html/api/ambre-multiagent-v2-external.php

<?php
/**
 * ambre-multiagent-v2-external.php · wave-260
 * MAX parallelism mobilisant IA EXTERNES + internes
 * Agents peuvent être : cascade fast | cascade think | ollama local | tool natif
 */
header("Content-Type: application/json; charset=utf-8");
header("Cache-Control: no-store, no-cache");
header("CDN-Cache-Control: no-store");
set_time_limit(120);

$raw = file_get_contents("php://input");
$in = json_decode($raw, true) ?: $_POST;
$goal = trim($in["goal"] ?? $in["message"] ?? "");
$max_agents = min(12, max(2, intval($in["max_agents"] ?? 6)));

if (!$goal) { echo json_encode(["error"=>"goal required"]); exit; }

$t0 = microtime(true);

// PLAN with richer agent diversity
$plan_sys = "Tu es un planificateur multi-agent WEVAL. Pour l'objectif donné, génère UN JSON STRICT :\n" .
    "{\n" .
    '  "objective": "<reformulation>",' . "\n" .
    '  "agents": [' . "\n" .
    '    {"role":"<nom>", "task":"<tâche autonome>", "tool":"<type>", "model_hint":"<fast|think|ollama|tool_spec>"},' . "\n" .
    "    ...\n" .
    "  ]\n" .
    "}\n" .
    "Tools disponibles:\n" .
    "- llm_fast (Cerebras/Groq rapide)\n" .
    "- llm_think (Cerebras think raisonnement profond)\n" .
    "- llm_ollama (local qwen3/llama3.2 souverain)\n" .
    "- pdf_premium (génère PDF 5 pages)\n" .
    "- mermaid (diagramme SVG)\n" .
    "- web_search (recherche web)\n" .
    "- kb_search (mémoire partagée)\n" .
    "- calc (calcul)\n" .
    "Max $max_agents agents · roles variés · tools diversifiés pour parallélisme optimal.";

$plan_raw = @file_get_contents("http://127.0.0.1:4000/v1/chat/completions", false, stream_context_create([
    "http" => ["method"=>"POST","header"=>"Content-Type: application/json\r\n",
        "content"=>json_encode(["model"=>"fast","messages"=>[["role"=>"system","content"=>$plan_sys],["role"=>"user","content"=>"Objectif: $goal"]],"max_tokens"=>1000,"temperature"=>0.4]),
        "timeout"=>25],
]));
$pd = @json_decode($plan_raw, true);
$pt = $pd["choices"][0]["message"]["content"] ?? "";
$pt = preg_replace('/^```(?:json)?\s*/m', '', $pt);
$pt = preg_replace('/\s*```\s*$/m', '', trim($pt));
$plan = @json_decode($pt, true);

if (!$plan || !isset($plan["agents"])) {
    echo json_encode(["error"=>"plan invalid", "raw"=>substr($pt, 0, 400)]);
    exit;
}

$plan_ms = round((microtime(true)-$t0)*1000);

// EXECUTE
$agents = array_slice($plan["agents"], 0, $max_agents);
$mh = curl_multi_init();
$handles = [];

foreach ($agents as $i => $agent) {
    $tool = strtolower($agent["tool"] ?? "llm_fast");
    $task = $agent["task"] ?? "";
    $ch = curl_init();
    
    if ($tool === "llm_ollama") {
        // Ollama LOCAL sovereign (qwen3:4b fastest)
        curl_setopt_array($ch, [
            CURLOPT_URL => "http://127.0.0.1:11434/api/generate",
            CURLOPT_POST => true,
            CURLOPT_HTTPHEADER => ["Content-Type: application/json"],
            CURLOPT_POSTFIELDS => json_encode(["model"=>"qwen3:4b","prompt"=>$task,"stream"=>false,"options"=>["num_predict"=>400]]),
            CURLOPT_RETURNTRANSFER => true, CURLOPT_TIMEOUT => 60, CURLOPT_CONNECTTIMEOUT => 3,
        ]);
    } elseif ($tool === "llm_think") {
        curl_setopt_array($ch, [
            CURLOPT_URL => "http://127.0.0.1:4000/v1/chat/completions",
            CURLOPT_POST => true,
            CURLOPT_HTTPHEADER => ["Content-Type: application/json"],
            CURLOPT_POSTFIELDS => json_encode(["model"=>"think","messages"=>[["role"=>"user","content"=>$task]],"max_tokens"=>500]),
            CURLOPT_RETURNTRANSFER => true, CURLOPT_TIMEOUT => 45, CURLOPT_CONNECTTIMEOUT => 3,
        ]);
    } elseif ($tool === "pdf_premium") {
        curl_setopt_array($ch, [
            CURLOPT_URL => "http://127.0.0.1/api/ambre-tool-pdf-premium.php",
            CURLOPT_POST => true, CURLOPT_HTTPHEADER => ["Content-Type: application/json"],
            CURLOPT_POSTFIELDS => json_encode(["topic"=>$task]),
            CURLOPT_RETURNTRANSFER => true, CURLOPT_TIMEOUT => 50, CURLOPT_CONNECTTIMEOUT => 3,
        ]);
    } elseif ($tool === "mermaid") {
        curl_setopt_array($ch, [
            CURLOPT_URL => "http://127.0.0.1/api/ambre-tool-mermaid.php",
            CURLOPT_POST => true, CURLOPT_HTTPHEADER => ["Content-Type: application/json"],
            CURLOPT_POSTFIELDS => json_encode(["topic"=>$task]),
            CURLOPT_RETURNTRANSFER => true, CURLOPT_TIMEOUT => 20, CURLOPT_CONNECTTIMEOUT => 3,
        ]);
    } elseif ($tool === "web_search") {
        curl_setopt_array($ch, [
            CURLOPT_URL => "http://127.0.0.1/api/ambre-tool-web-search.php",
            CURLOPT_POST => true, CURLOPT_HTTPHEADER => ["Content-Type: application/json"],
            CURLOPT_POSTFIELDS => json_encode(["query"=>$task]),
            CURLOPT_RETURNTRANSFER => true, CURLOPT_TIMEOUT => 30, CURLOPT_CONNECTTIMEOUT => 3,
        ]);
    } elseif ($tool === "kb_search") {
        curl_setopt_array($ch, [
            CURLOPT_URL => "http://127.0.0.1/api/ambre-mermaid-learn.php",
            CURLOPT_POST => true, CURLOPT_HTTPHEADER => ["Content-Type: application/json"],
            CURLOPT_POSTFIELDS => json_encode(["action"=>"search","query"=>$task]),
            CURLOPT_RETURNTRANSFER => true, CURLOPT_TIMEOUT => 10, CURLOPT_CONNECTTIMEOUT => 3,
        ]);
    } elseif ($tool === "calc") {
        curl_setopt_array($ch, [
            CURLOPT_URL => "http://127.0.0.1/api/ambre-tool-calc.php",
            CURLOPT_POST => true, CURLOPT_HTTPHEADER => ["Content-Type: application/json"],
            CURLOPT_POSTFIELDS => json_encode(["expression"=>$task]),
            CURLOPT_RETURNTRANSFER => true, CURLOPT_TIMEOUT => 10, CURLOPT_CONNECTTIMEOUT => 3,
        ]);
    } else {
        // default = llm_fast (Cerebras)
        curl_setopt_array($ch, [
            CURLOPT_URL => "http://127.0.0.1:4000/v1/chat/completions",
            CURLOPT_POST => true, CURLOPT_HTTPHEADER => ["Content-Type: application/json"],
            CURLOPT_POSTFIELDS => json_encode(["model"=>"fast","messages"=>[["role"=>"user","content"=>$task]],"max_tokens"=>400]),
            CURLOPT_RETURNTRANSFER => true, CURLOPT_TIMEOUT => 30, CURLOPT_CONNECTTIMEOUT => 3,
        ]);
        $tool = "llm_fast";
    }
    
    curl_multi_add_handle($mh, $ch);
    $handles[$i] = ["ch"=>$ch, "agent"=>$agent, "tool_used"=>$tool, "t0"=>microtime(true)];
}

$exec_t0 = microtime(true);
$running = null;
do { curl_multi_exec($mh, $running); curl_multi_select($mh, 0.5); } while ($running > 0);

$results = [];
foreach ($handles as $i => $h) {
    $output = curl_multi_getcontent($h["ch"]);
    $http = curl_getinfo($h["ch"], CURLINFO_HTTP_CODE);
    $ems = round((microtime(true)-$h["t0"])*1000);
    curl_multi_remove_handle($mh, $h["ch"]);
    curl_close($h["ch"]);
    
    $d = @json_decode($output, true);
    $summary = "";
    if ($d) {
        if (isset($d["mermaid_code"])) $summary = "📊 Mermaid " . strlen($d["mermaid_code"]) . "B · source=" . ($d["source"] ?? "?");
        elseif (isset($d["url"])) $summary = "📄 " . $d["url"] . " · " . ($d["size_kb"] ?? "?") . "KB";
        elseif (isset($d["choices"][0]["message"]["content"])) $summary = substr($d["choices"][0]["message"]["content"], 0, 350);
        elseif (isset($d["response"])) $summary = substr($d["response"], 0, 350);
        elseif (isset($d["answer"])) $summary = substr($d["answer"], 0, 350);
        elseif (isset($d["result"])) $summary = (string)$d["result"];
        else $summary = substr($output, 0, 200);
    } else $summary = substr($output ?: "empty", 0, 200);
    
    $results[] = [
        "agent" => $h["agent"]["role"] ?? "agent_$i",
        "task" => $h["agent"]["task"] ?? "",
        "tool" => $h["tool_used"],
        "http" => $http,
        "elapsed_ms" => $ems,
        "summary" => $summary,
    ];
}
curl_multi_close($mh);
$exec_ms = round((microtime(true)-$exec_t0)*1000);

// RECONCILE via Groq (souverain non-Cerebras pour diversité)
$synth_input = "Objectif: $goal\n\nRésultats " . count($results) . " agents :\n";
foreach ($results as $r) {
    $synth_input .= "• " . $r["agent"] . " (tool=" . $r["tool"] . ", " . $r["elapsed_ms"] . "ms): " . substr($r["summary"], 0, 250) . "\n";
}
$synth_input .= "\nProduis une synthèse finale en français, structurée, actionnable. Max 300 mots.";

$synth_raw = @file_get_contents("http://127.0.0.1:4000/v1/chat/completions", false, stream_context_create([
    "http"=>["method"=>"POST","header"=>"Content-Type: application/json\r\n",
        "content"=>json_encode(["model"=>"fast","messages"=>[["role"=>"user","content"=>$synth_input]],"max_tokens"=>600]),
        "timeout"=>20],
]));
$sd = @json_decode($synth_raw, true);
$reconciled = $sd["choices"][0]["message"]["content"] ?? "synthèse échouée";

// Register in learning pool
$shared_kb_file = "/opt/wevads/internal-memory/_shared-learning.json";
$shared_kb = @json_decode(@file_get_contents($shared_kb_file), true) ?: [];
$shared_kb[] = [
    "ts" => time(), "source" => "multiagent-v2",
    "topic" => $plan["objective"] ?? $goal,
    "synthesis_preview" => substr($reconciled, 0, 300),
    "agents" => count($results),
];
if (count($shared_kb) > 500) $shared_kb = array_slice($shared_kb, -500);
@file_put_contents($shared_kb_file, json_encode($shared_kb, JSON_UNESCAPED_UNICODE));

echo json_encode([
    "ok" => true,
    "goal" => $goal,
    "plan" => $plan,
    "plan_ms" => $plan_ms,
    "results" => $results,
    "exec_ms" => $exec_ms,
    "parallel_speedup" => round(array_sum(array_column($results, "elapsed_ms")) / max($exec_ms, 1), 2),
    "reconciled" => trim($reconciled),
    "total_ms" => round((microtime(true)-$t0)*1000),
    "agents_count" => count($results),
    "tools_diversity" => count(array_unique(array_column($results, "tool"))),
    "provider" => "WEVIA Multi-Agent V2 · external IA dispatch · wave-260",
    "shared_kb_size" => count($shared_kb),
], JSON_PRETTY_PRINT|JSON_UNESCAPED_UNICODE);