open-agents-ai 0.15.1 → 0.15.3

package/dist/index.js

@@ -8870,13 +8870,6 @@ ${newerSummary}` : newerSummary;
               acc.id = chunk.toolCallId;
             if (chunk.toolCallArgs) {
               acc.args += chunk.toolCallArgs;
-              this.emit({
-                type: "stream_token",
-                content: chunk.toolCallArgs,
-                streamKind: "tool_args",
-                turn,
-                timestamp: (/* @__PURE__ */ new Date()).toISOString()
-              });
             }
           }
         }
@@ -10552,844 +10545,939 @@ var init_oa_directory = __esm({
   }
 });
 
-// packages/cli/dist/tui/commands.js
-async function handleSlashCommand(input, ctx) {
-  const trimmed = input.trim();
-  if (!trimmed.startsWith("/"))
-    return "not_a_command";
-  const [cmd, ...rest] = trimmed.slice(1).split(/\s+/);
-  const hasLocal = rest.includes("--local");
-  const filteredRest = rest.filter((r) => r !== "--local");
-  const arg = filteredRest.join(" ").trim();
-  switch (cmd) {
-    case "help":
-    case "h":
-    case "?":
-      renderSlashHelp();
-      return "handled";
-    case "quit":
-    case "exit":
-    case "q":
-      return "exit";
-    case "clear":
-    case "cls":
-      ctx.clearScreen();
-      return "handled";
-    case "verbose":
-    case "v":
-      ctx.setVerbose(!ctx.config.verbose);
-      if (hasLocal) {
-        ctx.saveLocalSettings({ verbose: ctx.config.verbose });
-        renderInfo(`Verbose mode: ${ctx.config.verbose ? "on" : "off"} (project-local)`);
-      } else {
-        ctx.saveSettings({ verbose: ctx.config.verbose });
-        renderInfo(`Verbose mode: ${ctx.config.verbose ? "on" : "off"}`);
+// packages/cli/dist/tui/setup.js
+import * as readline from "node:readline";
+import { execSync as execSync10 } from "node:child_process";
+import { existsSync as existsSync13, writeFileSync as writeFileSync7, mkdirSync as mkdirSync7 } from "node:fs";
+import { join as join18 } from "node:path";
+import { homedir as homedir8 } from "node:os";
+function detectSystemSpecs() {
+  let totalRamGB = 0;
+  let availableRamGB = 0;
+  let gpuVramGB = 0;
+  let gpuName = "";
+  try {
+    const memInfo = execSync10("free -b 2>/dev/null || sysctl -n hw.memsize 2>/dev/null", {
+      encoding: "utf8",
+      timeout: 5e3
+    });
+    if (memInfo.includes("Mem:")) {
+      const match = memInfo.match(/^Mem:\s+(\d+)\s+\d+\s+\d+\s+\d+\s+\d+\s+(\d+)/m);
+      if (match) {
+        totalRamGB = parseInt(match[1], 10) / 1024 ** 3;
+        availableRamGB = parseInt(match[2], 10) / 1024 ** 3;
       }
-      return "handled";
-    case "config":
-    case "cfg":
-      renderConfig({
-        model: ctx.config.model,
-        backendType: ctx.config.backendType,
-        backendUrl: ctx.config.backendUrl,
-        timeoutMs: String(ctx.config.timeoutMs),
-        maxRetries: String(ctx.config.maxRetries),
-        verbose: String(ctx.config.verbose),
-        dryRun: String(ctx.config.dryRun)
-      });
-      return "handled";
-    case "model":
-      if (arg) {
-        await switchModel(arg, ctx, hasLocal);
-      } else {
-        await showModelPicker(ctx);
+    } else {
+      const bytes = parseInt(memInfo.trim(), 10);
+      if (!isNaN(bytes)) {
+        totalRamGB = bytes / 1024 ** 3;
+        availableRamGB = totalRamGB * 0.7;
       }
-      return "handled";
-    case "models":
-      await listModels(ctx);
-      return "handled";
-    case "endpoint":
-    case "ep":
-      await handleEndpoint(arg, ctx, hasLocal);
-      return "handled";
-    case "update":
-    case "upgrade":
-      await handleUpdate(arg, ctx.repoRoot);
-      return "handled";
-    case "voice": {
-      const save = hasLocal ? ctx.saveLocalSettings.bind(ctx) : ctx.saveSettings.bind(ctx);
-      if (arg) {
-        const msg = await ctx.voiceSetModel(arg);
-        save({ voice: true, voiceModel: arg });
-        renderInfo(msg + (hasLocal ? " (project-local)" : ""));
-      } else {
-        const msg = await ctx.voiceToggle();
-        const isOn = msg.toLowerCase().includes("enabled") || msg.toLowerCase().includes("on");
-        save({ voice: isOn });
-        renderInfo(msg + (hasLocal ? " (project-local)" : ""));
+    }
+  } catch {
+  }
+  try {
+    const nvidiaSmi = execSync10("nvidia-smi --query-gpu=memory.total,name --format=csv,noheader,nounits 2>/dev/null", { encoding: "utf8", timeout: 5e3 });
+    const lines = nvidiaSmi.trim().split("\n");
+    if (lines.length > 0) {
+      for (const line of lines) {
+        const parts = line.split(",").map((s) => s.trim());
+        const vramMB = parseInt(parts[0] ?? "0", 10);
+        if (!isNaN(vramMB))
+          gpuVramGB += vramMB / 1024;
+        if (!gpuName && parts[1])
+          gpuName = parts[1];
       }
-      return "handled";
     }
-    case "stream": {
-      const isOn = ctx.streamToggle();
-      const save = hasLocal ? ctx.saveLocalSettings.bind(ctx) : ctx.saveSettings.bind(ctx);
-      save({ stream: isOn });
-      renderInfo(`Token streaming: ${isOn ? "on" : "off"}${hasLocal ? " (project-local)" : ""}` + (isOn ? " \u2014 thinking tokens in grey italics, responses with pastel syntax highlighting" : ""));
-      return "handled";
+  } catch {
+  }
+  return {
+    totalRamGB: Math.round(totalRamGB * 10) / 10,
+    availableRamGB: Math.round(availableRamGB * 10) / 10,
+    gpuVramGB: Math.round(gpuVramGB * 10) / 10,
+    gpuName
+  };
+}
+function recommendModel(specs) {
+  const effectiveGB = Math.max(specs.gpuVramGB, specs.availableRamGB);
+  const budget = effectiveGB * 0.8;
+  const localVariants = QWEN_VARIANTS.filter((v) => !v.cloud);
+  for (let i = localVariants.length - 1; i >= 0; i--) {
+    if (localVariants[i].sizeGB <= budget) {
+      return localVariants[i];
     }
-    case "tools": {
-      const tools = listCustomToolFiles(ctx.repoRoot);
-      if (tools.length === 0) {
-        renderInfo("No custom tools installed.");
-        renderInfo("The agent will automatically create tools when it detects repeated workflows (3+ times).");
-        renderInfo('Or ask the agent: "create a tool for [workflow]"');
-      } else {
-        process.stdout.write(`
-  ${c2.bold("Custom Tools:")}
+  }
+  return QWEN_VARIANTS.find((v) => v.tag === "qwen3.5:cloud");
+}
+function calculateContextWindow(specs, modelSizeGB2) {
+  const totalAvail = Math.max(specs.gpuVramGB, specs.totalRamGB);
+  const remaining = totalAvail - modelSizeGB2;
+  if (remaining >= 200)
+    return { numCtx: 131072, label: "128K" };
+  if (remaining >= 100)
+    return { numCtx: 65536, label: "64K" };
+  if (remaining >= 50)
+    return { numCtx: 32768, label: "32K" };
+  if (remaining >= 20)
+    return { numCtx: 16384, label: "16K" };
+  if (remaining >= 8)
+    return { numCtx: 8192, label: "8K" };
+  return { numCtx: 4096, label: "4K" };
+}
+function modelSupportsToolCalling(modelName) {
+  const lower = modelName.toLowerCase();
+  for (const known of TOOL_CALLING_MODELS) {
+    if (lower.startsWith(known) || lower.includes(known))
+      return true;
+  }
+  return false;
+}
+function ask(rl, question) {
+  return new Promise((resolve16) => {
+    rl.question(question, (answer) => resolve16(answer.trim()));
+  });
+}
+function pullModelWithAutoUpdate(tag) {
+  try {
+    execSync10(`ollama pull ${tag}`, {
+      stdio: "inherit",
+      timeout: 36e5
+      // 1 hour max
+    });
+  } catch (err) {
+    const errMsg = err instanceof Error ? err.message : String(err);
+    const stderr = err?.stderr?.toString?.() ?? errMsg;
+    const combined = errMsg + "\n" + stderr;
+    if (combined.includes("412") || combined.includes("newer version") || combined.includes("requires a newer version")) {
+      process.stdout.write(`
+  ${c2.yellow("\u26A0")} Ollama needs to be updated for this model.
+`);
+      process.stdout.write(`  ${c2.cyan("\u25CF")} Updating Ollama via official install script...
 
 `);
-        for (const t of tools) {
-          process.stdout.write(`  ${c2.cyan(t.name.padEnd(28))} ${c2.dim(`(${t.scope}, v${t.version}, ${t.stepsCount} steps)`)}
+      try {
+        execSync10("curl -fsSL https://ollama.com/install.sh | sh", {
+          stdio: "inherit",
+          timeout: 3e5
+          // 5 min max for install
+        });
+        process.stdout.write(`
+  ${c2.green("\u2714")} Ollama updated successfully.
 `);
-          process.stdout.write(`  ${"".padEnd(28)} ${t.description}
+        process.stdout.write(`  ${c2.cyan("\u25CF")} Retrying pull of ${c2.bold(tag)}...
+
 `);
-        }
-        process.stdout.write("\n");
+        execSync10(`ollama pull ${tag}`, {
+          stdio: "inherit",
+          timeout: 36e5
+        });
+      } catch (updateErr) {
+        const updateMsg = updateErr instanceof Error ? updateErr.message : String(updateErr);
+        throw new Error(`Failed to update Ollama and retry pull: ${updateMsg}
+Try manually:
+  curl -fsSL https://ollama.com/install.sh | sh
+  ollama pull ${tag}`);
       }
-      return "handled";
+    } else {
+      throw err;
     }
-    case "skills":
-    case "skill": {
-      const skills = discoverSkills(ctx.repoRoot);
-      if (skills.length === 0) {
-        renderInfo("No skills found.");
-        renderInfo("Install AIWG to get skills: npm i -g aiwg && aiwg use sdlc");
-        renderInfo("Or add skills manually to .oa/skills/{name}/SKILL.md");
-      } else {
-        let filtered = skills;
-        if (arg) {
-          const q = arg.toLowerCase();
-          filtered = skills.filter((s) => s.name.toLowerCase().includes(q) || s.description.toLowerCase().includes(q) || s.triggers.some((t) => t.toLowerCase().includes(q)));
-        }
-        if (filtered.length === 0) {
-          renderWarning(`No skills matching "${arg}". Showing all ${skills.length} skills:`);
-          filtered = skills;
-        }
-        const bySource = /* @__PURE__ */ new Map();
-        for (const s of filtered) {
-          const group = bySource.get(s.source) ?? [];
-          group.push(s);
-          bySource.set(s.source, group);
-        }
-        process.stdout.write(`
-  ${c2.bold(`Available Skills (${filtered.length}):`)}
-`);
-        for (const [source, group] of bySource) {
-          process.stdout.write(`
-  ${c2.dim(`\u2500\u2500 ${source} (${group.length}) \u2500\u2500`)}
-`);
-          for (const s of group) {
-            process.stdout.write(`  ${c2.cyan(s.name.padEnd(32))} ${s.description.slice(0, 60)}
-`);
-            if (s.triggers.length > 0) {
-              process.stdout.write(`  ${"".padEnd(32)} ${c2.dim(`triggers: ${s.triggers.slice(0, 3).join(" | ")}`)}
-`);
-            }
-          }
-        }
-        process.stdout.write("\n");
-        renderInfo('Invoke directly: /<skill-name> [args]  (e.g. /ralph "fix tests" --completion "npm test passes")');
-        renderInfo("Filter with: /skills <keyword>");
-      }
-      return "handled";
-    }
-    case "dream": {
-      if (arg === "stop" || arg === "wake") {
-        if (ctx.isDreaming?.()) {
-          ctx.dreamStop?.();
-          renderInfo("Waking up from dream mode...");
-        } else {
-          renderWarning("Not currently dreaming.");
-        }
-      } else if (ctx.isDreaming?.()) {
-        renderWarning("Already dreaming. Use /dream stop to wake up first.");
-      } else {
-        const mode = arg === "lucid" ? "lucid" : arg === "deep" ? "deep" : "default";
-        ctx.dreamStart?.(mode);
-      }
-      return "handled";
-    }
-    case "listen":
-    case "mic": {
-      if (!ctx.listenToggle) {
-        renderWarning("Listen mode not available in this context.");
-        return "handled";
-      }
-      if (arg === "stop" || arg === "off") {
-        const msg2 = await (ctx.listenStop?.() ?? Promise.resolve("Not listening."));
-        renderInfo(msg2);
-        return "handled";
-      }
-      if (arg === "confirm") {
-        const msg2 = ctx.listenSetMode?.("confirm") ?? "Confirm mode set.";
-        renderInfo(msg2);
-        return "handled";
-      }
-      if (arg === "auto") {
-        const msg2 = ctx.listenSetMode?.("auto") ?? "Auto mode set.";
-        renderInfo(msg2);
-        return "handled";
-      }
-      const modelSizes = ["tiny", "base", "small", "medium", "large", "large-v3"];
-      if (arg && modelSizes.includes(arg.toLowerCase())) {
-        const model = arg.toLowerCase() === "large" ? "large-v3" : arg.toLowerCase();
-        const msg2 = await (ctx.listenSetModel?.(model) ?? Promise.resolve(`Model set to ${model}.`));
-        renderInfo(msg2);
-        return "handled";
-      }
-      const msg = await ctx.listenToggle();
-      renderInfo(msg);
-      return "handled";
-    }
-    case "bruteforce":
-    case "brute": {
-      const isOn = ctx.bruteForceToggle();
-      const save = hasLocal ? ctx.saveLocalSettings.bind(ctx) : ctx.saveSettings.bind(ctx);
-      save({ bruteforce: isOn });
-      renderInfo(`Brute-force mode: ${isOn ? "on" : "off"}${hasLocal ? " (project-local)" : ""}` + (isOn ? " \u2014 agent will auto re-engage when turn limit is hit, reassess and try creative strategies" : ""));
-      return "handled";
-    }
-    case "emojis":
-    case "emoji": {
-      const current = ctx.getEmojis?.() ?? true;
-      const next = !current;
-      ctx.setEmojis?.(next);
-      const save = hasLocal ? ctx.saveLocalSettings.bind(ctx) : ctx.saveSettings.bind(ctx);
-      save({ emojis: next });
-      renderInfo(`Emojis ${next ? "enabled" : "disabled"}.`);
-      return "handled";
-    }
-    case "colors":
-    case "color": {
-      const current = ctx.getColors?.() ?? true;
-      const next = !current;
-      ctx.setColors?.(next);
-      const save = hasLocal ? ctx.saveLocalSettings.bind(ctx) : ctx.saveSettings.bind(ctx);
-      save({ colors: next });
-      renderInfo(`Colors ${next ? "enabled" : "disabled"}.`);
-      return "handled";
-    }
-    default: {
-      const skills = discoverSkills(ctx.repoRoot);
-      const skill = skills.find((s) => s.name === cmd || s.name === cmd.replace(/_/g, "-"));
-      if (skill) {
-        const content = loadSkillContent(skill.filePath);
-        if (content) {
-          renderInfo(`Loading skill: ${c2.bold(skill.name)} (${skill.source})`);
-          return { type: "skill", name: skill.name, content, args: arg };
-        }
-      }
-      renderWarning(`Unknown command: /${cmd}. Type /help for available commands.`);
-      return "handled";
-    }
-  }
-}
-async function listModels(ctx) {
-  try {
-    const models = await fetchOllamaModels(ctx.config.backendUrl);
-    renderModelList(models.map((m) => ({ name: m.name, size: m.size, modified: m.modified })), ctx.config.model);
-  } catch (err) {
-    renderError(`Failed to fetch models: ${err instanceof Error ? err.message : String(err)}`);
   }
 }
-async function showModelPicker(ctx) {
+async function runSetupWizard(config) {
+  const rl = readline.createInterface({
+    input: process.stdin,
+    output: process.stdout,
+    terminal: true
+  });
   try {
-    const models = await fetchOllamaModels(ctx.config.backendUrl);
-    if (models.length === 0) {
-      renderWarning("No models found. Pull a model with: ollama pull <model>");
-      return;
-    }
-    renderModelList(models.map((m) => ({ name: m.name, size: m.size, modified: m.modified })), ctx.config.model);
-  } catch (err) {
-    renderError(`Failed to fetch models: ${err instanceof Error ? err.message : String(err)}`);
+    return await doSetup(config, rl);
+  } finally {
+    rl.close();
   }
 }
-async function handleEndpoint(arg, ctx, local = false) {
-  if (!arg) {
-    process.stdout.write(`
-  ${c2.bold("Current endpoint:")}
-
-`);
-    process.stdout.write(`  ${c2.cyan("URL".padEnd(12))} ${ctx.config.backendUrl}
+async function doSetup(config, rl) {
+  process.stdout.write(`
+  ${c2.bold(c2.cyan("open-agents"))}
 `);
-    process.stdout.write(`  ${c2.cyan("Type".padEnd(12))} ${ctx.config.backendType}
+  process.stdout.write(`  ${c2.dim("\u2500".repeat(60))}
 `);
-    process.stdout.write(`  ${c2.cyan("Auth".padEnd(12))} ${ctx.config.apiKey ? "Bearer token set" : "none"}
+  process.stdout.write(`  ${c2.bold("First-run setup")}
+
 `);
-    process.stdout.write(`
-  ${c2.dim("Usage: /endpoint <url> [--auth <token>]")}
+  process.stdout.write(`  ${c2.cyan("\u25CF")} Detecting system specs...
 `);
-    process.stdout.write(`  ${c2.dim("  /endpoint http://localhost:11434         (Ollama, no auth)")}
+  const specs = detectSystemSpecs();
+  process.stdout.write(`  ${c2.dim("  RAM:")} ${specs.totalRamGB.toFixed(1)} GB total, ${specs.availableRamGB.toFixed(1)} GB available
 `);
-    process.stdout.write(`  ${c2.dim("  /endpoint http://remote:8000/v1 --auth sk-...  (OpenAI-compatible)")}
+  if (specs.gpuVramGB > 0) {
+    process.stdout.write(`  ${c2.dim("  GPU:")} ${specs.gpuName || "NVIDIA"} \u2014 ${specs.gpuVramGB.toFixed(1)} GB VRAM
 `);
-    process.stdout.write(`  ${c2.dim("  /endpoint http://remote:8000/v1          (OpenAI-compatible, no auth)")}
-
+  } else {
+    process.stdout.write(`  ${c2.dim("  GPU:")} No NVIDIA GPU detected (CPU inference)
 `);
-    return;
-  }
-  const parts = arg.split(/\s+/);
-  const url = parts[0];
-  let apiKey;
-  const authIdx = parts.indexOf("--auth");
-  if (authIdx !== -1 && parts[authIdx + 1]) {
-    apiKey = parts[authIdx + 1];
   }
+  process.stdout.write("\n");
+  let models = [];
   try {
-    new URL(url);
+    models = await fetchOllamaModels(config.backendUrl);
   } catch {
-    renderError(`Invalid URL: "${url}"`);
-    return;
-  }
-  let backendType = "ollama";
-  if (url.includes("/v1") || url.includes(":8000") || apiKey) {
-    backendType = "vllm";
+    renderError(`Cannot reach Ollama at ${config.backendUrl}`);
+    renderInfo("Start Ollama with: ollama serve");
+    renderInfo("Or use /endpoint to configure a remote backend after startup.");
+    const answer = await ask(rl, `
+  ${c2.bold("Continue without Ollama?")} (y/n) `);
+    if (answer.toLowerCase() !== "y")
+      return null;
+    return config.model;
   }
-  process.stdout.write(`
-  ${c2.dim("Testing connection...")} `);
-  try {
-    const healthUrl = backendType === "ollama" ? `${url.replace(/\/$/, "")}/api/tags` : `${url.replace(/\/$/, "")}/models`;
-    const headers = {};
-    if (apiKey)
-      headers["Authorization"] = `Bearer ${apiKey}`;
-    const resp = await fetch(healthUrl, {
-      headers,
-      signal: AbortSignal.timeout(1e4)
-    });
-    if (!resp.ok)
-      throw new Error(`HTTP ${resp.status}`);
-    process.stdout.write(`${c2.green("\u2714")} Connected
-`);
-  } catch (err) {
-    process.stdout.write(`${c2.yellow("\u26A0")} Could not verify
+  const currentModel = findModel(models, config.model);
+  if (currentModel) {
+    process.stdout.write(`  ${c2.green("\u2714")} Model ${c2.bold(currentModel.name)} is available.
+
 `);
-    renderWarning(`Endpoint may not be reachable: ${err instanceof Error ? err.message : String(err)}`);
-    renderInfo("Setting endpoint anyway \u2014 it may come online later.");
-  }
-  ctx.setEndpoint(url, backendType, apiKey);
-  const endpointSettings = { backendUrl: url, backendType, ...apiKey ? { apiKey } : {} };
-  if (local) {
-    ctx.saveLocalSettings(endpointSettings);
-  } else {
-    setConfigValue("backendUrl", url);
-    setConfigValue("backendType", backendType);
-    if (apiKey) {
-      setConfigValue("apiKey", apiKey);
-    }
-    ctx.saveSettings(endpointSettings);
+    return currentModel.name;
   }
-  process.stdout.write(`
-  ${c2.green("\u2714")} Endpoint updated and saved${local ? " (project-local)" : ""}:
+  process.stdout.write(`  ${c2.yellow("\u26A0")} Default model ${c2.bold(config.model)} is not available.
+
 `);
-  process.stdout.write(`    ${c2.cyan("URL".padEnd(8))} ${url}
+  const toolCallingModels = models.filter((m) => modelSupportsToolCalling(m.name));
+  if (toolCallingModels.length > 0) {
+    process.stdout.write(`  ${c2.cyan("\u25CF")} Found ${toolCallingModels.length} model(s) with tool-calling support:
+
 `);
-  process.stdout.write(`    ${c2.cyan("Type".padEnd(8))} ${backendType}
+    for (let i = 0; i < Math.min(toolCallingModels.length, 10); i++) {
+      const m = toolCallingModels[i];
+      process.stdout.write(`    ${c2.bold(String(i + 1))}. ${m.name} ${c2.dim(`(${m.size})`)}
 `);
-  if (apiKey) {
-    process.stdout.write(`    ${c2.cyan("Auth".padEnd(8))} Bearer ${apiKey.slice(0, 8)}...
+    }
+    process.stdout.write(`
+    ${c2.dim("0")}. Pull a new qwen3.5 model instead
+`);
+    process.stdout.write("\n");
+    const choice = await ask(rl, `  ${c2.bold("Select a model")} (1-${Math.min(toolCallingModels.length, 10)}, or 0 to pull new): `);
+    const idx = parseInt(choice, 10);
+    if (idx > 0 && idx <= toolCallingModels.length) {
+      const selected = toolCallingModels[idx - 1];
+      setConfigValue("model", selected.name);
+      process.stdout.write(`
+  ${c2.green("\u2714")} Selected ${c2.bold(selected.name)}. Saved to config.
+
 `);
+      return selected.name;
+    }
   } else {
-    process.stdout.write(`    ${c2.cyan("Auth".padEnd(8))} none
+    process.stdout.write(`  ${c2.yellow("\u26A0")} No tool-calling capable models found on this system.
+
 `);
   }
-  process.stdout.write("\n");
-}
-async function handleUpdate(subcommand, repoRoot) {
-  if (subcommand === "auto") {
-    const settings = { updateMode: "auto" };
-    saveProjectSettings(repoRoot, settings);
-    saveGlobalSettings(settings);
-    renderInfo("Update mode: auto \u2014 updates will install automatically after task completion.");
-    return;
+  const recommended = recommendModel(specs);
+  process.stdout.write(`  ${c2.cyan("\u25CF")} Recommended model based on your system:
+
+`);
+  const localVariants = QWEN_VARIANTS.filter((v) => !v.cloud);
+  for (let i = 0; i < localVariants.length; i++) {
+    const v = localVariants[i];
+    const fits = v.sizeGB <= Math.max(specs.gpuVramGB, specs.availableRamGB) * 0.8;
+    const isRec = v.tag === recommended.tag;
+    const marker = isRec ? c2.green("\u2192") : fits ? c2.dim(" ") : c2.red("\u2716");
+    const name = isRec ? c2.bold(c2.green(v.tag)) : fits ? v.tag : c2.dim(v.tag);
+    const label = isRec ? c2.bold(v.label) : c2.dim(v.label);
+    const tooLarge = !fits && !v.cloud ? c2.red(" (exceeds available memory)") : "";
+    process.stdout.write(`    ${marker} ${String(i + 1).padStart(2)}. ${name.padEnd(isRec ? 45 : 25)} ${label}${tooLarge}
+`);
   }
-  if (subcommand === "manual") {
-    const settings = { updateMode: "manual" };
-    saveProjectSettings(repoRoot, settings);
-    saveGlobalSettings(settings);
-    renderInfo("Update mode: manual \u2014 updates only install when you run /update.");
-    return;
+  process.stdout.write(`
+    ${c2.dim(" ")} ${String(localVariants.length + 1).padStart(2)}. ${c2.dim("qwen3.5:cloud")}              ${c2.dim("Ollama Cloud")}
+`);
+  process.stdout.write(`    ${c2.dim(" ")} ${String(localVariants.length + 2).padStart(2)}. ${c2.dim("qwen3.5:397b-cloud")}         ${c2.dim("397B Ollama Cloud")}
+`);
+  process.stdout.write("\n");
+  const pullChoice = await ask(rl, `  ${c2.bold("Select a model to pull")} (1-${localVariants.length + 2}, or Enter for recommended): `);
+  const pullIdx = pullChoice ? parseInt(pullChoice, 10) : 0;
+  let selectedVariant;
+  if (pullIdx === 0 || isNaN(pullIdx)) {
+    selectedVariant = recommended;
+  } else if (pullIdx <= localVariants.length) {
+    selectedVariant = localVariants[pullIdx - 1];
+  } else if (pullIdx === localVariants.length + 1) {
+    selectedVariant = QWEN_VARIANTS.find((v) => v.tag === "qwen3.5:cloud");
+  } else {
+    selectedVariant = QWEN_VARIANTS.find((v) => v.tag === "qwen3.5:397b-cloud");
   }
-  let currentVersion = "0.0.0";
-  try {
-    const { createRequire: createRequire4 } = await import("node:module");
-    const { fileURLToPath: fileURLToPath3 } = await import("node:url");
-    const { dirname: dirname5, join: join28 } = await import("node:path");
-    const { existsSync: existsSync19 } = await import("node:fs");
-    const req = createRequire4(import.meta.url);
-    const thisDir = dirname5(fileURLToPath3(import.meta.url));
-    const candidates = [
-      join28(thisDir, "..", "package.json"),
-      join28(thisDir, "..", "..", "package.json"),
-      join28(thisDir, "..", "..", "..", "package.json")
-    ];
-    for (const pkgPath of candidates) {
-      if (existsSync19(pkgPath)) {
-        const pkg = req(pkgPath);
-        if (pkg.name === "open-agents-ai" || pkg.name === "@open-agents/cli") {
-          currentVersion = pkg.version ?? "0.0.0";
-          break;
-        }
-      }
-    }
-  } catch {
+  const confirmPull = await ask(rl, `
+  Pull ${c2.bold(selectedVariant.tag)} (${selectedVariant.label})? (Y/n) `);
+  if (confirmPull.toLowerCase() === "n") {
+    process.stdout.write(`
+  ${c2.dim("Skipping model pull. You can pull manually with: ollama pull <model>")}
+
+`);
+    return config.model;
   }
   process.stdout.write(`
-  ${c2.cyan("\u25CF")} Checking for updates... ${c2.dim(`(current: v${currentVersion})`)}
+  ${c2.cyan("\u25CF")} Pulling ${c2.bold(selectedVariant.tag)}... (this may take a while)
 `);
-  const info = await checkForUpdate(currentVersion, true);
-  if (!info) {
-    process.stdout.write(`  ${c2.green("\u2714")} You're on the latest version (v${currentVersion}).
+  try {
+    pullModelWithAutoUpdate(selectedVariant.tag);
+    process.stdout.write(`
+  ${c2.green("\u2714")} Model ${c2.bold(selectedVariant.tag)} pulled successfully.
 
 `);
-    return;
+  } catch (err) {
+    renderError(`Failed to pull model: ${err instanceof Error ? err.message : String(err)}`);
+    renderInfo("Try manually: ollama pull " + selectedVariant.tag);
+    return config.model;
   }
-  process.stdout.write(`  ${c2.yellow("\u26A0")} Update available: v${info.currentVersion} \u2192 v${c2.bold(c2.green(info.latestVersion))}
+  if (!selectedVariant.cloud) {
+    const ctx = calculateContextWindow(specs, selectedVariant.sizeGB);
+    const customName = `open-agents-${selectedVariant.tag.replace(":", "-").replace(".", "")}`;
+    process.stdout.write(`  ${c2.cyan("\u25CF")} Context window recommendation: ${c2.bold(ctx.label)} (${ctx.numCtx} tokens)
 `);
-  process.stdout.write(`  ${c2.cyan("\u25CF")} Installing in background...
+    process.stdout.write(`    ${c2.dim(`Based on ${specs.totalRamGB.toFixed(0)} GB RAM, ${selectedVariant.sizeGB} GB model`)}
 
 `);
-  const { exec } = await import("node:child_process");
-  exec(`npm cache clean --force open-agents-ai 2>/dev/null; npm install -g open-agents-ai@latest --force`, { timeout: 18e4 }, (err) => {
-    if (err) {
-      renderWarning("Update install failed. Try manually: npm i -g open-agents-ai");
-    } else {
-      renderInfo(`${c2.green("\u2714")} Updated to v${info.latestVersion}. Takes effect next session.`);
+    const createModelfile = await ask(rl, `  Create optimized model "${c2.bold(customName)}" with ${ctx.label} context? (Y/n) `);
+    if (createModelfile.toLowerCase() !== "n") {
+      try {
+        const modelfileContent = [
+          `FROM ${selectedVariant.tag}`,
+          `PARAMETER num_ctx ${ctx.numCtx}`,
+          `PARAMETER temperature 0`,
+          `PARAMETER num_predict 16384`,
+          `PARAMETER stop "<|endoftext|>"`
+        ].join("\n");
+        const modelDir2 = join18(homedir8(), ".open-agents", "models");
+        mkdirSync7(modelDir2, { recursive: true });
+        const modelfilePath = join18(modelDir2, `Modelfile.${customName}`);
+        writeFileSync7(modelfilePath, modelfileContent + "\n", "utf8");
+        process.stdout.write(`  ${c2.dim("Creating model...")} `);
+        execSync10(`ollama create ${customName} -f ${modelfilePath}`, {
+          stdio: "pipe",
+          timeout: 12e4
+        });
+        process.stdout.write(`${c2.green("\u2714")}
+`);
+        setConfigValue("model", customName);
+        process.stdout.write(`
+  ${c2.green("\u2714")} Model ${c2.bold(customName)} created with ${ctx.label} context.
+`);
+        process.stdout.write(`  ${c2.green("\u2714")} Saved as default model in config.
+
+`);
+        return customName;
+      } catch (err) {
+        renderWarning(`Could not create custom model: ${err instanceof Error ? err.message : String(err)}`);
+        renderInfo(`Using base model ${selectedVariant.tag} instead.`);
+      }
     }
-  });
+    setConfigValue("model", selectedVariant.tag);
+    process.stdout.write(`
+  ${c2.green("\u2714")} Saved ${c2.bold(selectedVariant.tag)} as default model.
+
+`);
+    return selectedVariant.tag;
+  }
+  setConfigValue("model", selectedVariant.tag);
+  process.stdout.write(`
+  ${c2.green("\u2714")} Saved ${c2.bold(selectedVariant.tag)} as default model.
+
+`);
+  return selectedVariant.tag;
 }
-async function switchModel(query, ctx, local = false) {
+async function isModelAvailable(config) {
   try {
-    const models = await fetchOllamaModels(ctx.config.backendUrl);
-    const match = findModel(models, query);
-    if (!match) {
-      renderError(`Model not found: "${query}"`);
-      renderInfo("Available models:");
-      for (const m of models.slice(0, 10)) {
-        renderInfo(`  ${m.name}`);
-      }
-      return;
-    }
-    const oldModel = ctx.config.model;
-    ctx.setModel(match.name);
-    if (local) {
-      ctx.saveLocalSettings({ model: match.name });
-    } else {
-      ctx.saveSettings({ model: match.name });
-    }
-    renderModelSwitch(oldModel, match.name);
-    if (local) {
-      renderInfo("Saved as project-local override.");
-    }
-  } catch (err) {
-    renderError(`Failed to switch model: ${err instanceof Error ? err.message : String(err)}`);
+    const models = await fetchOllamaModels(config.backendUrl);
+    return !!findModel(models, config.model);
+  } catch {
+    return false;
   }
 }
-var init_commands = __esm({
-  "packages/cli/dist/tui/commands.js"() {
-    "use strict";
-    init_model_picker();
-    init_render();
-    init_dist2();
-    init_config();
-    init_updater();
-    init_oa_directory();
+function isFirstRun() {
+  try {
+    return !existsSync13(join18(homedir8(), ".open-agents", "config.json"));
+  } catch {
+    return true;
   }
-});
-
-// packages/cli/dist/tui/setup.js
-import * as readline from "node:readline";
-import { execSync as execSync10 } from "node:child_process";
-import { existsSync as existsSync13, writeFileSync as writeFileSync7, mkdirSync as mkdirSync7 } from "node:fs";
-import { join as join18 } from "node:path";
-import { homedir as homedir8 } from "node:os";
-function detectSystemSpecs() {
-  let totalRamGB = 0;
-  let availableRamGB = 0;
-  let gpuVramGB = 0;
-  let gpuName = "";
+}
+function expandedModelName(baseModel) {
+  return `open-agents-${baseModel.replace(":", "-").replace(/\./g, "")}`;
+}
+async function checkExpandedVariant(modelName, backendUrl) {
+  if (modelName.startsWith("open-agents-"))
+    return null;
+  const target = expandedModelName(modelName);
   try {
-    const memInfo = execSync10("free -b 2>/dev/null || sysctl -n hw.memsize 2>/dev/null", {
-      encoding: "utf8",
-      timeout: 5e3
-    });
-    if (memInfo.includes("Mem:")) {
-      const match = memInfo.match(/^Mem:\s+(\d+)\s+\d+\s+\d+\s+\d+\s+\d+\s+(\d+)/m);
-      if (match) {
-        totalRamGB = parseInt(match[1], 10) / 1024 ** 3;
-        availableRamGB = parseInt(match[2], 10) / 1024 ** 3;
-      }
-    } else {
-      const bytes = parseInt(memInfo.trim(), 10);
-      if (!isNaN(bytes)) {
-        totalRamGB = bytes / 1024 ** 3;
-        availableRamGB = totalRamGB * 0.7;
-      }
-    }
+    const models = await fetchOllamaModels(backendUrl);
+    const found = models.find((m) => m.name === target || m.name.startsWith(target + ":"));
+    return found ? found.name : false;
   } catch {
+    return false;
   }
+}
+function modelSizeGB(models, modelName) {
+  const m = findModel(models, modelName);
+  if (m)
+    return m.sizeBytes / 1024 ** 3;
+  const known = QWEN_VARIANTS.find((v) => modelName.includes(v.tag.split(":")[1] ?? ""));
+  return known?.sizeGB ?? 4;
+}
+function createExpandedVariant(baseModel, specs, sizeGB) {
+  const customName = expandedModelName(baseModel);
+  const ctx = calculateContextWindow(specs, sizeGB);
   try {
-    const nvidiaSmi = execSync10("nvidia-smi --query-gpu=memory.total,name --format=csv,noheader,nounits 2>/dev/null", { encoding: "utf8", timeout: 5e3 });
-    const lines = nvidiaSmi.trim().split("\n");
-    if (lines.length > 0) {
-      for (const line of lines) {
-        const parts = line.split(",").map((s) => s.trim());
-        const vramMB = parseInt(parts[0] ?? "0", 10);
-        if (!isNaN(vramMB))
-          gpuVramGB += vramMB / 1024;
-        if (!gpuName && parts[1])
-          gpuName = parts[1];
-      }
-    }
+    const modelfileContent = [
+      `FROM ${baseModel}`,
+      `PARAMETER num_ctx ${ctx.numCtx}`,
+      `PARAMETER temperature 0`,
+      `PARAMETER num_predict 16384`,
+      `PARAMETER stop "<|endoftext|>"`
+    ].join("\n");
+    const modelDir2 = join18(homedir8(), ".open-agents", "models");
+    mkdirSync7(modelDir2, { recursive: true });
+    const modelfilePath = join18(modelDir2, `Modelfile.${customName}`);
+    writeFileSync7(modelfilePath, modelfileContent + "\n", "utf8");
+    execSync10(`ollama create ${customName} -f ${modelfilePath}`, {
+      stdio: "pipe",
+      timeout: 12e4
+    });
+    return customName;
   } catch {
+    return null;
   }
-  return {
-    totalRamGB: Math.round(totalRamGB * 10) / 10,
-    availableRamGB: Math.round(availableRamGB * 10) / 10,
-    gpuVramGB: Math.round(gpuVramGB * 10) / 10,
-    gpuName
-  };
 }
-function recommendModel(specs) {
-  const effectiveGB = Math.max(specs.gpuVramGB, specs.availableRamGB);
-  const budget = effectiveGB * 0.8;
-  const localVariants = QWEN_VARIANTS.filter((v) => !v.cloud);
-  for (let i = localVariants.length - 1; i >= 0; i--) {
-    if (localVariants[i].sizeGB <= budget) {
-      return localVariants[i];
+async function ensureExpandedContext(modelName, backendUrl) {
+  if (modelName.startsWith("open-agents-")) {
+    const specs2 = detectSystemSpecs();
+    const ctx2 = calculateContextWindow(specs2, 4);
+    return { model: modelName, created: false, contextLabel: ctx2.label, numCtx: ctx2.numCtx };
+  }
+  if (modelName.includes("cloud") || modelName.includes(":cloud")) {
+    return { model: modelName, created: false, contextLabel: "remote", numCtx: 0 };
+  }
+  const existing = await checkExpandedVariant(modelName, backendUrl);
+  if (existing === null) {
+    return { model: modelName, created: false, contextLabel: "", numCtx: 0 };
+  }
+  const specs = detectSystemSpecs();
+  if (typeof existing === "string") {
+    let sizeGB2 = 4;
+    try {
+      const models = await fetchOllamaModels(backendUrl);
+      sizeGB2 = modelSizeGB(models, modelName);
+    } catch {
     }
+    const ctx2 = calculateContextWindow(specs, sizeGB2);
+    return { model: existing, created: false, contextLabel: ctx2.label, numCtx: ctx2.numCtx };
   }
-  return QWEN_VARIANTS.find((v) => v.tag === "qwen3.5:cloud");
-}
-function calculateContextWindow(specs, modelSizeGB) {
-  const totalAvail = Math.max(specs.gpuVramGB, specs.totalRamGB);
-  const remaining = totalAvail - modelSizeGB;
-  if (remaining >= 200)
-    return { numCtx: 131072, label: "128K" };
-  if (remaining >= 100)
-    return { numCtx: 65536, label: "64K" };
-  if (remaining >= 50)
-    return { numCtx: 32768, label: "32K" };
-  if (remaining >= 20)
-    return { numCtx: 16384, label: "16K" };
-  if (remaining >= 8)
-    return { numCtx: 8192, label: "8K" };
-  return { numCtx: 4096, label: "4K" };
-}
-function modelSupportsToolCalling(modelName) {
-  const lower = modelName.toLowerCase();
-  for (const known of TOOL_CALLING_MODELS) {
-    if (lower.startsWith(known) || lower.includes(known))
-      return true;
+  let sizeGB = 4;
+  try {
+    const models = await fetchOllamaModels(backendUrl);
+    sizeGB = modelSizeGB(models, modelName);
+  } catch {
   }
-  return false;
-}
-function ask(rl, question) {
-  return new Promise((resolve16) => {
-    rl.question(question, (answer) => resolve16(answer.trim()));
-  });
+  const ctx = calculateContextWindow(specs, sizeGB);
+  const created = createExpandedVariant(modelName, specs, sizeGB);
+  if (created) {
+    return { model: created, created: true, contextLabel: ctx.label, numCtx: ctx.numCtx };
+  }
+  return { model: modelName, created: false, contextLabel: ctx.label, numCtx: ctx.numCtx };
 }
-function pullModelWithAutoUpdate(tag) {
-  try {
-    execSync10(`ollama pull ${tag}`, {
-      stdio: "inherit",
-      timeout: 36e5
-      // 1 hour max
-    });
-  } catch (err) {
-    const errMsg = err instanceof Error ? err.message : String(err);
-    const stderr = err?.stderr?.toString?.() ?? errMsg;
-    const combined = errMsg + "\n" + stderr;
-    if (combined.includes("412") || combined.includes("newer version") || combined.includes("requires a newer version")) {
-      process.stdout.write(`
-  ${c2.yellow("\u26A0")} Ollama needs to be updated for this model.
-`);
-      process.stdout.write(`  ${c2.cyan("\u25CF")} Updating Ollama via official install script...
-
-`);
-      try {
-        execSync10("curl -fsSL https://ollama.com/install.sh | sh", {
-          stdio: "inherit",
-          timeout: 3e5
-          // 5 min max for install
-        });
-        process.stdout.write(`
-  ${c2.green("\u2714")} Ollama updated successfully.
-`);
-        process.stdout.write(`  ${c2.cyan("\u25CF")} Retrying pull of ${c2.bold(tag)}...
+var QWEN_VARIANTS, TOOL_CALLING_MODELS;
+var init_setup = __esm({
+  "packages/cli/dist/tui/setup.js"() {
+    "use strict";
+    init_model_picker();
+    init_render();
+    init_config();
+    QWEN_VARIANTS = [
+      { tag: "qwen3.5:0.8b", sizeGB: 1, label: "0.8B params (1.0 GB)", cloud: false },
+      { tag: "qwen3.5:2b", sizeGB: 2.7, label: "2B params (2.7 GB)", cloud: false },
+      { tag: "qwen3.5:4b", sizeGB: 3.4, label: "4B params (3.4 GB)", cloud: false },
+      { tag: "qwen3.5:9b", sizeGB: 6.6, label: "9B params (6.6 GB) \u2014 recommended minimum", cloud: false },
+      { tag: "qwen3.5:27b", sizeGB: 17, label: "27B params (17 GB)", cloud: false },
+      { tag: "qwen3.5:35b", sizeGB: 24, label: "35B params (24 GB)", cloud: false },
+      { tag: "qwen3.5:122b", sizeGB: 81, label: "122B params (81 GB) \u2014 best local", cloud: false },
+      { tag: "qwen3.5:cloud", sizeGB: 0, label: "Cloud (Ollama Cloud)", cloud: true },
+      { tag: "qwen3.5:397b-cloud", sizeGB: 0, label: "397B Cloud (Ollama Cloud)", cloud: true }
+    ];
+    TOOL_CALLING_MODELS = /* @__PURE__ */ new Set([
+      "qwen3.5",
+      "qwen3",
+      "qwen2.5",
+      "llama3.3",
+      "llama3.1",
+      "mistral",
+      "mixtral",
+      "command-r",
+      "gemma3",
+      "devstral",
+      "deepseek"
+    ]);
+  }
+});
 
-`);
-        execSync10(`ollama pull ${tag}`, {
-          stdio: "inherit",
-          timeout: 36e5
-        });
-      } catch (updateErr) {
-        const updateMsg = updateErr instanceof Error ? updateErr.message : String(updateErr);
-        throw new Error(`Failed to update Ollama and retry pull: ${updateMsg}
-Try manually:
-  curl -fsSL https://ollama.com/install.sh | sh
-  ollama pull ${tag}`);
+// packages/cli/dist/tui/commands.js
+async function handleSlashCommand(input, ctx) {
+  const trimmed = input.trim();
+  if (!trimmed.startsWith("/"))
+    return "not_a_command";
+  const [cmd, ...rest] = trimmed.slice(1).split(/\s+/);
+  const hasLocal = rest.includes("--local");
+  const filteredRest = rest.filter((r) => r !== "--local");
+  const arg = filteredRest.join(" ").trim();
+  switch (cmd) {
+    case "help":
+    case "h":
+    case "?":
+      renderSlashHelp();
+      return "handled";
+    case "quit":
+    case "exit":
+    case "q":
+      return "exit";
+    case "clear":
+    case "cls":
+      ctx.clearScreen();
+      return "handled";
+    case "verbose":
+    case "v":
+      ctx.setVerbose(!ctx.config.verbose);
+      if (hasLocal) {
+        ctx.saveLocalSettings({ verbose: ctx.config.verbose });
+        renderInfo(`Verbose mode: ${ctx.config.verbose ? "on" : "off"} (project-local)`);
+      } else {
+        ctx.saveSettings({ verbose: ctx.config.verbose });
+        renderInfo(`Verbose mode: ${ctx.config.verbose ? "on" : "off"}`);
       }
-    } else {
-      throw err;
+      return "handled";
+    case "config":
+    case "cfg":
+      renderConfig({
+        model: ctx.config.model,
+        backendType: ctx.config.backendType,
+        backendUrl: ctx.config.backendUrl,
+        timeoutMs: String(ctx.config.timeoutMs),
+        maxRetries: String(ctx.config.maxRetries),
+        verbose: String(ctx.config.verbose),
+        dryRun: String(ctx.config.dryRun)
+      });
+      return "handled";
+    case "model":
+      if (arg) {
+        await switchModel(arg, ctx, hasLocal);
+      } else {
+        await showModelPicker(ctx);
+      }
+      return "handled";
+    case "models":
+      await listModels(ctx);
+      return "handled";
+    case "endpoint":
+    case "ep":
+      await handleEndpoint(arg, ctx, hasLocal);
+      return "handled";
+    case "update":
+    case "upgrade":
+      await handleUpdate(arg, ctx.repoRoot);
+      return "handled";
+    case "voice": {
+      const save = hasLocal ? ctx.saveLocalSettings.bind(ctx) : ctx.saveSettings.bind(ctx);
+      if (arg) {
+        const msg = await ctx.voiceSetModel(arg);
+        save({ voice: true, voiceModel: arg });
+        renderInfo(msg + (hasLocal ? " (project-local)" : ""));
+      } else {
+        const msg = await ctx.voiceToggle();
+        const isOn = msg.toLowerCase().includes("enabled") || msg.toLowerCase().includes("on");
+        save({ voice: isOn });
+        renderInfo(msg + (hasLocal ? " (project-local)" : ""));
+      }
+      return "handled";
     }
-  }
-}
-async function runSetupWizard(config) {
-  const rl = readline.createInterface({
-    input: process.stdin,
-    output: process.stdout,
-    terminal: true
-  });
-  try {
-    return await doSetup(config, rl);
-  } finally {
-    rl.close();
-  }
-}
-async function doSetup(config, rl) {
-  process.stdout.write(`
-  ${c2.bold(c2.cyan("open-agents"))}
+    case "stream": {
+      const isOn = ctx.streamToggle();
+      const save = hasLocal ? ctx.saveLocalSettings.bind(ctx) : ctx.saveSettings.bind(ctx);
+      save({ stream: isOn });
+      renderInfo(`Token streaming: ${isOn ? "on" : "off"}${hasLocal ? " (project-local)" : ""}` + (isOn ? " \u2014 thinking tokens in grey italics, responses with pastel syntax highlighting" : ""));
+      return "handled";
+    }
+    case "tools": {
+      const tools = listCustomToolFiles(ctx.repoRoot);
+      if (tools.length === 0) {
+        renderInfo("No custom tools installed.");
+        renderInfo("The agent will automatically create tools when it detects repeated workflows (3+ times).");
+        renderInfo('Or ask the agent: "create a tool for [workflow]"');
+      } else {
+        process.stdout.write(`
+  ${c2.bold("Custom Tools:")}
+
 `);
-  process.stdout.write(`  ${c2.dim("\u2500".repeat(60))}
+        for (const t of tools) {
+          process.stdout.write(`  ${c2.cyan(t.name.padEnd(28))} ${c2.dim(`(${t.scope}, v${t.version}, ${t.stepsCount} steps)`)}
 `);
-  process.stdout.write(`  ${c2.bold("First-run setup")}
-
+          process.stdout.write(`  ${"".padEnd(28)} ${t.description}
 `);
-  process.stdout.write(`  ${c2.cyan("\u25CF")} Detecting system specs...
+        }
+        process.stdout.write("\n");
+      }
+      return "handled";
+    }
+    case "skills":
+    case "skill": {
+      const skills = discoverSkills(ctx.repoRoot);
+      if (skills.length === 0) {
+        renderInfo("No skills found.");
+        renderInfo("Install AIWG to get skills: npm i -g aiwg && aiwg use sdlc");
+        renderInfo("Or add skills manually to .oa/skills/{name}/SKILL.md");
+      } else {
+        let filtered = skills;
+        if (arg) {
+          const q = arg.toLowerCase();
+          filtered = skills.filter((s) => s.name.toLowerCase().includes(q) || s.description.toLowerCase().includes(q) || s.triggers.some((t) => t.toLowerCase().includes(q)));
+        }
+        if (filtered.length === 0) {
+          renderWarning(`No skills matching "${arg}". Showing all ${skills.length} skills:`);
+          filtered = skills;
+        }
+        const bySource = /* @__PURE__ */ new Map();
+        for (const s of filtered) {
+          const group = bySource.get(s.source) ?? [];
+          group.push(s);
+          bySource.set(s.source, group);
+        }
+        process.stdout.write(`
+  ${c2.bold(`Available Skills (${filtered.length}):`)}
 `);
-  const specs = detectSystemSpecs();
-  process.stdout.write(`  ${c2.dim("  RAM:")} ${specs.totalRamGB.toFixed(1)} GB total, ${specs.availableRamGB.toFixed(1)} GB available
+        for (const [source, group] of bySource) {
+          process.stdout.write(`
+  ${c2.dim(`\u2500\u2500 ${source} (${group.length}) \u2500\u2500`)}
 `);
-  if (specs.gpuVramGB > 0) {
-    process.stdout.write(`  ${c2.dim("  GPU:")} ${specs.gpuName || "NVIDIA"} \u2014 ${specs.gpuVramGB.toFixed(1)} GB VRAM
+          for (const s of group) {
+            process.stdout.write(`  ${c2.cyan(s.name.padEnd(32))} ${s.description.slice(0, 60)}
 `);
-  } else {
-    process.stdout.write(`  ${c2.dim("  GPU:")} No NVIDIA GPU detected (CPU inference)
+            if (s.triggers.length > 0) {
+              process.stdout.write(`  ${"".padEnd(32)} ${c2.dim(`triggers: ${s.triggers.slice(0, 3).join(" | ")}`)}
 `);
+            }
+          }
+        }
+        process.stdout.write("\n");
+        renderInfo('Invoke directly: /<skill-name> [args]  (e.g. /ralph "fix tests" --completion "npm test passes")');
+        renderInfo("Filter with: /skills <keyword>");
+      }
+      return "handled";
+    }
+    case "dream": {
+      if (arg === "stop" || arg === "wake") {
+        if (ctx.isDreaming?.()) {
+          ctx.dreamStop?.();
+          renderInfo("Waking up from dream mode...");
+        } else {
+          renderWarning("Not currently dreaming.");
+        }
+      } else if (ctx.isDreaming?.()) {
+        renderWarning("Already dreaming. Use /dream stop to wake up first.");
+      } else {
+        const mode = arg === "lucid" ? "lucid" : arg === "deep" ? "deep" : "default";
+        ctx.dreamStart?.(mode);
+      }
+      return "handled";
+    }
+    case "listen":
+    case "mic": {
+      if (!ctx.listenToggle) {
+        renderWarning("Listen mode not available in this context.");
+        return "handled";
+      }
+      if (arg === "stop" || arg === "off") {
+        const msg2 = await (ctx.listenStop?.() ?? Promise.resolve("Not listening."));
+        renderInfo(msg2);
+        return "handled";
+      }
+      if (arg === "confirm") {
+        const msg2 = ctx.listenSetMode?.("confirm") ?? "Confirm mode set.";
+        renderInfo(msg2);
+        return "handled";
+      }
+      if (arg === "auto") {
+        const msg2 = ctx.listenSetMode?.("auto") ?? "Auto mode set.";
+        renderInfo(msg2);
+        return "handled";
+      }
+      const modelSizes = ["tiny", "base", "small", "medium", "large", "large-v3"];
+      if (arg && modelSizes.includes(arg.toLowerCase())) {
+        const model = arg.toLowerCase() === "large" ? "large-v3" : arg.toLowerCase();
+        const msg2 = await (ctx.listenSetModel?.(model) ?? Promise.resolve(`Model set to ${model}.`));
+        renderInfo(msg2);
+        return "handled";
+      }
+      const msg = await ctx.listenToggle();
+      renderInfo(msg);
+      return "handled";
+    }
+    case "bruteforce":
+    case "brute": {
+      const isOn = ctx.bruteForceToggle();
+      const save = hasLocal ? ctx.saveLocalSettings.bind(ctx) : ctx.saveSettings.bind(ctx);
+      save({ bruteforce: isOn });
+      renderInfo(`Brute-force mode: ${isOn ? "on" : "off"}${hasLocal ? " (project-local)" : ""}` + (isOn ? " \u2014 agent will auto re-engage when turn limit is hit, reassess and try creative strategies" : ""));
+      return "handled";
+    }
+    case "emojis":
+    case "emoji": {
+      const current = ctx.getEmojis?.() ?? true;
+      const next = !current;
+      ctx.setEmojis?.(next);
+      const save = hasLocal ? ctx.saveLocalSettings.bind(ctx) : ctx.saveSettings.bind(ctx);
+      save({ emojis: next });
+      renderInfo(`Emojis ${next ? "enabled" : "disabled"}.`);
+      return "handled";
+    }
+    case "colors":
+    case "color": {
+      const current = ctx.getColors?.() ?? true;
+      const next = !current;
+      ctx.setColors?.(next);
+      const save = hasLocal ? ctx.saveLocalSettings.bind(ctx) : ctx.saveSettings.bind(ctx);
+      save({ colors: next });
+      renderInfo(`Colors ${next ? "enabled" : "disabled"}.`);
+      return "handled";
+    }
+    default: {
+      const skills = discoverSkills(ctx.repoRoot);
+      const skill = skills.find((s) => s.name === cmd || s.name === cmd.replace(/_/g, "-"));
+      if (skill) {
+        const content = loadSkillContent(skill.filePath);
+        if (content) {
+          renderInfo(`Loading skill: ${c2.bold(skill.name)} (${skill.source})`);
+          return { type: "skill", name: skill.name, content, args: arg };
+        }
+      }
+      renderWarning(`Unknown command: /${cmd}. Type /help for available commands.`);
+      return "handled";
+    }
   }
-  process.stdout.write("\n");
-  let models = [];
+}
+async function listModels(ctx) {
   try {
-    models = await fetchOllamaModels(config.backendUrl);
-  } catch {
-    renderError(`Cannot reach Ollama at ${config.backendUrl}`);
-    renderInfo("Start Ollama with: ollama serve");
-    renderInfo("Or use /endpoint to configure a remote backend after startup.");
-    const answer = await ask(rl, `
-  ${c2.bold("Continue without Ollama?")} (y/n) `);
-    if (answer.toLowerCase() !== "y")
-      return null;
-    return config.model;
-  }
-  const currentModel = findModel(models, config.model);
-  if (currentModel) {
-    process.stdout.write(`  ${c2.green("\u2714")} Model ${c2.bold(currentModel.name)} is available.
-
-`);
-    return currentModel.name;
+    const models = await fetchOllamaModels(ctx.config.backendUrl);
+    renderModelList(models.map((m) => ({ name: m.name, size: m.size, modified: m.modified })), ctx.config.model);
+  } catch (err) {
+    renderError(`Failed to fetch models: ${err instanceof Error ? err.message : String(err)}`);
   }
-  process.stdout.write(`  ${c2.yellow("\u26A0")} Default model ${c2.bold(config.model)} is not available.
-
-`);
-  const toolCallingModels = models.filter((m) => modelSupportsToolCalling(m.name));
-  if (toolCallingModels.length > 0) {
-    process.stdout.write(`  ${c2.cyan("\u25CF")} Found ${toolCallingModels.length} model(s) with tool-calling support:
-
-`);
-    for (let i = 0; i < Math.min(toolCallingModels.length, 10); i++) {
-      const m = toolCallingModels[i];
-      process.stdout.write(`    ${c2.bold(String(i + 1))}. ${m.name} ${c2.dim(`(${m.size})`)}
-`);
-    }
-    process.stdout.write(`
-    ${c2.dim("0")}. Pull a new qwen3.5 model instead
-`);
-    process.stdout.write("\n");
-    const choice = await ask(rl, `  ${c2.bold("Select a model")} (1-${Math.min(toolCallingModels.length, 10)}, or 0 to pull new): `);
-    const idx = parseInt(choice, 10);
-    if (idx > 0 && idx <= toolCallingModels.length) {
-      const selected = toolCallingModels[idx - 1];
-      setConfigValue("model", selected.name);
-      process.stdout.write(`
-  ${c2.green("\u2714")} Selected ${c2.bold(selected.name)}. Saved to config.
-
-`);
-      return selected.name;
+}
+async function showModelPicker(ctx) {
+  try {
+    const models = await fetchOllamaModels(ctx.config.backendUrl);
+    if (models.length === 0) {
+      renderWarning("No models found. Pull a model with: ollama pull <model>");
+      return;
     }
-  } else {
-    process.stdout.write(`  ${c2.yellow("\u26A0")} No tool-calling capable models found on this system.
-
-`);
+    renderModelList(models.map((m) => ({ name: m.name, size: m.size, modified: m.modified })), ctx.config.model);
+  } catch (err) {
+    renderError(`Failed to fetch models: ${err instanceof Error ? err.message : String(err)}`);
   }
-  const recommended = recommendModel(specs);
-  process.stdout.write(`  ${c2.cyan("\u25CF")} Recommended model based on your system:
+}
+async function handleEndpoint(arg, ctx, local = false) {
+  if (!arg) {
+    process.stdout.write(`
+  ${c2.bold("Current endpoint:")}
 
 `);
-  const localVariants = QWEN_VARIANTS.filter((v) => !v.cloud);
-  for (let i = 0; i < localVariants.length; i++) {
-    const v = localVariants[i];
-    const fits = v.sizeGB <= Math.max(specs.gpuVramGB, specs.availableRamGB) * 0.8;
-    const isRec = v.tag === recommended.tag;
-    const marker = isRec ? c2.green("\u2192") : fits ? c2.dim(" ") : c2.red("\u2716");
-    const name = isRec ? c2.bold(c2.green(v.tag)) : fits ? v.tag : c2.dim(v.tag);
-    const label = isRec ? c2.bold(v.label) : c2.dim(v.label);
-    const tooLarge = !fits && !v.cloud ? c2.red(" (exceeds available memory)") : "";
-    process.stdout.write(`    ${marker} ${String(i + 1).padStart(2)}. ${name.padEnd(isRec ? 45 : 25)} ${label}${tooLarge}
-`);
-  }
-  process.stdout.write(`
-    ${c2.dim(" ")} ${String(localVariants.length + 1).padStart(2)}. ${c2.dim("qwen3.5:cloud")}              ${c2.dim("Ollama Cloud")}
-`);
-  process.stdout.write(`    ${c2.dim(" ")} ${String(localVariants.length + 2).padStart(2)}. ${c2.dim("qwen3.5:397b-cloud")}         ${c2.dim("397B Ollama Cloud")}
-`);
-  process.stdout.write("\n");
-  const pullChoice = await ask(rl, `  ${c2.bold("Select a model to pull")} (1-${localVariants.length + 2}, or Enter for recommended): `);
-  const pullIdx = pullChoice ? parseInt(pullChoice, 10) : 0;
-  let selectedVariant;
-  if (pullIdx === 0 || isNaN(pullIdx)) {
-    selectedVariant = recommended;
-  } else if (pullIdx <= localVariants.length) {
-    selectedVariant = localVariants[pullIdx - 1];
-  } else if (pullIdx === localVariants.length + 1) {
-    selectedVariant = QWEN_VARIANTS.find((v) => v.tag === "qwen3.5:cloud");
-  } else {
-    selectedVariant = QWEN_VARIANTS.find((v) => v.tag === "qwen3.5:397b-cloud");
-  }
-  const confirmPull = await ask(rl, `
-  Pull ${c2.bold(selectedVariant.tag)} (${selectedVariant.label})? (Y/n) `);
-  if (confirmPull.toLowerCase() === "n") {
+    process.stdout.write(`  ${c2.cyan("URL".padEnd(12))} ${ctx.config.backendUrl}
+`);
+    process.stdout.write(`  ${c2.cyan("Type".padEnd(12))} ${ctx.config.backendType}
+`);
+    process.stdout.write(`  ${c2.cyan("Auth".padEnd(12))} ${ctx.config.apiKey ? "Bearer token set" : "none"}
+`);
     process.stdout.write(`
-  ${c2.dim("Skipping model pull. You can pull manually with: ollama pull <model>")}
+  ${c2.dim("Usage: /endpoint <url> [--auth <token>]")}
+`);
+    process.stdout.write(`  ${c2.dim("  /endpoint http://localhost:11434         (Ollama, no auth)")}
+`);
+    process.stdout.write(`  ${c2.dim("  /endpoint http://remote:8000/v1 --auth sk-...  (OpenAI-compatible)")}
+`);
+    process.stdout.write(`  ${c2.dim("  /endpoint http://remote:8000/v1          (OpenAI-compatible, no auth)")}
 
 `);
-    return config.model;
+    return;
+  }
+  const parts = arg.split(/\s+/);
+  const url = parts[0];
+  let apiKey;
+  const authIdx = parts.indexOf("--auth");
+  if (authIdx !== -1 && parts[authIdx + 1]) {
+    apiKey = parts[authIdx + 1];
+  }
+  try {
+    new URL(url);
+  } catch {
+    renderError(`Invalid URL: "${url}"`);
+    return;
+  }
+  let backendType = "ollama";
+  if (url.includes("/v1") || url.includes(":8000") || apiKey) {
+    backendType = "vllm";
   }
   process.stdout.write(`
-  ${c2.cyan("\u25CF")} Pulling ${c2.bold(selectedVariant.tag)}... (this may take a while)
-`);
+  ${c2.dim("Testing connection...")} `);
   try {
-    pullModelWithAutoUpdate(selectedVariant.tag);
-    process.stdout.write(`
-  ${c2.green("\u2714")} Model ${c2.bold(selectedVariant.tag)} pulled successfully.
-
+    const healthUrl = backendType === "ollama" ? `${url.replace(/\/$/, "")}/api/tags` : `${url.replace(/\/$/, "")}/models`;
+    const headers = {};
+    if (apiKey)
+      headers["Authorization"] = `Bearer ${apiKey}`;
+    const resp = await fetch(healthUrl, {
+      headers,
+      signal: AbortSignal.timeout(1e4)
+    });
+    if (!resp.ok)
+      throw new Error(`HTTP ${resp.status}`);
+    process.stdout.write(`${c2.green("\u2714")} Connected
 `);
   } catch (err) {
-    renderError(`Failed to pull model: ${err instanceof Error ? err.message : String(err)}`);
-    renderInfo("Try manually: ollama pull " + selectedVariant.tag);
-    return config.model;
+    process.stdout.write(`${c2.yellow("\u26A0")} Could not verify
+`);
+    renderWarning(`Endpoint may not be reachable: ${err instanceof Error ? err.message : String(err)}`);
+    renderInfo("Setting endpoint anyway \u2014 it may come online later.");
   }
-  if (!selectedVariant.cloud) {
-    const ctx = calculateContextWindow(specs, selectedVariant.sizeGB);
-    const customName = `open-agents-${selectedVariant.tag.replace(":", "-").replace(".", "")}`;
-    process.stdout.write(`  ${c2.cyan("\u25CF")} Context window recommendation: ${c2.bold(ctx.label)} (${ctx.numCtx} tokens)
+  ctx.setEndpoint(url, backendType, apiKey);
+  const endpointSettings = { backendUrl: url, backendType, ...apiKey ? { apiKey } : {} };
+  if (local) {
+    ctx.saveLocalSettings(endpointSettings);
+  } else {
+    setConfigValue("backendUrl", url);
+    setConfigValue("backendType", backendType);
+    if (apiKey) {
+      setConfigValue("apiKey", apiKey);
+    }
+    ctx.saveSettings(endpointSettings);
+  }
+  process.stdout.write(`
+  ${c2.green("\u2714")} Endpoint updated and saved${local ? " (project-local)" : ""}:
 `);
-    process.stdout.write(`    ${c2.dim(`Based on ${specs.totalRamGB.toFixed(0)} GB RAM, ${selectedVariant.sizeGB} GB model`)}
-
+  process.stdout.write(`    ${c2.cyan("URL".padEnd(8))} ${url}
 `);
-    const createModelfile = await ask(rl, `  Create optimized model "${c2.bold(customName)}" with ${ctx.label} context? (Y/n) `);
-    if (createModelfile.toLowerCase() !== "n") {
-      try {
-        const modelfileContent = [
-          `FROM ${selectedVariant.tag}`,
-          `PARAMETER num_ctx ${ctx.numCtx}`,
-          `PARAMETER temperature 0`,
-          `PARAMETER num_predict 16384`,
-          `PARAMETER stop "<|endoftext|>"`
-        ].join("\n");
-        const modelDir2 = join18(homedir8(), ".open-agents", "models");
-        mkdirSync7(modelDir2, { recursive: true });
-        const modelfilePath = join18(modelDir2, `Modelfile.${customName}`);
-        writeFileSync7(modelfilePath, modelfileContent + "\n", "utf8");
-        process.stdout.write(`  ${c2.dim("Creating model...")} `);
-        execSync10(`ollama create ${customName} -f ${modelfilePath}`, {
-          stdio: "pipe",
-          timeout: 12e4
-        });
-        process.stdout.write(`${c2.green("\u2714")}
+  process.stdout.write(`    ${c2.cyan("Type".padEnd(8))} ${backendType}
 `);
-        setConfigValue("model", customName);
-        process.stdout.write(`
-  ${c2.green("\u2714")} Model ${c2.bold(customName)} created with ${ctx.label} context.
+  if (apiKey) {
+    process.stdout.write(`    ${c2.cyan("Auth".padEnd(8))} Bearer ${apiKey.slice(0, 8)}...
 `);
-        process.stdout.write(`  ${c2.green("\u2714")} Saved as default model in config.
-
+  } else {
+    process.stdout.write(`    ${c2.cyan("Auth".padEnd(8))} none
 `);
-        return customName;
-      } catch (err) {
-        renderWarning(`Could not create custom model: ${err instanceof Error ? err.message : String(err)}`);
-        renderInfo(`Using base model ${selectedVariant.tag} instead.`);
+  }
+  process.stdout.write("\n");
+}
+async function handleUpdate(subcommand, repoRoot) {
+  if (subcommand === "auto") {
+    const settings = { updateMode: "auto" };
+    saveProjectSettings(repoRoot, settings);
+    saveGlobalSettings(settings);
+    renderInfo("Update mode: auto \u2014 updates will install automatically after task completion.");
+    return;
+  }
+  if (subcommand === "manual") {
+    const settings = { updateMode: "manual" };
+    saveProjectSettings(repoRoot, settings);
+    saveGlobalSettings(settings);
+    renderInfo("Update mode: manual \u2014 updates only install when you run /update.");
+    return;
+  }
+  let currentVersion = "0.0.0";
+  try {
+    const { createRequire: createRequire4 } = await import("node:module");
+    const { fileURLToPath: fileURLToPath3 } = await import("node:url");
+    const { dirname: dirname5, join: join28 } = await import("node:path");
+    const { existsSync: existsSync19 } = await import("node:fs");
+    const req = createRequire4(import.meta.url);
+    const thisDir = dirname5(fileURLToPath3(import.meta.url));
+    const candidates = [
+      join28(thisDir, "..", "package.json"),
+      join28(thisDir, "..", "..", "package.json"),
+      join28(thisDir, "..", "..", "..", "package.json")
+    ];
+    for (const pkgPath of candidates) {
+      if (existsSync19(pkgPath)) {
+        const pkg = req(pkgPath);
+        if (pkg.name === "open-agents-ai" || pkg.name === "@open-agents/cli") {
+          currentVersion = pkg.version ?? "0.0.0";
+          break;
+        }
       }
     }
-    setConfigValue("model", selectedVariant.tag);
-    process.stdout.write(`
-  ${c2.green("\u2714")} Saved ${c2.bold(selectedVariant.tag)} as default model.
-
-`);
-    return selectedVariant.tag;
+  } catch {
   }
-  setConfigValue("model", selectedVariant.tag);
   process.stdout.write(`
-  ${c2.green("\u2714")} Saved ${c2.bold(selectedVariant.tag)} as default model.
+  ${c2.cyan("\u25CF")} Checking for updates... ${c2.dim(`(current: v${currentVersion})`)}
+`);
+  const info = await checkForUpdate(currentVersion, true);
+  if (!info) {
+    process.stdout.write(`  ${c2.green("\u2714")} You're on the latest version (v${currentVersion}).
 
 `);
-  return selectedVariant.tag;
-}
-async function isModelAvailable(config) {
-  try {
-    const models = await fetchOllamaModels(config.backendUrl);
-    return !!findModel(models, config.model);
-  } catch {
-    return false;
+    return;
   }
+  process.stdout.write(`  ${c2.yellow("\u26A0")} Update available: v${info.currentVersion} \u2192 v${c2.bold(c2.green(info.latestVersion))}
+`);
+  process.stdout.write(`  ${c2.cyan("\u25CF")} Installing in background...
+
+`);
+  const { exec } = await import("node:child_process");
+  exec(`npm cache clean --force open-agents-ai 2>/dev/null; npm install -g open-agents-ai@latest --force`, { timeout: 18e4 }, (err) => {
+    if (err) {
+      renderWarning("Update install failed. Try manually: npm i -g open-agents-ai");
+    } else {
+      renderInfo(`${c2.green("\u2714")} Updated to v${info.latestVersion}. Takes effect next session.`);
+    }
+  });
 }
-function isFirstRun() {
+async function switchModel(query, ctx, local = false) {
   try {
-    return !existsSync13(join18(homedir8(), ".open-agents", "config.json"));
-  } catch {
-    return true;
+    const models = await fetchOllamaModels(ctx.config.backendUrl);
+    const match = findModel(models, query);
+    if (!match) {
+      renderError(`Model not found: "${query}"`);
+      renderInfo("Available models:");
+      for (const m of models.slice(0, 10)) {
+        renderInfo(`  ${m.name}`);
+      }
+      return;
+    }
+    let finalModel = match.name;
+    if (ctx.config.backendType === "ollama") {
+      const result = await ensureExpandedContext(match.name, ctx.config.backendUrl);
+      if (result.created) {
+        renderInfo(`Created expanded context variant: ${c2.bold(result.model)} (${result.contextLabel}, ${result.numCtx} tokens)`);
+        finalModel = result.model;
+      } else if (result.model !== match.name) {
+        renderInfo(`Using expanded context variant: ${c2.bold(result.model)} (${result.contextLabel})`);
+        finalModel = result.model;
+      }
+    }
+    const oldModel = ctx.config.model;
+    ctx.setModel(finalModel);
+    if (local) {
+      ctx.saveLocalSettings({ model: finalModel });
+    } else {
+      ctx.saveSettings({ model: finalModel });
+    }
+    renderModelSwitch(oldModel, finalModel);
+    if (local) {
+      renderInfo("Saved as project-local override.");
+    }
+  } catch (err) {
+    renderError(`Failed to switch model: ${err instanceof Error ? err.message : String(err)}`);
   }
 }
-var QWEN_VARIANTS, TOOL_CALLING_MODELS;
-var init_setup = __esm({
-  "packages/cli/dist/tui/setup.js"() {
+var init_commands = __esm({
+  "packages/cli/dist/tui/commands.js"() {
     "use strict";
     init_model_picker();
     init_render();
+    init_dist2();
     init_config();
-    QWEN_VARIANTS = [
-      { tag: "qwen3.5:0.8b", sizeGB: 1, label: "0.8B params (1.0 GB)", cloud: false },
-      { tag: "qwen3.5:2b", sizeGB: 2.7, label: "2B params (2.7 GB)", cloud: false },
-      { tag: "qwen3.5:4b", sizeGB: 3.4, label: "4B params (3.4 GB)", cloud: false },
-      { tag: "qwen3.5:9b", sizeGB: 6.6, label: "9B params (6.6 GB) \u2014 recommended minimum", cloud: false },
-      { tag: "qwen3.5:27b", sizeGB: 17, label: "27B params (17 GB)", cloud: false },
-      { tag: "qwen3.5:35b", sizeGB: 24, label: "35B params (24 GB)", cloud: false },
-      { tag: "qwen3.5:122b", sizeGB: 81, label: "122B params (81 GB) \u2014 best local", cloud: false },
-      { tag: "qwen3.5:cloud", sizeGB: 0, label: "Cloud (Ollama Cloud)", cloud: true },
-      { tag: "qwen3.5:397b-cloud", sizeGB: 0, label: "397B Cloud (Ollama Cloud)", cloud: true }
-    ];
-    TOOL_CALLING_MODELS = /* @__PURE__ */ new Set([
-      "qwen3.5",
-      "qwen3",
-      "qwen2.5",
-      "llama3.3",
-      "llama3.1",
-      "mistral",
-      "mixtral",
-      "command-r",
-      "gemma3",
-      "devstral",
-      "deepseek"
-    ]);
+    init_updater();
+    init_oa_directory();
+    init_setup();
   }
 });
 
@@ -14965,6 +15053,19 @@ async function startInteractive(config, repoPath) {
       config = { ...config, model: setupModel };
     }
   }
+  if (config.backendType === "ollama" && !config.model.startsWith("open-agents-")) {
+    try {
+      const expandResult = await ensureExpandedContext(config.model, config.backendUrl);
+      if (expandResult.created) {
+        renderInfo(`Created expanded context model: ${expandResult.model} (${expandResult.contextLabel}, ${expandResult.numCtx} tokens)`);
+        config = { ...config, model: expandResult.model };
+      } else if (expandResult.model !== config.model) {
+        renderInfo(`Using expanded context model: ${expandResult.model} (${expandResult.contextLabel})`);
+        config = { ...config, model: expandResult.model };
+      }
+    } catch {
+    }
+  }
   if (!isResumed) {
     try {
       const healthUrl = config.backendType === "ollama" ? `${config.backendUrl}/api/tags` : `${config.backendUrl}/v1/models`;
@@ -15486,6 +15587,15 @@ async function runWithTUI(task, config, repoPath) {
     }
     config = { ...config, model: setupModel };
   }
+  if (config.backendType === "ollama" && !config.model.startsWith("open-agents-")) {
+    try {
+      const expandResult = await ensureExpandedContext(config.model, config.backendUrl);
+      if (expandResult.model !== config.model) {
+        config = { ...config, model: expandResult.model };
+      }
+    } catch {
+    }
+  }
   try {
     const healthUrl = config.backendType === "ollama" ? `${config.backendUrl}/api/tags` : `${config.backendUrl}/v1/models`;
     const resp = await fetch(healthUrl, { signal: AbortSignal.timeout(1e4) });