chat.ps1

<#
.SYNOPSIS
    AI Server quickstart — PowerShell (5.1 and 7+), no modules to install.

.DESCRIPTION
    Shows the client contract: Bearer auth, model discovery via GET /v1/models,
    a streaming chat completion, and the responses a real AI Server can give you
    (401, 429 free-tier metering, 503 backpressure with Retry-After).

.EXAMPLE
    $env:AISERVER_BASE_URL = "http://192.168.1.42:11436/v1"
    $env:AISERVER_API_KEY  = "ai-suite_..."
    .\chat.ps1 "Why is the sky blue?"
#>
[CmdletBinding()]
param(
    [Parameter(ValueFromRemainingArguments = $true)]
    [string[]]$Prompt
)

$ErrorActionPreference = 'Stop'

$BaseUrl = ($env:AISERVER_BASE_URL, 'http://localhost:11436/v1' -ne $null)[0].TrimEnd('/')
$ApiKey  = $env:AISERVER_API_KEY
$Model   = $env:AISERVER_MODEL     # optional override; otherwise auto-discovered
$Text    = if ($Prompt) { $Prompt -join ' ' } else { 'Say hello in one short sentence.' }

if (-not $ApiKey) {
    Write-Error 'Set AISERVER_API_KEY (issue a key in AI Server -> API keys).'
    exit 1
}
$Headers = @{ Authorization = "Bearer $ApiKey" }

# Turn AI Server's error responses into advice rather than a stack trace.
function Resolve-AiServerError {
    param($ErrorRecord)
    $resp = $ErrorRecord.Exception.Response
    if (-not $resp) {
        Write-Error ("Could not reach $BaseUrl : " + $ErrorRecord.Exception.Message + "`n" +
            "  * Is AI Server running, and does AISERVER_BASE_URL end in /v1?`n" +
            "  * Serving another machine? It must be bound to your network AND allowed`n" +
            "    through the firewall for that port.")
        exit 1
    }
    $code = [int]$resp.StatusCode
    switch ($code) {
        401 {
            Write-Error ("The server rejected the API key (401).`n" +
                "  * Issue a key in AI Server -> API keys and set AISERVER_API_KEY.`n" +
                '  * A server that serves a network always requires a key.')
        }
        429 {
            # Generic (non-AI-Suite) clients are metered on the free tier: a deliberate cap,
            # not a transient failure, so don't retry in a loop.
            if ($resp.Headers['X-AISuite-Upgrade'] -eq '1') {
                Write-Error ("Free-tier limit reached for generic API clients (429).`n" +
                    '  Upgrade the server to Pro to lift it, or try again later.')
            } else {
                Write-Error "Rate limited (429). Retry-After: $($resp.Headers['Retry-After'])s"
            }
        }
        503 { Write-Error "Server busy (503) — retry in $($resp.Headers['Retry-After'])s (a model may be loading)." }
        default { Write-Error "HTTP $code from the server." }
    }
    exit 1
}

# ── 1. Model discovery — ids are installation-specific, so never hard-code them ──
if (-not $Model) {
    try {
        $models = (Invoke-RestMethod -Uri "$BaseUrl/models" -Headers $Headers).data.id
    } catch { Resolve-AiServerError $_ }

    if (-not $models) { Write-Error 'The server reports no models. Install one in AI Server -> Models.'; exit 1 }
    # Prefer a locally-hosted model (no cloud key needed), and skip reasoning models by
    # default: they emit <think> blocks that would just confuse a first run. Any of them
    # still works — set AISERVER_MODEL to choose deliberately.
    $local = @($models | Where-Object { $_ -notlike 'cloud-*' })
    $plain = @($local | Where-Object { $_ -notmatch '(?i)-r1|r1-|qwq|reasoner|thinking' })
    $Model = if ($plain.Count) { $plain[0] } elseif ($local.Count) { $local[0] } else { @($models)[0] }
    Write-Host "[using model $Model; set AISERVER_MODEL to override]" -ForegroundColor DarkGray
}

# ── 2. Streaming chat completion ────────────────────────────────────────────────
# Invoke-RestMethod buffers the whole body, which would defeat streaming — so read
# the response stream directly. Each line is "data: {json}", ending with "data: [DONE]".
$body = @{
    model    = $Model
    messages = @(@{ role = 'user'; content = $Text })
    stream   = $true
} | ConvertTo-Json -Depth 5 -Compress

$req = [System.Net.HttpWebRequest]::Create("$BaseUrl/chat/completions")
$req.Method = 'POST'
$req.ContentType = 'application/json'
$req.Headers.Add('Authorization', "Bearer $ApiKey")
$req.Timeout = 300000
$bytes = [Text.Encoding]::UTF8.GetBytes($body)
$req.GetRequestStream().Write($bytes, 0, $bytes.Length)

try { $response = $req.GetResponse() } catch { Resolve-AiServerError $_ }

$backend = $response.Headers['X-AISuite-Backend']
if ($backend) { Write-Host "[served by $backend]" -ForegroundColor DarkGray }

$reader = [IO.StreamReader]::new($response.GetResponseStream())
while (-not $reader.EndOfStream) {
    $line = $reader.ReadLine()
    if (-not $line.StartsWith('data:')) { continue }
    $payload = $line.Substring(5).Trim()
    if ($payload -eq '[DONE]') { break }
    try {
        $delta = ($payload | ConvertFrom-Json).choices[0].delta.content
        if ($delta) { Write-Host -NoNewline $delta }
    } catch { }   # ignore keep-alive / partial frames
}
$reader.Dispose(); $response.Dispose()
Write-Host ''