@relayplane/proxy 1.5.2 → 1.5.4

package/dist/standalone-proxy.js

@@ -418,7 +418,7 @@ function shutdownHistory() {
     }
     flushHistoryBuffer();
 }
-function logRequest(originalModel, targetModel, provider, latencyMs, success, mode, escalated) {
+function logRequest(originalModel, targetModel, provider, latencyMs, success, mode, escalated, taskType, complexity) {
     const timestamp = new Date().toISOString();
     const status = success ? '✓' : '✗';
     const escalateTag = escalated ? ' [ESCALATED]' : '';
@@ -459,6 +459,8 @@ function logRequest(originalModel, targetModel, provider, latencyMs, success, mo
         tokensIn: 0,
         tokensOut: 0,
         costUsd: 0,
+        taskType: taskType || 'general',
+        complexity: complexity || 'simple',
     };
     requestHistory.push(entry);
     if (requestHistory.length > MAX_HISTORY) {
@@ -1111,6 +1113,15 @@ async function forwardToGemini(request, targetModel, apiKey) {
             parts: [{ text: systemMessage.content }],
         };
     }
+    if (request.tools && request.tools.length > 0) {
+        geminiBody["tools"] = [{
+                functionDeclarations: request.tools.map((t) => ({
+                    name: t.function.name,
+                    description: t.function.description || "",
+                    parameters: t.function.parameters || {}
+                }))
+            }];
+    }
     const response = await fetch(`https://generativelanguage.googleapis.com/v1beta/models/${targetModel}:generateContent?key=${apiKey}`, {
         method: 'POST',
         headers: {
@@ -1140,6 +1151,15 @@ async function forwardToGeminiStream(request, targetModel, apiKey) {
             parts: [{ text: systemMessage.content }],
         };
     }
+    if (request.tools && request.tools.length > 0) {
+        geminiBody["tools"] = [{
+                functionDeclarations: request.tools.map((t) => ({
+                    name: t.function.name,
+                    description: t.function.description || "",
+                    parameters: t.function.parameters || {}
+                }))
+            }];
+    }
     const response = await fetch(`https://generativelanguage.googleapis.com/v1beta/models/${targetModel}:streamGenerateContent?alt=sse&key=${apiKey}`, {
         method: 'POST',
         headers: {
@@ -1154,14 +1174,30 @@ async function forwardToGeminiStream(request, targetModel, apiKey) {
  */
 function convertGeminiResponse(geminiData, model) {
     const candidate = geminiData.candidates?.[0];
-    const text = candidate?.content?.parts?.map((p) => p.text ?? '').join('') ?? '';
+    const parts = candidate?.content?.parts ?? [];
+    const text = parts.map((p) => p.text ?? '').join('');
+    const functionCalls = parts.filter((p) => p.functionCall);
     let finishReason = 'stop';
-    if (candidate?.finishReason === 'MAX_TOKENS') {
+    if (functionCalls.length > 0) {
+        finishReason = 'tool_calls';
+    }
+    else if (candidate?.finishReason === 'MAX_TOKENS') {
         finishReason = 'length';
     }
     else if (candidate?.finishReason === 'SAFETY') {
         finishReason = 'content_filter';
     }
+    const message = { role: 'assistant', content: text || null };
+    if (functionCalls.length > 0) {
+        message['tool_calls'] = functionCalls.map((p, i) => ({
+            id: `call_${Date.now()}_${i}`,
+            type: 'function',
+            function: {
+                name: p.functionCall.name,
+                arguments: JSON.stringify(p.functionCall.args || {})
+            }
+        }));
+    }
     return {
         id: `chatcmpl-${Date.now()}`,
         object: 'chat.completion',
@@ -1170,10 +1206,7 @@ function convertGeminiResponse(geminiData, model) {
         choices: [
             {
                 index: 0,
-                message: {
-                    role: 'assistant',
-                    content: text,
-                },
+                message,
                 finish_reason: finishReason,
             },
         ],
@@ -1190,20 +1223,36 @@ function convertGeminiResponse(geminiData, model) {
  */
 function convertGeminiStreamEvent(eventData, messageId, model, isFirst) {
     const candidate = eventData.candidates?.[0];
-    const text = candidate?.content?.parts?.map((p) => p.text ?? '').join('') ?? '';
+    const parts = candidate?.content?.parts ?? [];
+    const text = parts.map((p) => p.text ?? '').join('');
+    const functionCalls = parts.filter((p) => p.functionCall);
     const choice = {
         index: 0,
         delta: {},
         finish_reason: null,
     };
+    const delta = {};
     if (isFirst) {
-        choice['delta'] = { role: 'assistant', content: text };
+        delta['role'] = 'assistant';
+    }
+    if (text) {
+        delta['content'] = text;
     }
-    else if (text) {
-        choice['delta'] = { content: text };
+    if (functionCalls.length > 0) {
+        delta['tool_calls'] = functionCalls.map((p, i) => ({
+            index: i,
+            id: `call_${messageId}_${i}`,
+            type: 'function',
+            function: {
+                name: p.functionCall.name,
+                arguments: JSON.stringify(p.functionCall.args || {})
+            }
+        }));
+        choice['finish_reason'] = 'tool_calls';
     }
+    choice['delta'] = delta;
     // Check for finish
-    if (candidate?.finishReason) {
+    if (candidate?.finishReason && choice['finish_reason'] === null) {
         let finishReason = 'stop';
         if (candidate.finishReason === 'MAX_TOKENS') {
             finishReason = 'length';
@@ -1773,6 +1822,9 @@ td{padding:8px 12px;border-bottom:1px solid #111318}
 .dot{display:inline-block;width:8px;height:8px;border-radius:50%;margin-right:6px}.dot.up{background:#34d399}.dot.down{background:#ef4444}
 .badge{display:inline-block;padding:2px 8px;border-radius:6px;font-size:.75rem;font-weight:500}
 .badge.ok{background:#052e1633;color:#34d399}.badge.err{background:#2d0a0a;color:#ef4444}
+.badge.tt-code{background:#1e3a5f;color:#60a5fa}.badge.tt-analysis{background:#3b1f6e;color:#a78bfa}.badge.tt-summarization{background:#1a3a2a;color:#6ee7b7}.badge.tt-qa{background:#3a2f1e;color:#fbbf24}.badge.tt-general{background:#1e293b;color:#94a3b8}
+.badge.cx-simple{background:#052e1633;color:#34d399}.badge.cx-moderate{background:#2d2a0a;color:#fbbf24}.badge.cx-complex{background:#2d0a0a;color:#ef4444}
+@media(max-width:768px){.col-tt,.col-cx{display:none}}
 .prov{display:flex;gap:16px;flex-wrap:wrap}.prov-item{display:flex;align-items:center;font-size:.85rem;background:#111318;padding:8px 14px;border-radius:8px;border:1px solid #1e293b}
 </style></head><body>
 <div class="header"><div><h1>⚡ RelayPlane Dashboard</h1></div><div class="meta"><span id="ver"></span> · up <span id="uptime"></span> · refreshes every 5s</div></div>
@@ -1786,7 +1838,7 @@ td{padding:8px 12px;border-bottom:1px solid #111318}
 <table><thead><tr><th>Model</th><th>Requests</th><th>Cost</th><th>% of Total</th></tr></thead><tbody id="models"></tbody></table></div>
 <div class="section"><h2>Provider Status</h2><div class="prov" id="providers"></div></div>
 <div class="section"><h2>Recent Runs</h2>
-<table><thead><tr><th>Time</th><th>Model</th><th>Tokens In</th><th>Tokens Out</th><th>Cost</th><th>Latency</th><th>Status</th></tr></thead><tbody id="runs"></tbody></table></div>
+<table><thead><tr><th>Time</th><th>Model</th><th class="col-tt">Task Type</th><th class="col-cx">Complexity</th><th>Tokens In</th><th>Tokens Out</th><th>Cost</th><th>Latency</th><th>Status</th></tr></thead><tbody id="runs"></tbody></table></div>
 <script>
 const $ = id => document.getElementById(id);
 function fmt(n,d=2){return typeof n==='number'?n.toFixed(d):'-'}
@@ -1811,9 +1863,11 @@ async function load(){
     $('models').innerHTML=(stats.byModel||[]).map(m=>
       '<tr><td>'+m.model+'</td><td>'+m.count+'</td><td>$'+fmt(m.costUsd,4)+'</td><td>'+fmt(m.count/total*100,1)+'%</td></tr>'
     ).join('')||'<tr><td colspan=4 style="color:#64748b">No data yet</td></tr>';
+    function ttCls(t){const m={code_generation:'tt-code',analysis:'tt-analysis',summarization:'tt-summarization',question_answering:'tt-qa'};return m[t]||'tt-general'}
+    function cxCls(c){const m={simple:'cx-simple',moderate:'cx-moderate',complex:'cx-complex'};return m[c]||'cx-simple'}
     $('runs').innerHTML=(runsR.runs||[]).map(r=>
-      '<tr><td>'+fmtTime(r.started_at)+'</td><td>'+r.model+'</td><td>'+(r.tokensIn||0)+'</td><td>'+(r.tokensOut||0)+'</td><td>$'+fmt(r.costUsd,4)+'</td><td>'+r.latencyMs+'ms</td><td><span class="badge '+(r.status==='success'?'ok':'err')+'">'+r.status+'</span></td></tr>'
-    ).join('')||'<tr><td colspan=7 style="color:#64748b">No runs yet</td></tr>';
+      '<tr><td>'+fmtTime(r.started_at)+'</td><td>'+r.model+'</td><td class="col-tt"><span class="badge '+ttCls(r.taskType)+'">'+(r.taskType||'general').replace(/_/g,' ')+'</span></td><td class="col-cx"><span class="badge '+cxCls(r.complexity)+'">'+(r.complexity||'simple')+'</span></td><td>'+(r.tokensIn||0)+'</td><td>'+(r.tokensOut||0)+'</td><td>$'+fmt(r.costUsd,4)+'</td><td>'+r.latencyMs+'ms</td><td><span class="badge '+(r.status==='success'?'ok':'err')+'">'+r.status+'</span></td></tr>'
+    ).join('')||'<tr><td colspan=9 style="color:#64748b">No runs yet</td></tr>';
     $('providers').innerHTML=(provH.providers||[]).map(p=>
       '<div class="prov-item"><span class="dot '+(p.status==='healthy'?'up':'down')+'"></span>'+p.provider+'</div>'
     ).join('');
@@ -2036,7 +2090,8 @@ async function startProxy(config = {}) {
                     started_at: r.timestamp,
                     model: r.targetModel,
                     routed_to: `${r.provider}/${r.targetModel}`,
-                    taskType: r.mode,
+                    taskType: r.taskType || 'general',
+                    complexity: r.complexity || 'simple',
                     costUsd: r.costUsd,
                     latencyMs: r.latencyMs,
                     tokensIn: r.tokensIn,
@@ -2366,7 +2421,7 @@ async function startProxy(config = {}) {
                             cooldownManager.recordFailure(targetProvider, JSON.stringify(errorPayload));
                         }
                         const durationMs = Date.now() - startTime;
-                        logRequest(originalModel ?? 'unknown', targetModel || requestedModel, targetProvider, durationMs, false, routingMode);
+                        logRequest(originalModel ?? 'unknown', targetModel || requestedModel, targetProvider, durationMs, false, routingMode, undefined, taskType, complexity);
                         res.writeHead(providerResponse.status, { 'Content-Type': 'application/json' });
                         res.end(JSON.stringify(errorPayload));
                         return;
@@ -2437,7 +2492,7 @@ async function startProxy(config = {}) {
                     }
                 }
                 const durationMs = Date.now() - startTime;
-                logRequest(originalModel ?? 'unknown', targetModel || requestedModel, targetProvider, durationMs, true, routingMode, useCascade && cascadeConfig ? undefined : false);
+                logRequest(originalModel ?? 'unknown', targetModel || requestedModel, targetProvider, durationMs, true, routingMode, useCascade && cascadeConfig ? undefined : false, taskType, complexity);
                 if (recordTelemetry) {
                     relay
                         .run({
@@ -2454,7 +2509,7 @@ async function startProxy(config = {}) {
             }
             catch (err) {
                 const durationMs = Date.now() - startTime;
-                logRequest(originalModel ?? 'unknown', targetModel || requestedModel, targetProvider, durationMs, false, routingMode);
+                logRequest(originalModel ?? 'unknown', targetModel || requestedModel, targetProvider, durationMs, false, routingMode, undefined, taskType, complexity);
                 if (err instanceof ProviderResponseError) {
                     res.writeHead(err.status, { 'Content-Type': 'application/json' });
                     res.end(JSON.stringify(err.payload));
@@ -2738,7 +2793,7 @@ async function startProxy(config = {}) {
         const startTime = Date.now();
         // Handle streaming vs non-streaming
         if (isStreaming) {
-            await handleStreamingRequest(res, request, targetProvider, targetModel, apiKey, ctx, relay, promptText, taskType, confidence, useCascade ? 'cascade' : routingMode, recordTelemetry, startTime, log, cooldownManager, cooldownsEnabled);
+            await handleStreamingRequest(res, request, targetProvider, targetModel, apiKey, ctx, relay, promptText, taskType, confidence, useCascade ? 'cascade' : routingMode, recordTelemetry, startTime, log, cooldownManager, cooldownsEnabled, complexity);
         }
         else {
             if (useCascade && cascadeConfig) {
@@ -2770,7 +2825,7 @@ async function startProxy(config = {}) {
                     const durationMs = Date.now() - startTime;
                     let responseData = cascadeResult.responseData;
                     // Log cascade request for stats tracking
-                    logRequest(originalRequestedModel ?? 'unknown', cascadeResult.model, cascadeResult.provider, durationMs, true, 'cascade', cascadeResult.escalations > 0);
+                    logRequest(originalRequestedModel ?? 'unknown', cascadeResult.model, cascadeResult.provider, durationMs, true, 'cascade', cascadeResult.escalations > 0, taskType, complexity);
                     const cascadeUsage = responseData?.usage;
                     const cascadeTokensIn = cascadeUsage?.input_tokens ?? cascadeUsage?.prompt_tokens ?? 0;
                     const cascadeTokensOut = cascadeUsage?.output_tokens ?? cascadeUsage?.completion_tokens ?? 0;
@@ -2804,7 +2859,7 @@ async function startProxy(config = {}) {
                 }
                 catch (err) {
                     const durationMs = Date.now() - startTime;
-                    logRequest(originalRequestedModel ?? 'unknown', targetModel || 'unknown', targetProvider, durationMs, false, 'cascade');
+                    logRequest(originalRequestedModel ?? 'unknown', targetModel || 'unknown', targetProvider, durationMs, false, 'cascade', undefined, taskType, complexity);
                     if (err instanceof ProviderResponseError) {
                         res.writeHead(err.status, { 'Content-Type': 'application/json' });
                         res.end(JSON.stringify(err.payload));
@@ -2816,7 +2871,7 @@ async function startProxy(config = {}) {
                 }
             }
             else {
-                await handleNonStreamingRequest(res, request, targetProvider, targetModel, apiKey, ctx, relay, promptText, taskType, confidence, routingMode, recordTelemetry, startTime, log, cooldownManager, cooldownsEnabled);
+                await handleNonStreamingRequest(res, request, targetProvider, targetModel, apiKey, ctx, relay, promptText, taskType, confidence, routingMode, recordTelemetry, startTime, log, cooldownManager, cooldownsEnabled, complexity);
             }
         }
     });
@@ -2889,7 +2944,7 @@ async function executeNonStreamingProviderRequest(request, targetProvider, targe
     }
     return { responseData, ok: true, status: 200 };
 }
-async function handleStreamingRequest(res, request, targetProvider, targetModel, apiKey, ctx, relay, promptText, taskType, confidence, routingMode, recordTelemetry, startTime, log, cooldownManager, cooldownsEnabled) {
+async function handleStreamingRequest(res, request, targetProvider, targetModel, apiKey, ctx, relay, promptText, taskType, confidence, routingMode, recordTelemetry, startTime, log, cooldownManager, cooldownsEnabled, complexity = 'simple') {
     let providerResponse;
     try {
         switch (targetProvider) {
@@ -2917,7 +2972,7 @@ async function handleStreamingRequest(res, request, targetProvider, targetModel,
                 cooldownManager.recordFailure(targetProvider, JSON.stringify(errorData));
             }
             const durationMs = Date.now() - startTime;
-            logRequest(request.model ?? 'unknown', targetModel, targetProvider, durationMs, false, routingMode);
+            logRequest(request.model ?? 'unknown', targetModel, targetProvider, durationMs, false, routingMode, undefined, taskType, complexity);
             res.writeHead(providerResponse.status, { 'Content-Type': 'application/json' });
             res.end(JSON.stringify(errorData));
             return;
@@ -2929,7 +2984,7 @@ async function handleStreamingRequest(res, request, targetProvider, targetModel,
             cooldownManager.recordFailure(targetProvider, errorMsg);
         }
         const durationMs = Date.now() - startTime;
-        logRequest(request.model ?? 'unknown', targetModel, targetProvider, durationMs, false, routingMode);
+        logRequest(request.model ?? 'unknown', targetModel, targetProvider, durationMs, false, routingMode, undefined, taskType, complexity);
         res.writeHead(500, { 'Content-Type': 'application/json' });
         res.end(JSON.stringify({ error: `Provider error: ${errorMsg}` }));
         return;
@@ -3013,7 +3068,7 @@ async function handleStreamingRequest(res, request, targetProvider, targetModel,
     }
     const durationMs = Date.now() - startTime;
     // Always log the request for stats/telemetry tracking
-    logRequest(request.model ?? 'unknown', targetModel, targetProvider, durationMs, true, routingMode);
+    logRequest(request.model ?? 'unknown', targetModel, targetProvider, durationMs, true, routingMode, undefined, taskType, complexity);
     // Update token/cost info on the history entry
     const streamCost = (0, telemetry_js_1.estimateCost)(targetModel, streamTokensIn, streamTokensOut);
     updateLastHistoryEntry(streamTokensIn, streamTokensOut, streamCost);
@@ -3038,7 +3093,7 @@ async function handleStreamingRequest(res, request, targetProvider, targetModel,
 /**
  * Handle non-streaming request
  */
-async function handleNonStreamingRequest(res, request, targetProvider, targetModel, apiKey, ctx, relay, promptText, taskType, confidence, routingMode, recordTelemetry, startTime, log, cooldownManager, cooldownsEnabled) {
+async function handleNonStreamingRequest(res, request, targetProvider, targetModel, apiKey, ctx, relay, promptText, taskType, confidence, routingMode, recordTelemetry, startTime, log, cooldownManager, cooldownsEnabled, complexity = 'simple') {
     let responseData;
     try {
         const result = await executeNonStreamingProviderRequest(request, targetProvider, targetModel, apiKey, ctx);
@@ -3048,7 +3103,7 @@ async function handleNonStreamingRequest(res, request, targetProvider, targetMod
                 cooldownManager.recordFailure(targetProvider, JSON.stringify(responseData));
             }
             const durationMs = Date.now() - startTime;
-            logRequest(request.model ?? 'unknown', targetModel, targetProvider, durationMs, false, routingMode);
+            logRequest(request.model ?? 'unknown', targetModel, targetProvider, durationMs, false, routingMode, undefined, taskType, complexity);
             res.writeHead(result.status, { 'Content-Type': 'application/json' });
             res.end(JSON.stringify(responseData));
             return;
@@ -3060,7 +3115,7 @@ async function handleNonStreamingRequest(res, request, targetProvider, targetMod
             cooldownManager.recordFailure(targetProvider, errorMsg);
         }
         const durationMs = Date.now() - startTime;
-        logRequest(request.model ?? 'unknown', targetModel, targetProvider, durationMs, false, routingMode);
+        logRequest(request.model ?? 'unknown', targetModel, targetProvider, durationMs, false, routingMode, undefined, taskType, complexity);
         res.writeHead(500, { 'Content-Type': 'application/json' });
         res.end(JSON.stringify({ error: `Provider error: ${errorMsg}` }));
         return;
@@ -3070,7 +3125,7 @@ async function handleNonStreamingRequest(res, request, targetProvider, targetMod
     }
     const durationMs = Date.now() - startTime;
     // Log the successful request
-    logRequest(request.model ?? 'unknown', targetModel, targetProvider, durationMs, true, routingMode);
+    logRequest(request.model ?? 'unknown', targetModel, targetProvider, durationMs, true, routingMode, undefined, taskType, complexity);
     // Update token/cost info
     const usage = responseData?.usage;
     const tokensIn = usage?.input_tokens ?? usage?.prompt_tokens ?? 0;