@hasna/terminal 0.5.2 → 0.5.3

package/dist/compression.js

@@ -74,8 +74,8 @@ export function compress(command, output, options = {}) {
             const json = JSON.stringify(parsed.data, null, format === "summary" ? 0 : 2);
             const savings = tokenSavings(output, parsed.data);
             const compressedTokens = estimateTokens(json);
-            // If within budget or no budget, return structured
-            if (!maxTokens || compressedTokens <= maxTokens) {
+            // ONLY use JSON if it actually saves tokens (never return larger output)
+            if (savings.saved > 0 && (!maxTokens || compressedTokens <= maxTokens)) {
                 return {
                     content: json,
                     format: "json",

package/dist/mcp/server.js

@@ -64,17 +64,20 @@ export function createServer() {
                         }) }],
             };
         }
-        // JSON mode — structured parsing
+        // JSON mode — structured parsing (only if it actually saves tokens)
         if (format === "json") {
             const parsed = parseOutput(command, output);
             if (parsed) {
                 const savings = tokenSavings(output, parsed.data);
-                return {
-                    content: [{ type: "text", text: JSON.stringify({
-                                exitCode: result.exitCode, parsed: parsed.data, parser: parsed.parser,
-                                duration: result.duration, tokensSaved: savings.saved, savingsPercent: savings.percent,
-                            }) }],
-                };
+                if (savings.saved > 0) {
+                    return {
+                        content: [{ type: "text", text: JSON.stringify({
+                                    exitCode: result.exitCode, parsed: parsed.data, parser: parsed.parser,
+                                    duration: result.duration, tokensSaved: savings.saved, savingsPercent: savings.percent,
+                                }) }],
+                    };
+                }
+                // JSON was larger — fall through to compression
             }
         }
         // Compressed mode (also fallback for json when no parser matches)

package/dist/output-processor.js

@@ -32,7 +32,11 @@ export async function processOutput(command, output) {
             summary: output,
             full: output,
             tokensSaved: 0,
+            aiTokensUsed: 0,
             aiProcessed: false,
+            aiCostUsd: 0,
+            savingsValueUsd: 0,
+            netSavingsUsd: 0,
         };
     }
     // Truncate very long output before sending to AI
@@ -65,12 +69,30 @@ export async function processOutput(command, output) {
             }
         }
         catch { /* not JSON, that's fine */ }
+        // Cost calculation
+        // AI input: system prompt (~200 tokens) + command + output sent to AI
+        const aiInputTokens = estimateTokens(SUMMARIZE_PROMPT) + estimateTokens(toSummarize) + 20;
+        const aiOutputTokens = summaryTokens;
+        const aiTokensUsed = aiInputTokens + aiOutputTokens;
+        // Cerebras qwen-3-235b pricing: $0.60/M input, $1.20/M output
+        const aiCostUsd = (aiInputTokens * 0.60 + aiOutputTokens * 1.20) / 1_000_000;
+        // Value of tokens saved (at Claude Sonnet $3/M input — what the agent would pay)
+        const savingsValueUsd = (saved * 3.0) / 1_000_000;
+        const netSavingsUsd = savingsValueUsd - aiCostUsd;
+        // Only record savings if net positive (AI cost < token savings value)
+        if (netSavingsUsd > 0 && saved > 0) {
+            recordSaving("compressed", saved);
+        }
         return {
             summary,
             full: output,
             structured,
             tokensSaved: saved,
+            aiTokensUsed,
             aiProcessed: true,
+            aiCostUsd,
+            savingsValueUsd,
+            netSavingsUsd,
         };
     }
     catch {
@@ -82,7 +104,11 @@ export async function processOutput(command, output) {
             summary: fallback,
             full: output,
             tokensSaved: Math.max(0, estimateTokens(output) - estimateTokens(fallback)),
+            aiTokensUsed: 0,
             aiProcessed: false,
+            aiCostUsd: 0,
+            savingsValueUsd: 0,
+            netSavingsUsd: 0,
         };
     }
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hasna/terminal",
-  "version": "0.5.2",
+  "version": "0.5.3",
   "description": "Smart terminal wrapper for AI agents and humans — structured output, token compression, MCP server, natural language",
   "type": "module",
   "bin": {

package/src/compression.ts

@@ -104,8 +104,8 @@ export function compress(command: string, output: string, options: CompressOptio
       const savings = tokenSavings(output, parsed.data);
       const compressedTokens = estimateTokens(json);
 
-      // If within budget or no budget, return structured
-      if (!maxTokens || compressedTokens <= maxTokens) {
+      // ONLY use JSON if it actually saves tokens (never return larger output)
+      if (savings.saved > 0 && (!maxTokens || compressedTokens <= maxTokens)) {
         return {
           content: json,
           format: "json",

package/src/mcp/server.ts

@@ -77,17 +77,20 @@ export function createServer(): McpServer {
         };
       }
 
-      // JSON mode — structured parsing
+      // JSON mode — structured parsing (only if it actually saves tokens)
       if (format === "json") {
         const parsed = parseOutput(command, output);
         if (parsed) {
           const savings = tokenSavings(output, parsed.data);
-          return {
-            content: [{ type: "text" as const, text: JSON.stringify({
-              exitCode: result.exitCode, parsed: parsed.data, parser: parsed.parser,
-              duration: result.duration, tokensSaved: savings.saved, savingsPercent: savings.percent,
-            }) }],
-          };
+          if (savings.saved > 0) {
+            return {
+              content: [{ type: "text" as const, text: JSON.stringify({
+                exitCode: result.exitCode, parsed: parsed.data, parser: parsed.parser,
+                duration: result.duration, tokensSaved: savings.saved, savingsPercent: savings.percent,
+              }) }],
+            };
+          }
+          // JSON was larger — fall through to compression
         }
       }
 

package/src/output-processor.ts

@@ -12,10 +12,18 @@ export interface ProcessedOutput {
   full: string;
   /** Structured JSON if the AI could extract it */
   structured?: Record<string, unknown>;
-  /** How many tokens were saved */
+  /** How many tokens were saved (net, after subtracting AI cost) */
   tokensSaved: number;
+  /** Tokens used by the AI summarization call */
+  aiTokensUsed: number;
   /** Whether AI processing was used (vs passthrough) */
   aiProcessed: boolean;
+  /** Cost of the AI call in USD (Cerebras pricing) */
+  aiCostUsd: number;
+  /** Value of tokens saved in USD (at Claude Sonnet rates) */
+  savingsValueUsd: number;
+  /** Net ROI: savings minus AI cost */
+  netSavingsUsd: number;
 }
 
 const MIN_LINES_TO_PROCESS = 15;
@@ -53,7 +61,11 @@ export async function processOutput(
       summary: output,
       full: output,
       tokensSaved: 0,
+      aiTokensUsed: 0,
       aiProcessed: false,
+      aiCostUsd: 0,
+      savingsValueUsd: 0,
+      netSavingsUsd: 0,
     };
   }
 
@@ -94,12 +106,34 @@ export async function processOutput(
       }
     } catch { /* not JSON, that's fine */ }
 
+    // Cost calculation
+    // AI input: system prompt (~200 tokens) + command + output sent to AI
+    const aiInputTokens = estimateTokens(SUMMARIZE_PROMPT) + estimateTokens(toSummarize) + 20;
+    const aiOutputTokens = summaryTokens;
+    const aiTokensUsed = aiInputTokens + aiOutputTokens;
+
+    // Cerebras qwen-3-235b pricing: $0.60/M input, $1.20/M output
+    const aiCostUsd = (aiInputTokens * 0.60 + aiOutputTokens * 1.20) / 1_000_000;
+
+    // Value of tokens saved (at Claude Sonnet $3/M input — what the agent would pay)
+    const savingsValueUsd = (saved * 3.0) / 1_000_000;
+    const netSavingsUsd = savingsValueUsd - aiCostUsd;
+
+    // Only record savings if net positive (AI cost < token savings value)
+    if (netSavingsUsd > 0 && saved > 0) {
+      recordSaving("compressed", saved);
+    }
+
     return {
       summary,
       full: output,
       structured,
       tokensSaved: saved,
+      aiTokensUsed,
       aiProcessed: true,
+      aiCostUsd,
+      savingsValueUsd,
+      netSavingsUsd,
     };
   } catch {
     // AI unavailable — fall back to simple truncation
@@ -111,7 +145,11 @@ export async function processOutput(
       summary: fallback,
       full: output,
       tokensSaved: Math.max(0, estimateTokens(output) - estimateTokens(fallback)),
+      aiTokensUsed: 0,
       aiProcessed: false,
+      aiCostUsd: 0,
+      savingsValueUsd: 0,
+      netSavingsUsd: 0,
     };
   }
 }