okengine 0.9.0 → 0.10.0

package/src/cli/ai-setup/detect-ollama.ts

@@ -138,6 +138,13 @@ export async function detectOllama(
   };
 }
 
+/** Detected host hardware for the Ollama banner. */
+export type MachineInfo = {
+  readonly osName: string;
+  readonly cpuCount: number | null;
+  readonly ramGb: number | null;
+};
+
 /**
  * Best-effort total system RAM in GB.
  */
@@ -164,3 +171,43 @@ export function detectTotalRamGb(): number | null {
   }
   return null;
 }
+
+/**
+ * Best-effort OS name · CPU count · RAM for the Ollama banner.
+ */
+export function detectMachineInfo(): MachineInfo {
+  const osName =
+    process.platform === "darwin"
+      ? "macOS"
+      : process.platform === "linux"
+        ? "Linux"
+        : process.platform === "win32"
+          ? "Windows"
+          : process.platform;
+
+  let cpuCount: number | null = null;
+  try {
+    if (process.platform === "darwin") {
+      const proc = Bun.spawnSync(["sysctl", "-n", "hw.ncpu"], { stdout: "pipe" });
+      if (proc.exitCode === 0) {
+        const n = Number(proc.stdout.toString().trim());
+        if (Number.isFinite(n) && n > 0) cpuCount = n;
+      }
+    } else if (typeof navigator !== "undefined" && "hardwareConcurrency" in navigator) {
+      const n = Number(navigator.hardwareConcurrency);
+      if (Number.isFinite(n) && n > 0) cpuCount = n;
+    }
+  } catch {
+    // ignore
+  }
+  if (cpuCount === null) {
+    try {
+      const n = (process as { availableParallelism?: () => number }).availableParallelism?.();
+      if (typeof n === "number" && Number.isFinite(n) && n > 0) cpuCount = n;
+    } catch {
+      // ignore
+    }
+  }
+
+  return { osName, cpuCount, ramGb: detectTotalRamGb() };
+}

package/src/cli/ai-setup/index.ts

@@ -100,11 +100,12 @@ export function aiSetupHelp(): string {
 
 Usage:
   oke ai setup
+  oke ai setup --provider llama-cpp --yes
   oke ai setup --provider ollama --yes
   oke ai setup --provider anthropic --chat claude-sonnet-4-20250514 --yes
 
 Options:
-  --provider <id>   ollama | openai | anthropic | gemini | lmstudio | openrouter | custom
+  --provider <id>   llama-cpp | ollama | vllm | sglang | openai | anthropic | gemini | lmstudio | openrouter | custom
   --chat <model>    Chat model id
   --vision <model>  Vision model id (ollama)
   --embed <model>   Embedding model id (ollama)
@@ -151,16 +152,21 @@ export async function runAiSetup(args: AiSetupCliArgs): Promise<number> {
   if (input === null) return 1;
 
   if (args.pull && input.driver === "ollama" && (args.yes || !tty) && args.chat) {
-    // Non-interactive pull of requested models
+    // Non-interactive pull via the server HTTP API (never a host `ollama` CLI).
+    const { ensureOllamaModel } = await import("../../docker/ollama-pull.ts");
+    const baseUrl =
+      input.baseUrl?.trim() || process.env.OKE_AI_URL?.trim() || "http://127.0.0.1:11434";
     const toPull = [input.chatModel, input.visionModel, input.embedModel].filter(
       (m): m is string => typeof m === "string" && m.length > 0,
     );
     for (const id of toPull) {
-      const proc = Bun.spawn(["ollama", "pull", id], {
-        stdout: "inherit",
-        stderr: "inherit",
-      });
-      await proc.exited;
+      console.log(`oke ai setup: pulling ${id} via ${baseUrl}/api/pull…`);
+      try {
+        await ensureOllamaModel({ url: baseUrl, model: id });
+      } catch (err) {
+        const msg = err instanceof Error ? err.message : String(err);
+        console.error(`oke ai setup: pull ${id} failed (continuing) — ${msg}`);
+      }
     }
   }
 
@@ -192,6 +198,16 @@ export async function runAiSetup(args: AiSetupCliArgs): Promise<number> {
  */
 function nonInteractiveInput(args: AiSetupCliArgs): AiSetupApplyInput {
   const provider = args.provider!;
+  if (provider === "llama-cpp") {
+    return {
+      driver: "openai-compatible",
+      baseUrl: process.env.OKE_AI_URL ?? "http://127.0.0.1:8080/v1",
+      chatModel: args.chat ?? "smollm2",
+      visionModel: null,
+      embedModel: null,
+      image: "ghcr.io/ggml-org/llama.cpp:server-b10290",
+    };
+  }
   if (provider === "ollama") {
     return {
       driver: "ollama",
@@ -199,6 +215,27 @@ function nonInteractiveInput(args: AiSetupCliArgs): AiSetupApplyInput {
       chatModel: args.chat ?? "gemma4:e4b",
       visionModel: args.vision === undefined ? "qwen3-vl:4b" : args.vision || null,
       embedModel: args.embed ?? "nomic-embed-text",
+      image: "ollama/ollama:0.32.6",
+    };
+  }
+  if (provider === "vllm") {
+    return {
+      driver: "openai-compatible",
+      baseUrl: process.env.OKE_AI_URL ?? "http://127.0.0.1:8000/v1",
+      chatModel: args.chat ?? "Qwen/Qwen3-0.6B",
+      visionModel: null,
+      embedModel: null,
+      image: "vllm/vllm-openai:v0.26.0",
+    };
+  }
+  if (provider === "sglang") {
+    return {
+      driver: "openai-compatible",
+      baseUrl: process.env.OKE_AI_URL ?? "http://127.0.0.1:30000/v1",
+      chatModel: args.chat ?? "Qwen/Qwen3-0.6B",
+      visionModel: null,
+      embedModel: null,
+      image: "lmsysorg/sglang:v0.5.16-runtime",
     };
   }
   if (provider === "anthropic") {