dsh-codebase-chat 0.25.3 → 0.25.6

package/dist/cli.js

@@ -1623,6 +1623,10 @@ function formatHealthReport(r, lang = "fr") {
   };
   const out = [];
   out.push(`== ${t2.title} \u2014 ${basename(r.projectPath)} ==`);
+  if (r.analyzedFiles === 0) {
+    out.push(lang === "en" ? "No code files detected in this project \u2014 check the path or your .codebase-chat.json ignore rules." : "Aucun fichier de code d\xE9tect\xE9 dans ce projet \u2014 v\xE9rifie le chemin ou les r\xE8gles ignore de .codebase-chat.json.");
+    return out.join("\n");
+  }
   out.push(`${t2.score}: ${r.score}/100 (${r.grade}) \xB7 ${r.analyzedFiles} ${t2.files} \xB7 ${r.importEdges} ${t2.edges}`);
   out.push("");
   out.push(`\u25CF ${t2.cycles} (${r.cycles.length})`);
@@ -2566,8 +2570,16 @@ function loadLocalModel() {
         throw new Error(`Local model not found: ${spec}`);
       }
       const { getLlama } = await importLlama();
-      const llama = await getLlama();
-      return llama.loadModel({ modelPath });
+      const cpuOnly = /^(0|off|false|cpu)$/i.test(process.env.CODEBASE_LOCAL_GPU || "");
+      const llama = await getLlama(cpuOnly ? { gpu: false } : {});
+      try {
+        return await llama.loadModel({ modelPath });
+      } catch (err) {
+        if (cpuOnly) throw err;
+        console.error(`[local-llm] GPU load failed (${err instanceof Error ? err.message : err}) \u2014 falling back to CPU. Slower, but works. Set CODEBASE_LOCAL_GPU=off to skip GPU entirely.`);
+        const cpuLlama = await getLlama({ gpu: false });
+        return cpuLlama.loadModel({ modelPath });
+      }
     })();
     modelPromise.catch(() => {
       modelPromise = null;
@@ -2584,13 +2596,42 @@ async function callLocalLlm(prompt, lang = "fr") {
       contextSequence: context.getSequence(),
       systemPrompt: lang === "en" ? "You are a senior codebase analyst. Be precise and cite files with [source: path:line]." : "Tu es un analyste codebase senior. Sois precis et cite les fichiers avec [source: chemin:ligne]."
     });
-    return await session.prompt(prompt, { temperature: 0.2, maxTokens: LOCAL_MAX_TOKENS });
+    return await session.prompt(prompt, {
+      temperature: 0.2,
+      maxTokens: LOCAL_MAX_TOKENS,
+      // Small models degenerate into loops at low temperature — penalize
+      // repeated tokens so the answer moves forward instead of echoing.
+      repeatPenalty: { lastTokens: 128, penalty: 1.2, penalizeNewLine: false }
+    });
   } finally {
     await context.dispose();
   }
 }
 
 // src/cli.ts
+var ANSI = { reset: "\x1B[0m", bold: "\x1B[1m", dim: "\x1B[2m", cyan: "\x1B[36m", yellow: "\x1B[33m", magenta: "\x1B[35m", green: "\x1B[32m" };
+function renderAnswerTerminal(md) {
+  let inCode = false;
+  const inline = (s) => s.replace(/\*\*([^*]+)\*\*/g, `${ANSI.bold}$1${ANSI.reset}`).replace(/`([^`]+)`/g, `${ANSI.cyan}$1${ANSI.reset}`).replace(/\[source: ([^\]]+)\]/g, `${ANSI.dim}${ANSI.cyan}[source: $1]${ANSI.reset}`);
+  return md.split("\n").map((line) => {
+    if (/^\s*```/.test(line)) {
+      inCode = !inCode;
+      return `${ANSI.dim}  \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500${ANSI.reset}`;
+    }
+    if (inCode) return `${ANSI.dim}  ${line}${ANSI.reset}`;
+    const h = line.match(/^(#{1,6})\s+(.*)/);
+    if (h) {
+      const color = h[1].length <= 2 ? `${ANSI.bold}${ANSI.magenta}` : `${ANSI.bold}${ANSI.yellow}`;
+      return `
+${color}${h[2]}${ANSI.reset}`;
+    }
+    const bullet = line.match(/^(\s*)([-*•]|\d+\.)\s+(.*)/);
+    if (bullet) return `${bullet[1]}${ANSI.cyan}${bullet[2]}${ANSI.reset} ${inline(bullet[3])}`;
+    if (/^\s*>/.test(line)) return `${ANSI.dim}${ANSI.green}\u2502${ANSI.reset}${ANSI.dim} ${inline(line.replace(/^\s*>\s?/, ""))}${ANSI.reset}`;
+    if (/^\s*(---+|\*\*\*+)\s*$/.test(line)) return `${ANSI.dim}${"\u2500".repeat(60)}${ANSI.reset}`;
+    return inline(line);
+  }).join("\n");
+}
 var ExitSignal = class {
   constructor(code) {
     this.code = code;
@@ -2813,7 +2854,8 @@ ${lang === "en" ? "Answer" : "R\xE9ponse"} :` : buildToolPrompt(tool, {
     }
     if (values.local || isLocalLlmEnabled()) {
       console.error(lang === "en" ? "Answering with the embedded local model (first run downloads ~1 GB)..." : "R\xE9ponse via le mod\xE8le local embarqu\xE9 (premier lancement : ~1 Go de t\xE9l\xE9chargement)...");
-      console.log(await callLocalLlm(prompt, lang));
+      const answer = await callLocalLlm(prompt, lang);
+      console.log(process.stdout.isTTY ? renderAnswerTerminal(answer) : answer);
       exit(0);
     }
     if (values.call) {
@@ -2844,7 +2886,8 @@ ${lang === "en" ? "Answer" : "R\xE9ponse"} :` : buildToolPrompt(tool, {
         exit(1);
       }
       const data = await res.json();
-      console.log(data.choices?.[0]?.message?.content || "");
+      const answer = data.choices?.[0]?.message?.content || "";
+      console.log(process.stdout.isTTY ? renderAnswerTerminal(answer) : answer);
       exit(0);
     }
     console.log(prompt);
@@ -2869,7 +2912,7 @@ ${lang === "en" ? "Answer" : "R\xE9ponse"} :` : buildToolPrompt(tool, {
     exit(0);
   }
   printHelp();
-  exit(1);
+  exit(0);
 }
 main().catch((err) => {
   disposeTreeSitter();