@relayplane/proxy 0.1.4 → 0.1.6

package/dist/cli.mjs

@@ -132,11 +132,13 @@ CREATE TABLE IF NOT EXISTS schema_version (
 INSERT OR IGNORE INTO schema_version (version) VALUES (1);
 `;
 var DEFAULT_ROUTING_RULES = [
-  { taskType: "code_generation", preferredModel: "anthropic:claude-3-5-haiku-latest" },
-  { taskType: "code_review", preferredModel: "anthropic:claude-3-5-haiku-latest" },
+  // Complex tasks → Sonnet (need reasoning & quality)
+  { taskType: "code_generation", preferredModel: "anthropic:claude-sonnet-4-20250514" },
+  { taskType: "code_review", preferredModel: "anthropic:claude-sonnet-4-20250514" },
+  { taskType: "analysis", preferredModel: "anthropic:claude-sonnet-4-20250514" },
+  { taskType: "creative_writing", preferredModel: "anthropic:claude-sonnet-4-20250514" },
+  // Simple tasks → Haiku (cost efficient)
   { taskType: "summarization", preferredModel: "anthropic:claude-3-5-haiku-latest" },
-  { taskType: "analysis", preferredModel: "anthropic:claude-3-5-haiku-latest" },
-  { taskType: "creative_writing", preferredModel: "anthropic:claude-3-5-haiku-latest" },
   { taskType: "data_extraction", preferredModel: "anthropic:claude-3-5-haiku-latest" },
   { taskType: "translation", preferredModel: "anthropic:claude-3-5-haiku-latest" },
   { taskType: "question_answering", preferredModel: "anthropic:claude-3-5-haiku-latest" },
@@ -1603,11 +1605,14 @@ var MODEL_MAPPING = {
   "gpt-4.1": { provider: "openai", model: "gpt-4.1" }
 };
 var DEFAULT_ROUTING = {
-  code_generation: { provider: "anthropic", model: "claude-3-5-haiku-latest" },
-  code_review: { provider: "anthropic", model: "claude-3-5-haiku-latest" },
+  // Complex tasks → Sonnet (need reasoning & quality)
+  code_review: { provider: "anthropic", model: "claude-sonnet-4-20250514" },
+  analysis: { provider: "anthropic", model: "claude-sonnet-4-20250514" },
+  creative_writing: { provider: "anthropic", model: "claude-sonnet-4-20250514" },
+  // Medium tasks → Sonnet (benefit from better model)
+  code_generation: { provider: "anthropic", model: "claude-sonnet-4-20250514" },
+  // Simple tasks → Haiku (cost efficient)
   summarization: { provider: "anthropic", model: "claude-3-5-haiku-latest" },
-  analysis: { provider: "anthropic", model: "claude-3-5-haiku-latest" },
-  creative_writing: { provider: "anthropic", model: "claude-3-5-haiku-latest" },
   data_extraction: { provider: "anthropic", model: "claude-3-5-haiku-latest" },
   translation: { provider: "anthropic", model: "claude-3-5-haiku-latest" },
   question_answering: { provider: "anthropic", model: "claude-3-5-haiku-latest" },
@@ -2592,18 +2597,28 @@ async function handleNonStreamingRequest(res, request, targetProvider, targetMod
 }
 
 // src/cli.ts
+import Database2 from "better-sqlite3";
 function printHelp() {
   console.log(`
 RelayPlane Proxy - Intelligent AI Model Routing
 
 Usage:
-  npx @relayplane/proxy [options]
-  relayplane-proxy [options]
+  npx @relayplane/proxy [command] [options]
+  relayplane-proxy [command] [options]
 
-Options:
+Commands:
+  (default)          Start the proxy server
+  stats              Show routing statistics
+
+Server Options:
   --port <number>    Port to listen on (default: 3001)
   --host <string>    Host to bind to (default: 127.0.0.1)
   -v, --verbose      Enable verbose logging
+
+Stats Options:
+  --days <number>    Days of history to show (default: 7)
+
+General:
   -h, --help         Show this help message
 
 Environment Variables:
@@ -2613,26 +2628,108 @@ Environment Variables:
   XAI_API_KEY        xAI/Grok API key (optional)
   MOONSHOT_API_KEY   Moonshot API key (optional)
 
-Example:
+Examples:
   # Start proxy on default port
   npx @relayplane/proxy
 
   # Start on custom port with verbose logging
   npx @relayplane/proxy --port 8080 -v
 
-  # Then point your SDKs to the proxy
-  export ANTHROPIC_BASE_URL=http://localhost:3001
-  export OPENAI_BASE_URL=http://localhost:3001
+  # View routing stats for last 7 days
+  npx @relayplane/proxy stats
+
+  # View stats for last 30 days
+  npx @relayplane/proxy stats --days 30
 
 Learn more: https://relayplane.com/integrations/openclaw
 `);
 }
+function showStats(days) {
+  const dbPath = getDefaultDbPath();
+  try {
+    const db = new Database2(dbPath, { readonly: true });
+    const cutoff = new Date(Date.now() - days * 24 * 60 * 60 * 1e3).toISOString();
+    const runs = db.prepare(`
+      SELECT 
+        model,
+        task_type,
+        COUNT(*) as count,
+        SUM(tokens_in) as total_in,
+        SUM(tokens_out) as total_out,
+        AVG(duration_ms) as avg_duration,
+        SUM(CASE WHEN success = 1 THEN 1 ELSE 0 END) as successes
+      FROM runs 
+      WHERE created_at >= ?
+      GROUP BY model
+      ORDER BY count DESC
+    `).all(cutoff);
+    const totalRuns = runs.reduce((sum, r) => sum + r.count, 0);
+    const totalTokensIn = runs.reduce((sum, r) => sum + (r.total_in || 0), 0);
+    const totalTokensOut = runs.reduce((sum, r) => sum + (r.total_out || 0), 0);
+    console.log("");
+    console.log(`  \u256D\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u256E`);
+    console.log(`  \u2502      RelayPlane Routing Stats          \u2502`);
+    console.log(`  \u2502          Last ${String(days).padStart(2)} days                  \u2502`);
+    console.log(`  \u2570\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u256F`);
+    console.log("");
+    if (totalRuns === 0) {
+      console.log("  No routing data found for this period.");
+      console.log("  Start using the proxy to collect stats!");
+      console.log("");
+      return;
+    }
+    console.log("  Summary:");
+    console.log(`    Total requests:     ${totalRuns.toLocaleString()}`);
+    console.log(`    Total tokens in:    ${totalTokensIn.toLocaleString()}`);
+    console.log(`    Total tokens out:   ${totalTokensOut.toLocaleString()}`);
+    console.log("");
+    console.log("  By Model:");
+    console.log("  \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500");
+    for (const row of runs) {
+      const pct = (row.count / totalRuns * 100).toFixed(1);
+      const successRate = row.count > 0 ? (row.successes / row.count * 100).toFixed(0) : "0";
+      console.log(`    ${row.model.padEnd(35)} ${String(row.count).padStart(6)} (${pct.padStart(5)}%)  ${successRate}% ok`);
+    }
+    console.log("");
+    const haikuRuns = runs.filter((r) => r.model.includes("haiku"));
+    const haikuTokensIn = haikuRuns.reduce((sum, r) => sum + (r.total_in || 0), 0);
+    const haikuTokensOut = haikuRuns.reduce((sum, r) => sum + (r.total_out || 0), 0);
+    const opusCost = totalTokensIn * 15 / 1e6 + totalTokensOut * 75 / 1e6;
+    const haikuCost = haikuTokensIn * 0.25 / 1e6 + haikuTokensOut * 1.25 / 1e6;
+    const nonHaikuCost = (totalTokensIn - haikuTokensIn) * 3 / 1e6 + (totalTokensOut - haikuTokensOut) * 15 / 1e6;
+    const actualCost = haikuCost + nonHaikuCost;
+    const savings = opusCost - actualCost;
+    if (savings > 0) {
+      console.log("  Estimated Savings:");
+      console.log(`    If all Opus:        $${opusCost.toFixed(2)}`);
+      console.log(`    With routing:       $${actualCost.toFixed(2)}`);
+      console.log(`    Saved:              $${savings.toFixed(2)} (${(savings / opusCost * 100).toFixed(0)}%)`);
+      console.log("");
+    }
+    db.close();
+  } catch (err) {
+    console.error("Error reading stats:", err);
+    console.log("");
+    console.log("  No data found. The proxy stores data at:");
+    console.log(`    ${dbPath}`);
+    console.log("");
+  }
+}
 async function main() {
   const args = process.argv.slice(2);
   if (args.includes("-h") || args.includes("--help")) {
     printHelp();
     process.exit(0);
   }
+  if (args[0] === "stats") {
+    let days = 7;
+    const daysIdx = args.indexOf("--days");
+    if (daysIdx !== -1 && args[daysIdx + 1]) {
+      days = parseInt(args[daysIdx + 1], 10) || 7;
+    }
+    showStats(days);
+    process.exit(0);
+  }
   let port = 3001;
   let host = "127.0.0.1";
   let verbose = false;