''; $WpPass = ''
+    $rng = [System.Security.Cryptography.RandomNumberGenerator]::Create()
+    $rng.GetBytes($Bytes); $RootPass = ($Bytes | ForEach-Object { $_.ToString('x2') }) -join ''
+    $rng.GetBytes($Bytes); $WpPass   = ($Bytes | ForEach-Object { $_.ToString('x2') }) -join ''
+    $rng.Dispose()
+
+    # 1a coincidencia en el fichero = MYSQL_ROOT_PASSWORD, 2a = MYSQL_PASSWORD.
+    # -replace no acepta tercer operando (count) en ninguna version de PS,
+    # asi que se recorre linea a linea.
+    $Lines = Get-Content $EnvFile
+    $Replaced = 0
+    for ($i = 0; $i -lt $Lines.Count; $i++) {
+        if ($Lines[$i] -like '*change-me-to-a-strong-password*') {
+            if ($Replaced -eq 0) {
+                $Lines[$i] = $Lines[$i] -replace 'change-me-to-a-strong-password', $RootPass
+            } else {
+                $Lines[$i] = $Lines[$i] -replace 'change-me-to-a-strong-password', $WpPass
+            }
+            $Replaced++
+        }
+    }
+    Set-Content $EnvFile -Value $Lines
+    Info "Contrasenas generadas en $EnvFile"
+}
+
+# --- GPU NVIDIA? (si hay, Ollama la usa y va 5-10x mas rapido) ---------------
+$HasGpu = $false
+if (Get-Command nvidia-smi -ErrorAction SilentlyContinue) {
+    $GpuInfo = & nvidia-smi --query-gpu=name,memory.total --format=csv,noheader 2>$null
+    if ($LASTEXITCODE -eq 0 -and $GpuInfo) {
+        $HasGpu = $true
+        $GpuName = ($GpuInfo -split ',')[0].Trim()
+        Info "GPU NVIDIA detectada: $GpuName (aceleracion activada)"
+    }
+}
+if (-not $HasGpu) {
+    Warn 'Sin GPU NVIDIA: modo CPU (~10-20 tokens/s). Con una GPU NVIDIA va 5-10x mas rapido.'
+}
+# --- Auto-detect ports ----------------------------------------------------
+# --- Tumbar deploy previo (antes de detectar puertos) ----------------------
+cmd /c docker compose -f $ComposeFile down '>' NUL '2>&1'
+Info 'Deploy previo detenido (down; volumenes con el modelo intactos).'
+
+$OllamaPort = Find-FreePort 11434
+$WpPort     = Find-FreePort 8080
+$WebUiPort  = Find-FreePort 3000
+
+# --- Mode info ------------------------------------------------------------
+if ($Mode -eq 'local') {
+    Warn 'Modo: local (tu ordenador)'
+    Write-Host "  WordPress: ${Cyan}http://localhost:${WpPort}${NC}"
+    Write-Host "  IA (chat): ${Cyan}http://localhost:${WebUiPort}${NC}"
+} else {
+    Warn 'Modo: VPS (accesible desde fuera)'
+    # IP real de LAN: la de la interfaz con la ruta por defecto (evita
+    # vEthernet WSL/Hyper-V, Docker, VPN)
+    $GwIf = (Get-NetRoute -DestinationPrefix '0.0.0.0/0' -ErrorAction SilentlyContinue |
+        Sort-Object RouteMetric | Select-Object -First 1).ifIndex
+    $Ip = (Get-NetIPAddress -AddressFamily IPv4 -ErrorAction SilentlyContinue |
+        Where-Object { $_.ifIndex -eq $GwIf -and $_.IPAddress -notlike '127.*' } |
+        Select-Object -First 1).IPAddress
+    if (-not $Ip) { $Ip = 'TU_IP' }
+    Write-Host "  WordPress: ${Cyan}http://${Ip}:${WpPort}${NC}"
+    Write-Host "  IA (chat): ${Cyan}http://${Ip}:${WebUiPort}${NC}"
+    Write-Host ''
+    Write-Host "  ${Bold}¿No tienes VPS?${NC} Consigue uno con descuento:"
+    Write-Host "  ${Cyan}https://d0a1.es/go/vultr/${NC}  - Vultr"
+    Write-Host "  ${Cyan}https://d0a1.es/go/digitalocean/${NC}  - DigitalOcean (`$200 credit)"
+}
+
+if ($OllamaPort -ne 11434) { Warn "Puerto 11434 ocupado, usando $OllamaPort" }
+if ($WpPort     -ne 8080)  { Warn "Puerto 8080 ocupado, usando $WpPort" }
+if ($WebUiPort  -ne 3000)  { Warn "Puerto 3000 ocupado, usando $WebUiPort" }
+Write-Host ''
+
+# --- Deploy ----------------------------------------------------------------
+Warn 'Descargando contenedores (3-5 min la primera vez)...'
+cmd /c docker compose -f $ComposeFile down '>' NUL '2>&1'
+$env:OLLAMA_PORT = "$OllamaPort"
+$env:WP_PORT     = "$WpPort"
+$env:WEBUI_PORT  = "$WebUiPort"
+$ComposeArgs = @('-f', $ComposeFile)
+if ($HasGpu) {
+    $OverrideFile = Join-Path $ScriptDir 'docker-compose.gpu.yml'
+    Set-Content -Path $OverrideFile -Value "services:`r`n  ollama:`r`n    gpus: all`r`n"
+    $ComposeArgs += @('-f', $OverrideFile)
+}
+docker compose @ComposeArgs up -d
+if ($LASTEXITCODE -ne 0) { Err "docker compose up fallo (exit $LASTEXITCODE)."; exit 1 }
+
+# --- Pull model -------------------------------------------------------------
+Info 'Esperando a que Ollama arranque...'
+$MaxWait = 60; $Waited = 0
+while ($true) {
+    # Descarte de stderr via cmd /c (ver nota de PS 5.1 / PS #4002 arriba)
+    cmd /c "docker exec d0a1-wp-ollama ollama ps >NUL 2>&1"
+    if ($LASTEXITCODE -eq 0) { break }
+    Start-Sleep -Seconds 2
+    $Waited += 2
+    if ($Waited -ge $MaxWait) {
+        Err 'Ollama no arranco. Revisa: docker logs d0a1-wp-ollama'
+        exit 1
+    }
+}
+# Si el usuario define $env:HTTPS_PROXY, Ollama lo usa para el pull (doc oficial):
+$ProxyEnv = @()
+if ($env:HTTPS_PROXY) { $ProxyEnv = @('-e', "HTTPS_PROXY=$env:HTTPS_PROXY", '-e', "HTTP_PROXY=$env:HTTP_PROXY") }
+# --- Modelo ya presente? (volumen persistente: no re-descargar) ------------
+$ModelList = docker exec d0a1-wp-ollama ollama list 2>$null
+if ("$ModelList" -match 'qwen3:4b') {
+    Info 'Modelo qwen3:4b ya descargado (volumen intacto). Omitiendo descarga.'
+} else {
+    # --- Modelo: HF primero (CDN propio, rapido) y registry de Ollama como ultimo recurso ---
+
+$Gguf = Join-Path $ScriptDir 'qwen3-4b-q4.gguf'
+$HfUrls = @(
+    'https://huggingface.co/Qwen/Qwen3-4B-GGUF/resolve/main/Qwen3-4B-Q4_K_M.gguf',
+    'https://huggingface.co/unsloth/Qwen3-4B-GGUF/resolve/main/Qwen3-4B-Q4_K_M.gguf'
+)
+Info 'Descargando modelo de IA (qwen3:4b, ~2.5 GB)...'
+foreach ($Url in $HfUrls) {
+    curl.exe -# -L --retry 3 --max-time 1800 -o "$Gguf" $Url
+    if ($LASTEXITCODE -eq 0 -and (Test-Path $Gguf) -and (Get-Item $Gguf).Length -gt 2400000000) { break }
+    Remove-Item "$Gguf" -Force -ErrorAction SilentlyContinue
+    if ($Url -ne $HfUrls[-1]) { Warn 'Mirror no accesible, probando el siguiente...' }
+}
+if ((Test-Path $Gguf) -and (Get-Item $Gguf).Length -gt 2400000000) {
+    Info 'Importando modelo en Ollama...'
+    docker cp "$Gguf" d0a1-wp-ollama:/root/.ollama/import.gguf
+    docker exec d0a1-wp-ollama sh -c 'printf "FROM /root/.ollama/import.gguf\\nPARAMETER num_ctx 16384\\n" > /tmp/mf && ollama create qwen3:4b -f /tmp/mf'
+    $CreateOk = ($LASTEXITCODE -eq 0)
+    docker exec d0a1-wp-ollama rm -f /root/.ollama/import.gguf | Out-Null
+    Remove-Item "$Gguf" -Force -ErrorAction SilentlyContinue
+    $PullOk = $CreateOk
+    if (-not $CreateOk) { Err 'Fallo al importar el modelo desde el GGUF.' }
+} else {
+    # Ultimo recurso: pull directo (1 intento, ya sin esperar 3 rondas)
+    Warn 'HuggingFace no accesible. Probando el registry de Ollama (ultimo recurso)...'
+    docker exec @ProxyEnv d0a1-wp-ollama ollama pull qwen3:4b
+    if ($LASTEXITCODE -eq 0) {
+        $PullOk = $true
+        # Contexto ampliado (el default 4096 da 400 en Open WebUI con prompts largos)
+        docker exec d0a1-wp-ollama sh -c 'printf "FROM qwen3:4b\\nPARAMETER num_ctx 16384\\n" > /tmp/mf && ollama create qwen3:4b -f /tmp/mf'
+    }
+}
+if (-not $PullOk) {
+    Err 'No se pudo descargar el modelo (ni HuggingFace ni registry).'
+    if (-not $env:HTTPS_PROXY) {
+        Write-Host '  Tu red bloquea ambos. Si tienes proxy/VPN:'
+        Write-Host '    $env:HTTPS_PROXY = "http://127.0.0.1:7890"   # (puerto de tu proxy)'
+        Write-Host '    .\deploy.ps1 vps'
+    }
+    exit 1
+}
+}
+
+# --- Precalentar el modelo (load a GPU/RAM ya, no en el primer mensaje) -----
+Info 'Precargando el modelo en memoria...'
+cmd /c docker exec d0a1-wp-ollama sh -c 'ollama run qwen3:4b ""' '>' NUL '2>&1'
+
+# --- Done -------------------------------------------------------------------
+Write-Host ''
+Info 'Listo! Tu WordPress con IA esta corriendo.'
+Write-Host ''
+if ($Mode -eq 'local') {
+    Write-Host "  ${Bold}WordPress:${NC} ${Cyan}http://localhost:${WpPort}${NC}"
+    Write-Host "  ${Bold}IA (chat):${NC} ${Cyan}http://localhost:${WebUiPort}${NC}"
+} else {
+    Write-Host "  ${Bold}WordPress:${NC} ${Cyan}http://${Ip}:${WpPort}${NC}"
+    Write-Host "  ${Bold}IA (chat):${NC} ${Cyan}http://${Ip}:${WebUiPort}${NC}"
+}
+Write-Host ''
+Write-Host "  ${Bold}Configurar WordPress:${NC}"
+Write-Host '  1. Abre la URL de WordPress en tu navegador'
+Write-Host '  2. Sigue el asistente de instalacion (2 minutos)'
+Write-Host "  3. Instala el plugin 'AI Agent Bridge' desde el panel"
+Write-Host ''
+Write-Host "  ${Bold}Parar:${NC} .\deploy.ps1 stop"
+Write-Host "  ${Bold}¿Quieres mas?${NC} ${Cyan}https://d0a1.es/packs/${NC}"
+Write-Host ''

wordpress/deploy.sh

@@ -64,9 +64,8 @@ if [ ! -f "$SCRIPT_DIR/.env" ]; then
     # Generate random passwords for local dev
     ROOT_PASS=$(openssl rand -hex 16 2>/dev/null || cat /dev/urandom | tr -dc 'a-zA-Z0-9' | fold -w 32 | head -n1)
     WP_PASS=$(openssl rand -hex 16 2>/dev/null || cat /dev/urandom | tr -dc 'a-zA-Z0-9' | fold -w 32 | head -n1)
-    sed -i "s/change-me-to-a-strong-password/${ROOT_PASS}/g" "$SCRIPT_DIR/.env"
+    sed -i "0,/change-me-to-a-strong-password/{s/change-me-to-a-strong-password/${ROOT_PASS}/}" "$SCRIPT_DIR/.env"
-    # Second occurrence is MYSQL_PASSWORD
+    sed -i "s/change-me-to-a-strong-password/${WP_PASS}/" "$SCRIPT_DIR/.env"
-    sed -i "0,/change-me-to-a-strong-password/{s/change-me-to-a-strong-password/${WP_PASS}/}" "$SCRIPT_DIR/.env"
     echo -e "${GREEN}✓${NC} Contraseñas generadas en ${SCRIPT_DIR}/.env"
 fi
 
@@ -125,7 +124,7 @@ echo -e "${YELLOW}Descargando contenedores (3-5 min la primera vez)...${NC}"
 docker compose -f "$COMPOSE_FILE" up -d
 
 # ── Pull model ───────────────────────────────
-echo -e "${YELLOW}Descargando modelo de IA (llama3.2:3b, ~2 GB)...${NC}"
+echo -e "${YELLOW}Descargando modelo de IA (qwen3:4b, ~2.5 GB)...${NC}"
 MAX_WAIT=60
 WAITED=0
 until docker exec d0a1-wp-ollama ollama ps &>/dev/null; do
@@ -136,7 +135,7 @@ until docker exec d0a1-wp-ollama ollama ps &>/dev/null; do
         exit 1
     fi
 done
-docker exec d0a1-wp-ollama ollama pull llama3.2:3b
+docker exec d0a1-wp-ollama ollama pull qwen3:4b
 
 # ── Done ────────────────────────────────────
 echo ""

wordpress/docker-compose.yml

@@ -52,13 +52,17 @@ services:
       - ollama_data:/root/.ollama
     environment:
       OLLAMA_HOST: "0.0.0.0"
-      OLLAMA_KEEP_ALIVE: "5m"
+      OLLAMA_KEEP_ALIVE: "-1"
     healthcheck:
       test: ["CMD", "ollama", "ps"]
       interval: 15s
       timeout: 5s
       retries: 10
       start_period: 30s
+    deploy:
+      resources:
+        limits:
+          memory: 6G
 
   openwebui:
     container_name: d0a1-wp-openwebui
@@ -73,6 +77,8 @@ services:
       WEBUI_NAME: "d0a1"
       WEBUI_AUTH: "false"
       DEFAULT_LOCALE: "es"
+      CHAT_RESPONSE_MAX_TOOL_CALL_ITERATIONS: "1"
+      DEFAULT_MODEL_PARAMS: '{"function_calling": "legacy"}'
     depends_on:
       ollama:
         condition: service_healthy