okengine 0.18.4 → 0.19.0

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

@@ -1,213 +0,0 @@
-/**
- * Detect local Ollama — CLI (`ollama list` / `ps`) and HTTP `/api/tags`.
- */
-
-import { OLLAMA_DEFAULT_BASE_URL, normalizeOllamaBaseUrl } from "../../drivers/ai-ollama.ts";
-import { ALL_CURATED, type CatalogModel } from "./catalog.ts";
-
-/** Result of probing the local Ollama install. */
-export type OllamaDetectResult = {
-  readonly available: boolean;
-  readonly baseUrl: string;
-  readonly installed: readonly string[];
-  readonly running: readonly string[];
-  readonly curatedInstalled: readonly CatalogModel[];
-  readonly curatedMissing: readonly CatalogModel[];
-};
-
-/**
- * Parse `ollama list` tabular output into model names.
- *
- * @param stdout - CLI stdout
- */
-export function parseOllamaList(stdout: string): string[] {
-  const lines = stdout
-    .split("\n")
-    .map((l) => l.trim())
-    .filter(Boolean);
-  const out: string[] = [];
-  for (const line of lines) {
-    if (/^NAME\b/i.test(line)) continue;
-    const name = line.split(/\s+/)[0];
-    if (name) out.push(name);
-  }
-  return out;
-}
-
-/**
- * Parse `ollama ps` tabular output into running model names.
- *
- * @param stdout - CLI stdout
- */
-export function parseOllamaPs(stdout: string): string[] {
-  return parseOllamaList(stdout);
-}
-
-/**
- * Normalize installed tag for comparison (`qwen3.5:9b` vs `qwen3.5:9b-mlx`).
- *
- * @param id - Catalog or installed id
- * @param installed - Installed names
- */
-export function isInstalled(id: string, installed: readonly string[]): boolean {
-  const base = id.split(":")[0] ?? id;
-  return installed.some((name) => {
-    if (name === id) return true;
-    if (name.startsWith(`${id}-`)) return true;
-    if (name.startsWith(`${base}:`) && id.startsWith(`${base}:`)) {
-      // treat mlx / quant suffixes as matching the same family:tag when prefix matches
-      const instTag = name.slice(base.length + 1);
-      const wantTag = id.slice(base.length + 1);
-      return (
-        instTag === wantTag || instTag.startsWith(`${wantTag}-`) || wantTag.startsWith(instTag)
-      );
-    }
-    return false;
-  });
-}
-
-/**
- * Detect Ollama via CLI then HTTP fallback.
- *
- * @param options - Base URL / spawn / fetch seams
- */
-export async function detectOllama(
-  options: {
-    readonly baseUrl?: string;
-    readonly fetch?: typeof globalThis.fetch;
-    readonly run?: (cmd: readonly string[]) => Promise<{ code: number; stdout: string }>;
-  } = {},
-): Promise<OllamaDetectResult> {
-  const baseUrl = normalizeOllamaBaseUrl(options.baseUrl ?? OLLAMA_DEFAULT_BASE_URL);
-  const run =
-    options.run ??
-    (async (cmd) => {
-      const proc = Bun.spawn([...cmd], { stdout: "pipe", stderr: "pipe" });
-      const stdout = await new Response(proc.stdout).text();
-      const code = await proc.exited;
-      return { code, stdout };
-    });
-
-  let installed: string[] = [];
-  let running: string[] = [];
-  let available = false;
-
-  try {
-    const list = await run(["ollama", "list"]);
-    if (list.code === 0) {
-      available = true;
-      installed = parseOllamaList(list.stdout);
-    }
-    const ps = await run(["ollama", "ps"]);
-    if (ps.code === 0) {
-      running = parseOllamaPs(ps.stdout);
-    }
-  } catch {
-    // CLI missing — try HTTP
-  }
-
-  if (!available) {
-    const fetchFn = options.fetch ?? globalThis.fetch;
-    try {
-      const res = await fetchFn(`${baseUrl}/api/tags`, {
-        method: "GET",
-        signal: AbortSignal.timeout(2_000),
-      });
-      if (res.ok) {
-        available = true;
-        const json = (await res.json()) as { models?: readonly { name?: string }[] };
-        installed = (json.models ?? [])
-          .map((m) => m.name)
-          .filter((n): n is string => typeof n === "string");
-      }
-    } catch {
-      // unavailable
-    }
-  }
-
-  const curatedInstalled = ALL_CURATED.filter((m) => isInstalled(m.id, installed));
-  const curatedMissing = ALL_CURATED.filter((m) => !isInstalled(m.id, installed));
-
-  return {
-    available,
-    baseUrl,
-    installed,
-    running,
-    curatedInstalled,
-    curatedMissing,
-  };
-}
-
-/** 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.
- */
-export function detectTotalRamGb(): number | null {
-  try {
-    if (process.platform === "darwin") {
-      const proc = Bun.spawnSync(["sysctl", "-n", "hw.memsize"], { stdout: "pipe" });
-      if (proc.exitCode === 0) {
-        const bytes = Number(proc.stdout.toString().trim());
-        if (Number.isFinite(bytes) && bytes > 0) return Math.round(bytes / 1024 ** 3);
-      }
-    }
-    if (process.platform === "linux") {
-      const proc = Bun.spawnSync(["awk", "/MemTotal/ {print $2}", "/proc/meminfo"], {
-        stdout: "pipe",
-      });
-      if (proc.exitCode === 0) {
-        const kb = Number(proc.stdout.toString().trim());
-        if (Number.isFinite(kb) && kb > 0) return Math.round(kb / 1024 ** 2);
-      }
-    }
-  } catch {
-    // ignore
-  }
-  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/recommend.test.ts

@@ -1,225 +0,0 @@
-/**
- * Smart Ollama recommendation — RAM tiers (no needs quiz).
- */
-
-import { describe, expect, test } from "bun:test";
-import {
-  LLAMA_CPP_CHAT_MODELS,
-  llamaCppModelsForTier,
-  modelsForTier,
-  recommendChatModel,
-  recommendForTier,
-  recommendLlamaCppForTier,
-} from "./catalog.ts";
-import {
-  fittingChatModels,
-  formatLlamaCppBanner,
-  formatModelRow,
-  formatModelTableHeader,
-  formatOllamaBanner,
-  isTightFit,
-  modelFitsComfortably,
-  modelFitsOnMachine,
-  recommendChatForNeeds,
-  suggestTierForRam,
-  usableRamGb,
-} from "./recommend.ts";
-
-describe("modelFitsOnMachine", () => {
-  test("catalog ramGb is machine tier, not download size", () => {
-    expect(
-      modelFitsOnMachine(
-        {
-          id: "x",
-          label: "x",
-          hint: "",
-          role: "chat",
-          ramGb: 16,
-          tier: "balanced",
-          modalities: ["text"],
-        },
-        24,
-      ),
-    ).toBe(true);
-    expect(
-      modelFitsOnMachine(
-        {
-          id: "x",
-          label: "x",
-          hint: "",
-          role: "chat",
-          ramGb: 32,
-          tier: "smart",
-          modalities: ["text"],
-        },
-        24,
-      ),
-    ).toBe(false);
-  });
-});
-
-describe("tiers", () => {
-  test("each tier exposes up to 10 models", () => {
-    for (const tier of ["ultra-fast", "fast", "balanced", "smart"] as const) {
-      expect(modelsForTier(tier).length).toBeGreaterThan(0);
-      expect(modelsForTier(tier).length).toBeLessThanOrEqual(10);
-    }
-  });
-
-  test("suggestTierForRam maps fit budget", () => {
-    expect(suggestTierForRam(4)).toBe("ultra-fast");
-    expect(suggestTierForRam(8)).toBe("fast");
-    expect(suggestTierForRam(16)).toBe("balanced");
-    expect(suggestTierForRam(32)).toBe("smart");
-  });
-
-  test("recommendForTier returns a model in that tier", () => {
-    const pick = recommendForTier("fast", 16);
-    expect(pick.tier).toBe("fast");
-  });
-});
-
-describe("recommendChatForNeeds", () => {
-  test("8GB → ultra-fast or fast entry", () => {
-    const pick = recommendChatForNeeds(8);
-    expect(pick.ramGb).toBeLessThanOrEqual(8);
-  });
-
-  test("16GB → never recommends 32GB-class", () => {
-    const pick = recommendChatForNeeds(16);
-    expect(pick.ramGb).toBeLessThanOrEqual(16);
-    expect(pick.id).not.toBe("qwen3.5:27b");
-  });
-
-  test("24GB leaves headroom", () => {
-    const pick = recommendChatForNeeds(24);
-    expect(pick.ramGb).toBeLessThanOrEqual(usableRamGb(24));
-  });
-});
-
-describe("comfortable vs tight", () => {
-  test("16GB-class on 16GB machine is tight", () => {
-    const qwen = fittingChatModels(16).find((m) => m.id === "qwen3.5:9b");
-    expect(qwen).toBeDefined();
-    expect(modelFitsOnMachine(qwen!, 16)).toBe(true);
-    expect(modelFitsComfortably(qwen!, 16)).toBe(false);
-    expect(isTightFit(qwen!, 16)).toBe(true);
-  });
-
-  test("16GB-class on 24GB machine is comfortable", () => {
-    const qwen = fittingChatModels(24).find((m) => m.id === "qwen3.5:9b")!;
-    expect(modelFitsComfortably(qwen, 24)).toBe(true);
-    expect(isTightFit(qwen, 24)).toBe(false);
-  });
-});
-
-describe("recommendChatModel", () => {
-  test("respects tiers with headroom preference", () => {
-    expect(recommendChatModel(8).ramGb).toBeLessThanOrEqual(8);
-    expect(recommendChatModel(16).ramGb).toBeLessThanOrEqual(16);
-    expect(recommendChatModel(24).ramGb).toBeLessThanOrEqual(16);
-  });
-});
-
-describe("formatOllamaBanner", () => {
-  test("shows OS · CPU · RAM · Fit RAM · detected — no comfortable list", () => {
-    const text = formatOllamaBanner({ osName: "macOS", cpuCount: 10, ramGb: 24 }, ["gemma4:e4b"]);
-    expect(text).toContain("OS macOS  ·  CPU 10  ·  RAM ~24GB RAM");
-    expect(text).toContain("Fit RAM ~20GB RAM");
-    expect(text).toContain("Detected local models");
-    expect(text).toContain("gemma4:e4b");
-    expect(text).not.toContain("Fits comfortably");
-    expect(text).toContain("Tiers  ·  Ultra Fast");
-  });
-});
-
-describe("llama.cpp catalog", () => {
-  test("each tier has ≥10 curated Docker Hub ai/ models (UI shows up to 20)", () => {
-    for (const tier of ["ultra-fast", "fast", "balanced", "smart"] as const) {
-      const all = LLAMA_CPP_CHAT_MODELS.filter((m) => m.tier === tier);
-      expect(all.length).toBeGreaterThanOrEqual(10);
-      const list = llamaCppModelsForTier(tier);
-      expect(list.length).toBeGreaterThanOrEqual(10);
-      expect(list.length).toBeLessThanOrEqual(20);
-      expect(list.length).toBe(Math.min(all.length, 20));
-      expect(list.every((m) => !m.id.startsWith("ai/"))).toBe(true);
-      for (let i = 1; i < list.length; i++) {
-        expect(list[i]!.ramGb).toBeGreaterThanOrEqual(list[i - 1]!.ramGb);
-      }
-    }
-  });
-
-  test("recommendLlamaCppForTier returns a model in that tier", () => {
-    expect(recommendLlamaCppForTier("ultra-fast", 8).id).toBe("granite3.3:2b");
-    expect(recommendLlamaCppForTier("fast", 16).tier).toBe("fast");
-  });
-
-  test("visible list surfaces Gemma 4 and Qwen", () => {
-    const ids = (tier: "fast" | "balanced" | "smart") =>
-      llamaCppModelsForTier(tier).map((m) => m.id);
-    expect(ids("fast").some((id) => id.startsWith("gemma4:"))).toBe(true);
-    expect(ids("fast").some((id) => id.startsWith("qwen"))).toBe(true);
-    expect(ids("balanced")).toContain("gemma4:e4b");
-    expect(ids("balanced").some((id) => id.startsWith("qwen3"))).toBe(true);
-    expect(ids("smart")).toContain("gemma4:31b");
-    expect(ids("smart").some((id) => id.startsWith("qwen"))).toBe(true);
-  });
-
-  test("formatLlamaCppBanner shows Docker Hub ai/ source — not Ollama detect", () => {
-    const text = formatLlamaCppBanner({ osName: "macOS", cpuCount: 10, ramGb: 24 });
-    expect(text).toContain("OS macOS  ·  CPU 10  ·  RAM ~24GB RAM");
-    expect(text).toContain("Fit RAM ~20GB RAM");
-    expect(text).toContain("Docker Hub ai/");
-    expect(text).not.toContain("Detected local models");
-    expect(text).toContain("Tiers  ·  Ultra Fast");
-  });
-});
-
-describe("formatModelRow", () => {
-  test("aligned Model | RAM | Caps columns", () => {
-    const row = formatModelRow({
-      id: "x",
-      label: "Gemma 4 4B",
-      hint: "",
-      role: "chat",
-      ramGb: 8,
-      tier: "fast",
-      modalities: ["text", "code"],
-    });
-    const header = formatModelTableHeader();
-    expect(header.startsWith("Model")).toBe(true);
-    expect(header).toContain("RAM");
-    expect(header).toContain("Caps");
-    // Caps column starts at the same offset as the header.
-    expect(row.indexOf("text")).toBe(header.indexOf("Caps"));
-    expect(row).toContain("Gemma 4 4B");
-    expect(row).toContain("≈8GB");
-    expect(row).toContain("text · code");
-    expect(row).not.toContain("  ·  ≈");
-  });
-
-  test("Arabic and CJK labels keep Caps aligned", () => {
-    const header = formatModelTableHeader();
-    const capsWidth = Bun.stringWidth(header.slice(0, header.indexOf("Caps")));
-    const ar = formatModelRow({
-      id: "ar",
-      label: "مرحبا بالعالم",
-      hint: "",
-      role: "chat",
-      ramGb: 8,
-      tier: "fast",
-      modalities: ["text"],
-    });
-    const cjk = formatModelRow({
-      id: "cjk",
-      label: "你好世界模型",
-      hint: "",
-      role: "chat",
-      ramGb: 8,
-      tier: "fast",
-      modalities: ["text"],
-    });
-    expect(Bun.stringWidth(ar.slice(0, ar.indexOf("text")))).toBe(capsWidth);
-    expect(Bun.stringWidth(cjk.slice(0, cjk.indexOf("text")))).toBe(capsWidth);
-  });
-});

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

@@ -1,225 +0,0 @@
-/**
- * Machine fit helpers for local Ollama model picks.
- *
- * Catalog `ramGb` is the **recommended machine tier** for that model (not the
- * download size). Prefer picks that leave ~4GB for OS / IDE / browser.
- */
-
-import {
-  CHAT_MODELS,
-  MODEL_TIERS,
-  type CatalogModel,
-  type ModelTier,
-  recommendForTier,
-} from "./catalog.ts";
-
-/** Reserved for OS + IDE + browser while the model runs. */
-export const OS_HEADROOM_GB = 4;
-
-/**
- * RAM left after reserving OS/IDE headroom.
- *
- * @param totalRamGb - Detected machine RAM
- */
-export function usableRamGb(totalRamGb: number): number {
-  return Math.max(0, totalRamGb - OS_HEADROOM_GB);
-}
-
-/**
- * Whether the machine meets the model's recommended tier (`model.ramGb`).
- *
- * @param model - Catalog entry
- * @param totalRamGb - Machine RAM (null → only the catalog default recommended)
- */
-export function modelFitsOnMachine(model: CatalogModel, totalRamGb: number | null): boolean {
-  if (totalRamGb === null || !Number.isFinite(totalRamGb)) {
-    return Boolean(model.recommended);
-  }
-  return model.ramGb <= totalRamGb;
-}
-
-/**
- * Whether the model fits with comfortable headroom for OS/IDE.
- *
- * @param model - Catalog entry
- * @param totalRamGb - Machine RAM
- */
-export function modelFitsComfortably(model: CatalogModel, totalRamGb: number | null): boolean {
-  if (totalRamGb === null || !Number.isFinite(totalRamGb)) {
-    return Boolean(model.recommended);
-  }
-  const smallest = [...CHAT_MODELS].sort((a, b) => a.ramGb - b.ramGb)[0]!;
-  if (model.id === smallest.id) return model.ramGb <= totalRamGb;
-  return model.ramGb + OS_HEADROOM_GB <= totalRamGb;
-}
-
-/**
- * True when the model meets the tier but leaves little room for OS/IDE.
- *
- * @param model - Catalog entry
- * @param totalRamGb - Machine RAM
- */
-export function isTightFit(model: CatalogModel, totalRamGb: number | null): boolean {
-  if (totalRamGb === null || !Number.isFinite(totalRamGb)) return false;
-  return modelFitsOnMachine(model, totalRamGb) && !modelFitsComfortably(model, totalRamGb);
-}
-
-/**
- * Chat models that meet the machine tier, smallest → largest.
- *
- * @param totalRamGb - Machine RAM
- */
-export function fittingChatModels(totalRamGb: number | null): readonly CatalogModel[] {
-  const fits = CHAT_MODELS.filter((m) => modelFitsOnMachine(m, totalRamGb));
-  if (fits.length > 0) return [...fits].sort((a, b) => a.ramGb - b.ramGb);
-  return [[...CHAT_MODELS].sort((a, b) => a.ramGb - b.ramGb)[0]!];
-}
-
-/**
- * Prefer comfortable fits; fall back to any tier-fit if needed.
- *
- * @param totalRamGb - Machine RAM
- */
-export function comfortableChatModels(totalRamGb: number | null): readonly CatalogModel[] {
-  const comfortable = CHAT_MODELS.filter((m) => modelFitsComfortably(m, totalRamGb));
-  if (comfortable.length > 0) {
-    return [...comfortable].sort((a, b) => a.ramGb - b.ramGb);
-  }
-  return fittingChatModels(totalRamGb);
-}
-
-/**
- * Suggest a tier for the machine's fit RAM budget.
- *
- * @param totalRamGb - Machine RAM
- */
-export function suggestTierForRam(totalRamGb: number | null): ModelTier {
-  if (totalRamGb === null || !Number.isFinite(totalRamGb)) return "fast";
-  const budget = usableRamGb(totalRamGb);
-  if (budget >= 24) return "smart";
-  if (budget >= 8) return "balanced";
-  if (budget >= 4) return "fast";
-  return "ultra-fast";
-}
-
-/**
- * Recommend a chat model for host RAM (balanced default — no quiz).
- *
- * @param totalRamGb - Machine RAM
- */
-export function recommendChatForNeeds(totalRamGb: number | null): CatalogModel {
-  return recommendForTier(suggestTierForRam(totalRamGb), totalRamGb);
-}
-
-/**
- * Shared hardware lines for local AI banners.
- *
- * @param machine - Detected hardware
- */
-function formatMachineBannerLines(machine: {
-  readonly osName: string;
-  readonly cpuCount: number | null;
-  readonly ramGb: number | null;
-}): readonly string[] {
-  const ram =
-    machine.ramGb !== null && Number.isFinite(machine.ramGb)
-      ? `~${machine.ramGb}GB RAM`
-      : "RAM unknown";
-  const cpu =
-    machine.cpuCount !== null && Number.isFinite(machine.cpuCount) ? String(machine.cpuCount) : "?";
-  const fit =
-    machine.ramGb !== null && Number.isFinite(machine.ramGb)
-      ? `~${usableRamGb(machine.ramGb)}GB RAM`
-      : "unknown";
-  const sep = "  ·  ";
-  return [[`OS ${machine.osName}`, `CPU ${cpu}`, `RAM ${ram}`].join(sep), `Fit RAM ${fit}`];
-}
-
-/**
- * Ollama banner lines — OS · CPU · RAM · fit · detected · tiers.
- *
- * @param machine - Detected hardware
- * @param detectedIds - Installed model ids
- */
-export function formatOllamaBanner(
-  machine: {
-    readonly osName: string;
-    readonly cpuCount: number | null;
-    readonly ramGb: number | null;
-  },
-  detectedIds: readonly string[],
-): string {
-  const sep = "  ·  ";
-  const lines = [
-    ...formatMachineBannerLines(machine),
-    "",
-    "Detected local models",
-    ...(detectedIds.length > 0 ? detectedIds.map((id) => id) : ["(none detected)"]),
-    "",
-    `Tiers${sep}${MODEL_TIERS.map((t) => t.label).join(sep)}`,
-  ];
-  return lines.join("\n");
-}
-
-/**
- * llama.cpp banner — OS · CPU · RAM · fit · Docker Hub `ai/` source · tiers.
- *
- * @param machine - Detected hardware
- */
-export function formatLlamaCppBanner(machine: {
-  readonly osName: string;
-  readonly cpuCount: number | null;
-  readonly ramGb: number | null;
-}): string {
-  const sep = "  ·  ";
-  const lines = [
-    ...formatMachineBannerLines(machine),
-    "",
-    "Model source",
-    "Docker Hub ai/ (curated) — pulled on first container start",
-    "",
-    `Tiers${sep}${MODEL_TIERS.map((t) => t.label).join(sep)}`,
-  ];
-  return lines.join("\n");
-}
-
-/** Model name column width (monospace CLI table). */
-const MODEL_COL_NAME = 26;
-/** RAM column width, right-aligned (`≈32GB`). */
-const MODEL_COL_RAM = 7;
-
-/**
- * Clip or pad a cell for aligned CLI columns.
- *
- * @param value - Cell text
- * @param width - Fixed width
- */
-function clipPad(value: string, width: number): string {
-  const bunAnsi = Bun as typeof Bun & {
-    stringWidth(text: string): number;
-    sliceAnsi(text: string, start: number, end: number, omission?: string): string;
-  };
-  const display = bunAnsi.stringWidth(value);
-  if (display === width) return value;
-  if (display < width) return value + " ".repeat(width - display);
-  return bunAnsi.sliceAnsi(value, 0, width, "…");
-}
-
-/**
- * Column header for the model pick list (aligns with {@link formatModelRow}).
- */
-export function formatModelTableHeader(): string {
-  return `${clipPad("Model", MODEL_COL_NAME)}  ${"RAM".padStart(MODEL_COL_RAM)}  Caps`;
-}
-
-/**
- * One model row as aligned columns: Model | RAM | Caps.
- *
- * @param model - Catalog entry
- */
-export function formatModelRow(model: CatalogModel): string {
-  const name = clipPad(model.label, MODEL_COL_NAME);
-  const ram = `≈${model.ramGb}GB`.padStart(MODEL_COL_RAM);
-  const caps = model.modalities.join(" · ");
-  return `${name}  ${ram}  ${caps}`;
-}