dsh-codebase-chat 0.25.0 → 0.25.1

package/dist/cli.js

@@ -2528,6 +2528,68 @@ function buildToolPrompt(tool, opts) {
   return normalizeLabels(builder(opts), opts.lang || "fr");
 }
 
+// src/local-llm.ts
+import { existsSync as existsSync2 } from "fs";
+import { join as join8 } from "path";
+var DEFAULT_MODEL_URI = "hf:Qwen/Qwen2.5-1.5B-Instruct-GGUF:Q4_K_M";
+var LOCAL_CONTEXT_SIZE = 8192;
+var LOCAL_MAX_TOKENS = 1024;
+function isLocalLlmEnabled() {
+  return !!(process.env.CODEBASE_LOCAL_LLM || "").trim();
+}
+function configuredModel() {
+  const v = (process.env.CODEBASE_LOCAL_LLM || "").trim();
+  if (!v || v === "1" || v.toLowerCase() === "true") return DEFAULT_MODEL_URI;
+  return v;
+}
+async function importLlama() {
+  try {
+    return await import("node-llama-cpp");
+  } catch {
+    throw new Error("Local LLM needs the optional dependency: npm i node-llama-cpp");
+  }
+}
+var modelPromise = null;
+function loadLocalModel() {
+  if (!modelPromise) {
+    modelPromise = (async () => {
+      const spec = configuredModel();
+      let modelPath = spec;
+      if (spec.startsWith("hf:")) {
+        const { createModelDownloader } = await importLlama();
+        const downloader = await createModelDownloader({
+          modelUri: spec,
+          dirPath: join8(getCacheDir(), "models")
+        });
+        modelPath = await downloader.download();
+      } else if (!existsSync2(spec)) {
+        throw new Error(`Local model not found: ${spec}`);
+      }
+      const { getLlama } = await importLlama();
+      const llama = await getLlama();
+      return llama.loadModel({ modelPath });
+    })();
+    modelPromise.catch(() => {
+      modelPromise = null;
+    });
+  }
+  return modelPromise;
+}
+async function callLocalLlm(prompt, lang = "fr") {
+  const model = await loadLocalModel();
+  const { LlamaChatSession } = await importLlama();
+  const context = await model.createContext({ contextSize: LOCAL_CONTEXT_SIZE });
+  try {
+    const session = new LlamaChatSession({
+      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 });
+  } finally {
+    await context.dispose();
+  }
+}
+
 // src/cli.ts
 var ExitSignal = class {
   constructor(code) {
@@ -2555,6 +2617,8 @@ Usage:
   npx dsh-codebase-chat --project <path> --health --diff main
   npx dsh-codebase-chat --project <path> --watch
   npx dsh-codebase-chat --project <path> --prompt intelligence
+  npx dsh-codebase-chat --project <path> --prompt intelligence --call   # answered via DEEPSEEK_API_KEY
+  npx dsh-codebase-chat --project <path> --prompt intelligence --local  # answered 100% offline
 
 Options:
   -p, --project <path>   Project directory (default: current directory)
@@ -2572,6 +2636,11 @@ Options:
                          Pipe it to any LLM (e.g. ... --prompt intelligence | dsh).
                          Query modes read --ask/--search/--file; --focus and
                          --style (ouf|punchy|dense|pedagogique|minimal) apply.
+  --call                 With --prompt: send it to the API (needs DEEPSEEK_API_KEY
+                         or OPENAI_API_KEY; DEEPSEEK_BASE_URL / CODEBASE_MODEL
+                         customize endpoint/model) instead of printing it.
+  --local                With --prompt: answer with the embedded local model
+                         (node-llama-cpp, ~1 GB download on first use, offline).
   -w, --watch            Keep the index hot \u2014 rebuild incrementally on file changes
   -e, --embed            Enable local semantic embeddings (slower, more relevant)
   --lang <en|fr>         Language for headings (default: .codebase-chat.json lang, else fr)
@@ -2600,6 +2669,8 @@ async function main() {
       prompt: { type: "string" },
       focus: { type: "string" },
       style: { type: "string" },
+      call: { type: "boolean", default: false },
+      local: { type: "boolean", default: false },
       watch: { type: "boolean", short: "w", default: false },
       embed: { type: "boolean", short: "e", default: false },
       lang: { type: "string" },
@@ -2718,8 +2789,9 @@ ${formatHealthReport(report, lang)}`;
       }
     }
     const projectName = basename3(result.absProject);
+    let prompt;
     try {
-      console.log(buildToolPrompt(tool, {
+      prompt = buildToolPrompt(tool, {
         context: `${result.context}${staticSection}`,
         projectName,
         lang,
@@ -2728,11 +2800,48 @@ ${formatHealthReport(report, lang)}`;
         focus: values.focus ?? query,
         filePath: values.file ?? "",
         description: query
-      }));
+      });
     } catch {
       console.error(lang === "en" ? `unknown prompt mode "${mode}" \u2014 expected: intelligence, report, audit, tasks, ceo, player, chat, search, explain, refactor, crea` : `mode de prompt inconnu "${mode}" \u2014 attendu : intelligence, report, audit, tasks, ceo, player, chat, search, explain, refactor, crea`);
       exit(1);
     }
+    if (values.local || isLocalLlmEnabled()) {
+      console.error(lang === "en" ? "Answering with the embedded local model (first run downloads ~1 GB)..." : "R\xE9ponse via le mod\xE8le local embarqu\xE9 (premier lancement : ~1 Go de t\xE9l\xE9chargement)...");
+      console.log(await callLocalLlm(prompt, lang));
+      exit(0);
+    }
+    if (values.call) {
+      const apiKey = process.env.DEEPSEEK_API_KEY || process.env.OPENAI_API_KEY || "";
+      if (!apiKey) {
+        console.error(lang === "en" ? "--call needs DEEPSEEK_API_KEY or OPENAI_API_KEY in the environment" : "--call n\xE9cessite DEEPSEEK_API_KEY ou OPENAI_API_KEY dans l\u2019environnement");
+        exit(1);
+      }
+      const baseUrl = process.env.DEEPSEEK_BASE_URL || process.env.OPENAI_BASE_URL || "https://api.deepseek.com/v1";
+      const model = process.env.CODEBASE_MODEL || "deepseek-chat";
+      const res = await fetch(`${baseUrl}/chat/completions`, {
+        method: "POST",
+        signal: AbortSignal.timeout(12e4),
+        headers: { "Content-Type": "application/json", Authorization: `Bearer ${apiKey}` },
+        body: JSON.stringify({
+          model,
+          messages: [
+            { role: "system", content: lang === "en" ? "You are a senior codebase analyst. Be precise and cite files." : "Tu es un analyste codebase senior. Sois pr\xE9cis et cite les fichiers." },
+            { role: "user", content: prompt }
+          ],
+          temperature: 0.3,
+          max_tokens: 8192
+        })
+      });
+      if (!res.ok) {
+        const text = await res.text().catch(() => "");
+        console.error(`API error ${res.status}: ${text}`);
+        exit(1);
+      }
+      const data = await res.json();
+      console.log(data.choices?.[0]?.message?.content || "");
+      exit(0);
+    }
+    console.log(prompt);
     exit(0);
   }
   if (values.ask || values.search || values.file) {