open-agents-ai 0.103.99 → 0.104.2

package/dist/index.js

@@ -30893,6 +30893,7 @@ ${activitySummary}
 });
 
 // packages/cli/dist/tui/model-picker.js
+import { totalmem as totalmem2 } from "node:os";
 async function fetchOllamaModels(baseUrl) {
   const url = `${normalizeBaseUrl(baseUrl)}/api/tags`;
   const resp = await fetch(url, {
@@ -30932,10 +30933,22 @@ async function fetchOllamaModels(baseUrl) {
       }
     }
     if (show.model_info) {
-      for (const [key, value] of Object.entries(show.model_info)) {
-        if (key.endsWith(".context_length") && typeof value === "number") {
-          result[i].contextLength = value;
-          break;
+      const info = show.model_info;
+      const arch = info["general.architecture"];
+      const paramCount = info["general.parameter_count"];
+      const fileSizeGB = result[i].sizeBytes > 0 ? result[i].sizeBytes / 1024 ** 3 : paramCount ? paramCount * 0.6 / 1024 ** 3 : 4;
+      if (arch) {
+        const archMax = info[`${arch}.context_length`];
+        const nLayers = info[`${arch}.block_count`];
+        const nKVHeads = info[`${arch}.attention.head_count_kv`] ?? info[`${arch}.attention.head_count`];
+        const keyDim = info[`${arch}.attention.key_length`];
+        const valDim = info[`${arch}.attention.value_length`] ?? keyDim;
+        if (archMax && nLayers && nKVHeads && keyDim && valDim) {
+          const kvBytesPerToken = nLayers * nKVHeads * (keyDim + valDim) * 2;
+          result[i].contextLength = estimateRealisticContext(kvBytesPerToken, archMax, fileSizeGB);
+        } else if (archMax) {
+          const kvEstimate = fileSizeGB <= 5 ? 524288 : fileSizeGB <= 20 ? 1048576 : 1572864;
+          result[i].contextLength = estimateRealisticContext(kvEstimate, archMax, fileSizeGB);
         }
       }
     }
@@ -31231,9 +31244,23 @@ async function queryModelContextSize(baseUrl, modelName) {
         return parseInt(match[1], 10);
     }
     if (data.model_info) {
-      for (const [key, value] of Object.entries(data.model_info)) {
-        if (key.endsWith(".context_length") && typeof value === "number") {
-          return value;
+      const info = data.model_info;
+      const arch = info["general.architecture"];
+      const paramCount = info["general.parameter_count"];
+      const modelSizeGB2 = paramCount ? paramCount * 0.6 / 1024 ** 3 : 4;
+      if (arch) {
+        const archMax = info[`${arch}.context_length`];
+        const nLayers = info[`${arch}.block_count`];
+        const nKVHeads = info[`${arch}.attention.head_count_kv`] ?? info[`${arch}.attention.head_count`];
+        const keyDim = info[`${arch}.attention.key_length`];
+        const valDim = info[`${arch}.attention.value_length`] ?? keyDim;
+        if (archMax && nLayers && nKVHeads && keyDim && valDim) {
+          const kvBytesPerToken = nLayers * nKVHeads * (keyDim + valDim) * 2;
+          return estimateRealisticContext(kvBytesPerToken, archMax, modelSizeGB2);
+        }
+        if (archMax) {
+          const kvEstimate = modelSizeGB2 <= 5 ? 524288 : modelSizeGB2 <= 20 ? 1048576 : 1572864;
+          return estimateRealisticContext(kvEstimate, archMax, modelSizeGB2);
         }
       }
     }
@@ -31242,10 +31269,25 @@ async function queryModelContextSize(baseUrl, modelName) {
     return null;
   }
 }
+function estimateRealisticContext(kvBytesPerToken, archMax, modelSizeGB2) {
+  const totalMemGB = totalmem2() / 1024 ** 3;
+  const usableBytes = totalMemGB * 0.7 * 1024 ** 3;
+  const maxTokens = Math.floor(usableBytes / kvBytesPerToken);
+  let numCtx = Math.max(2048, Math.floor(maxTokens / 1024) * 1024);
+  numCtx = Math.min(numCtx, 131072, archMax);
+  if (modelSizeGB2 && modelSizeGB2 > 0) {
+    const maxKVBytes = modelSizeGB2 * 4 * 1024 ** 3;
+    const budgetCap = Math.max(2048, Math.floor(maxKVBytes / kvBytesPerToken / 1024) * 1024);
+    numCtx = Math.min(numCtx, budgetCap);
+  }
+  return numCtx;
+}
 async function queryOpenAIContextSize(baseUrl, modelName, apiKey) {
   try {
     const models = await fetchOpenAIModels(baseUrl, apiKey);
     const model = models.find((m) => m.name === modelName);
+    if (model?.contextLength)
+      return model.contextLength;
     if (model?.size) {
       const match = model.size.match(/(\d+)K ctx/);
       if (match)
@@ -32155,20 +32197,30 @@ function recommendModel(specs) {
   }
   return QWEN_VARIANTS.find((v) => v.tag === "qwen3.5:cloud");
 }
-function calculateContextWindow(specs, modelSizeGB2) {
+function calculateContextWindow(specs, modelSizeGB2, kvBytesPerToken, archMax) {
   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" };
+  const remaining = Math.max(0, totalAvail - modelSizeGB2);
+  const usableGB = remaining * 0.85;
+  let numCtx;
+  if (kvBytesPerToken && kvBytesPerToken > 0) {
+    const maxTokens = Math.floor(usableGB * 1024 ** 3 / kvBytesPerToken);
+    numCtx = Math.max(2048, Math.floor(maxTokens / 1024) * 1024);
+  } else {
+    const kvEstimate = modelSizeGB2 <= 5 ? 524288 : modelSizeGB2 <= 20 ? 1048576 : 1572864;
+    const maxTokens = Math.floor(usableGB * 1024 ** 3 / kvEstimate);
+    numCtx = Math.max(2048, Math.floor(maxTokens / 1024) * 1024);
+  }
+  numCtx = Math.min(numCtx, 131072);
+  if (archMax && archMax > 0)
+    numCtx = Math.min(numCtx, archMax);
+  if (kvBytesPerToken && kvBytesPerToken > 0 && modelSizeGB2 > 0) {
+    const maxKVBytes = modelSizeGB2 * 4 * 1024 ** 3;
+    const maxTokensFromBudget = Math.floor(maxKVBytes / kvBytesPerToken);
+    const budgetCap = Math.max(2048, Math.floor(maxTokensFromBudget / 1024) * 1024);
+    numCtx = Math.min(numCtx, budgetCap);
+  }
+  const label = numCtx >= 1024 ? `${Math.floor(numCtx / 1024)}K` : String(numCtx);
+  return { numCtx, label };
 }
 function modelSupportsToolCalling(modelName) {
   const lower = modelName.toLowerCase();
@@ -33512,9 +33564,40 @@ function modelSizeGB(models, modelName) {
   const known = QWEN_VARIANTS.find((v) => modelName.includes(v.tag.split(":")[1] ?? ""));
   return known?.sizeGB ?? 4;
 }
-async function createExpandedVariantAsync(baseModel, specs, sizeGB) {
+async function queryModelKVInfo(backendUrl, modelName) {
+  try {
+    const normalized = backendUrl.replace(/\/+$/, "");
+    const res = await fetch(`${normalized}/api/show`, {
+      method: "POST",
+      headers: { "Content-Type": "application/json" },
+      body: JSON.stringify({ name: modelName }),
+      signal: AbortSignal.timeout(5e3)
+    });
+    if (!res.ok)
+      return null;
+    const data = await res.json();
+    if (!data.model_info)
+      return null;
+    const info = data.model_info;
+    const arch = info["general.architecture"];
+    if (!arch)
+      return null;
+    const nLayers = info[`${arch}.block_count`];
+    const nKVHeads = info[`${arch}.attention.head_count_kv`] ?? info[`${arch}.attention.head_count`];
+    const keyDim = info[`${arch}.attention.key_length`];
+    const valDim = info[`${arch}.attention.value_length`] ?? keyDim;
+    const archMax = info[`${arch}.context_length`];
+    if (!nLayers || !nKVHeads || !keyDim || !valDim || !archMax)
+      return null;
+    const kvBytesPerToken = nLayers * nKVHeads * (keyDim + valDim) * 2;
+    return { kvBytesPerToken, archMax };
+  } catch {
+    return null;
+  }
+}
+async function createExpandedVariantAsync(baseModel, specs, sizeGB, kvBytesPerToken, archMax) {
   const customName = expandedModelName(baseModel);
-  const ctx = calculateContextWindow(specs, sizeGB);
+  const ctx = calculateContextWindow(specs, sizeGB, kvBytesPerToken, archMax);
   try {
     const numPredict = Math.min(16384, Math.max(2048, Math.floor(ctx.numCtx * 0.25)));
     const modelfileContent = [
@@ -33537,37 +33620,51 @@ async function createExpandedVariantAsync(baseModel, specs, sizeGB) {
   }
 }
 async function ensureExpandedContext(modelName, backendUrl) {
-  if (modelName.startsWith("open-agents-")) {
-    const specs2 = await detectSystemSpecsAsync();
-    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 = await detectSystemSpecsAsync();
-  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 };
-  }
+  const kvInfo = await queryModelKVInfo(backendUrl, modelName);
   let sizeGB = 4;
   try {
     const models = await fetchOllamaModels(backendUrl);
     sizeGB = modelSizeGB(models, modelName);
   } catch {
   }
-  const ctx = calculateContextWindow(specs, sizeGB);
-  const created = await createExpandedVariantAsync(modelName, specs, sizeGB);
+  const ctx = calculateContextWindow(specs, sizeGB, kvInfo?.kvBytesPerToken, kvInfo?.archMax);
+  if (modelName.startsWith("open-agents-")) {
+    try {
+      const normalized = backendUrl.replace(/\/+$/, "");
+      const showRes = await fetch(`${normalized}/api/show`, {
+        method: "POST",
+        headers: { "Content-Type": "application/json" },
+        body: JSON.stringify({ name: modelName }),
+        signal: AbortSignal.timeout(5e3)
+      });
+      if (showRes.ok) {
+        const showData = await showRes.json();
+        const numCtxMatch = showData.parameters?.match(/num_ctx\s+(\d+)/);
+        const currentNumCtx = numCtxMatch ? parseInt(numCtxMatch[1], 10) : 0;
+        if (currentNumCtx !== ctx.numCtx) {
+          const fromMatch = showData.modelfile?.match(/^FROM\s+(.+)$/m);
+          const baseModel = fromMatch?.[1]?.trim();
+          if (baseModel && !baseModel.startsWith("open-agents-")) {
+            await createExpandedVariantAsync(baseModel, specs, sizeGB, kvInfo?.kvBytesPerToken, kvInfo?.archMax);
+          }
+        }
+      }
+    } catch {
+    }
+    return { model: modelName, created: false, contextLabel: ctx.label, numCtx: ctx.numCtx };
+  }
+  const existing = await checkExpandedVariant(modelName, backendUrl);
+  if (existing === null) {
+    return { model: modelName, created: false, contextLabel: "", numCtx: 0 };
+  }
+  if (typeof existing === "string") {
+    return { model: existing, created: false, contextLabel: ctx.label, numCtx: ctx.numCtx };
+  }
+  const created = await createExpandedVariantAsync(modelName, specs, sizeGB, kvInfo?.kvBytesPerToken, kvInfo?.archMax);
   if (created) {
     return { model: created, created: true, contextLabel: ctx.label, numCtx: ctx.numCtx };
   }
@@ -33610,6 +33707,45 @@ var init_setup = __esm({
   }
 });
 
+// packages/cli/dist/tui/overlay-lock.js
+function isOverlayActive() {
+  return _overlayActive;
+}
+function bufferIfOverlay(stream, data) {
+  if (!_overlayActive)
+    return false;
+  _buffer.push({ stream, data });
+  return true;
+}
+function enterOverlay() {
+  _depth++;
+  _overlayActive = true;
+}
+function leaveOverlay() {
+  _depth = Math.max(0, _depth - 1);
+  if (_depth === 0) {
+    _overlayActive = false;
+    const pending = _buffer;
+    _buffer = [];
+    for (const { stream, data } of pending) {
+      if (stream === "stdout") {
+        process.stdout.write(data);
+      } else {
+        process.stderr.write(data);
+      }
+    }
+  }
+}
+var _overlayActive, _depth, _buffer;
+var init_overlay_lock = __esm({
+  "packages/cli/dist/tui/overlay-lock.js"() {
+    "use strict";
+    _overlayActive = false;
+    _depth = 0;
+    _buffer = [];
+  }
+});
+
 // packages/cli/dist/tui/tui-select.js
 function ansi3(code, text) {
   return isTTY3 ? `\x1B[${code}m${text}\x1B[0m` : text;
@@ -33681,7 +33817,7 @@ function tuiSelect(opts) {
     cursor = first >= 0 ? first : 0;
   }
   const selectChrome = 3;
-  const contentArea = process.stdout.rows ?? 24;
+  const contentArea = opts.availableRows ?? process.stdout.rows ?? 24;
   const maxVisible = opts.maxVisible ?? Math.max(3, contentArea - selectChrome);
   let scrollOffset = 0;
   let lastRenderedLines = 0;
@@ -33704,6 +33840,7 @@ function tuiSelect(opts) {
     }
     stdin.resume();
     process.stdout.write("\x1B[?1049h\x1B[2J\x1B[H\x1B[?25l");
+    enterOverlay();
     function clampScroll(displayList) {
       const cursorPos = displayList.indexOf(cursor);
       if (cursorPos < 0)
@@ -33797,6 +33934,7 @@ function tuiSelect(opts) {
     function cleanup() {
       stdin.removeListener("data", onData);
       process.stdout.removeListener("resize", onResize);
+      leaveOverlay();
       process.stdout.write("\x1B[?1049l\x1B[?25h");
       if (typeof stdin.setRawMode === "function") {
         stdin.setRawMode(hadRawMode ?? false);
@@ -34052,6 +34190,7 @@ var isTTY3, selectColors;
 var init_tui_select = __esm({
   "packages/cli/dist/tui/tui-select.js"() {
     "use strict";
+    init_overlay_lock();
     isTTY3 = process.stdout.isTTY ?? false;
     selectColors = {
       orange: (t) => fg2562(208, t),
@@ -34094,6 +34233,7 @@ function showDropPanel(opts) {
     }
     stdin.resume();
     process.stdout.write("\x1B[?1049h\x1B[2J\x1B[H\x1B[?25h");
+    enterOverlay();
     let inputBuf = "";
     let errorMsg = "";
     function render() {
@@ -34137,6 +34277,7 @@ function showDropPanel(opts) {
     }
     function cleanup() {
       stdin.removeListener("data", onData);
+      leaveOverlay();
       process.stdout.write("\x1B[?1049l\x1B[?25h");
       if (typeof stdin.setRawMode === "function") {
         stdin.setRawMode(hadRawMode ?? false);
@@ -34212,6 +34353,7 @@ var isTTY4, dc;
 var init_drop_panel = __esm({
   "packages/cli/dist/tui/drop-panel.js"() {
     "use strict";
+    init_overlay_lock();
     isTTY4 = process.stdout.isTTY ?? false;
     dc = {
       bold: (t) => ansi4("1", t),
@@ -37180,9 +37322,7 @@ async function handleSlashCommand(input, ctx) {
           return "handled";
         }
         if (arg === "list" || arg === "ls" || arg === "voices") {
-          ctx.setOverlayActive?.(true);
           await handleVoiceList(ctx);
-          ctx.setOverlayActive?.(false);
           return "handled";
         }
         const msg = await ctx.voiceSetModel(arg);
@@ -38268,7 +38408,6 @@ async function handleVoiceMenu(ctx, save, hasLocal) {
       { key: "clone", label: "Clone Voice", detail: "drag & drop audio file" },
       { key: "list", label: "Manage Clone Voices", detail: "rename, play, delete" }
     ];
-    ctx.setOverlayActive?.(true);
     const result = await tuiSelect({
       items,
       title: "Voice Configuration",
@@ -38276,7 +38415,6 @@ async function handleVoiceMenu(ctx, save, hasLocal) {
       availableRows: ctx.availableContentRows?.(),
       skipKeys: ["header-status", "header-config"]
     });
-    ctx.setOverlayActive?.(false);
     if (!result.confirmed || !result.key)
       return;
     switch (result.key) {
@@ -38287,7 +38425,6 @@ async function handleVoiceMenu(ctx, save, hasLocal) {
         continue;
       }
       case "mode": {
-        ctx.setOverlayActive?.(true);
         const modeItems = [
           { key: "chat", label: "Chat", detail: "speaks streamed model text at normal pitch" },
           { key: "action", label: "Action", detail: "speaks tool call narrations with emotion pitch" },
@@ -38300,7 +38437,6 @@ async function handleVoiceMenu(ctx, save, hasLocal) {
           rl: ctx.rl,
           availableRows: ctx.availableContentRows?.()
         });
-        ctx.setOverlayActive?.(false);
         if (modeResult.confirmed && modeResult.key) {
           const mode = modeResult.key;
           ctx.voiceSetMode?.(mode);
@@ -38309,7 +38445,6 @@ async function handleVoiceMenu(ctx, save, hasLocal) {
         continue;
       }
       case "voice": {
-        ctx.setOverlayActive?.(true);
         const voiceItems = [
           { key: "header-onnx", label: selectColors.dim("\u2500\u2500\u2500 ONNX Voices \u2500\u2500\u2500") },
           { key: "glados", label: "GLaDOS", detail: "Portal AI voice (ONNX)" },
@@ -38330,15 +38465,13 @@ async function handleVoiceMenu(ctx, save, hasLocal) {
           availableRows: ctx.availableContentRows?.(),
           skipKeys: ["header-onnx", "header-mlx", "header-clone"]
         });
-        ctx.setOverlayActive?.(false);
         if (voiceResult.confirmed && voiceResult.key) {
-          const msg = await ctx.voiceSetModel(voiceResult.key);
+          await ctx.voiceSetModel(voiceResult.key);
           save({ voice: true, voiceModel: voiceResult.key });
         }
         continue;
       }
       case "system": {
-        ctx.setOverlayActive?.(true);
         const sysItems = [
           { key: "onnx", label: "Piper ONNX", detail: "cross-platform, fast, offline" },
           { key: "luxtts", label: "LuxTTS", detail: "voice cloning, Python venv" },
@@ -38355,10 +38488,8 @@ async function handleVoiceMenu(ctx, save, hasLocal) {
           availableRows: ctx.availableContentRows?.(),
           skipKeys: ["header-import"]
         });
-        ctx.setOverlayActive?.(false);
         if (sysResult.confirmed && sysResult.key) {
           if (sysResult.key === "import-onnx") {
-            ctx.setOverlayActive?.(true);
             const onnxDrop = await showDropPanel({
               title: "Import Piper Voice \u2014 Drop .onnx File",
               instruction: "Drag and drop the .onnx model file",
@@ -38366,11 +38497,9 @@ async function handleVoiceMenu(ctx, save, hasLocal) {
               typeLabel: "ONNX model",
               rl: ctx.rl
             });
-            ctx.setOverlayActive?.(false);
             if (!onnxDrop.confirmed || !onnxDrop.path) {
               continue;
             }
-            ctx.setOverlayActive?.(true);
             const jsonDrop = await showDropPanel({
               title: "Import Piper Voice \u2014 Drop .onnx.json Config",
               instruction: "Drag and drop the .onnx.json config file for the model",
@@ -38378,7 +38507,6 @@ async function handleVoiceMenu(ctx, save, hasLocal) {
               typeLabel: "JSON config",
               rl: ctx.rl
             });
-            ctx.setOverlayActive?.(false);
             if (!jsonDrop.confirmed || !jsonDrop.path) {
               continue;
             }
@@ -38409,7 +38537,6 @@ async function handleVoiceMenu(ctx, save, hasLocal) {
         continue;
       }
       case "clone": {
-        ctx.setOverlayActive?.(true);
         const dropResult = await showDropPanel({
           title: "Voice Clone \u2014 Drop Audio File",
           instruction: "Drag and drop an audio file to use as voice clone reference",
@@ -38417,20 +38544,15 @@ async function handleVoiceMenu(ctx, save, hasLocal) {
           typeLabel: "Audio files",
           rl: ctx.rl
         });
-        ctx.setOverlayActive?.(false);
         if (dropResult.confirmed && dropResult.path && ctx.voiceClone) {
           await ctx.voiceClone(dropResult.path);
           const newFilename = ctx.voiceGetLastClonedFilename?.();
-          ctx.setOverlayActive?.(true);
           await handleVoiceList(ctx, newFilename ?? void 0);
-          ctx.setOverlayActive?.(false);
         }
         continue;
       }
       case "list": {
-        ctx.setOverlayActive?.(true);
         await handleVoiceList(ctx);
-        ctx.setOverlayActive?.(false);
         continue;
       }
     }
@@ -46312,7 +46434,7 @@ var init_braille_spinner = __esm({
 });
 
 // packages/cli/dist/tui/system-metrics.js
-import { loadavg as loadavg2, cpus as cpus2, totalmem as totalmem2, freemem as freemem2, platform as platform4 } from "node:os";
+import { loadavg as loadavg2, cpus as cpus2, totalmem as totalmem3, freemem as freemem2, platform as platform4 } from "node:os";
 import { exec as exec3 } from "node:child_process";
 import { readFile as readFile16 } from "node:fs/promises";
 function formatRate(bytesPerSec) {
@@ -46446,7 +46568,7 @@ function collectCpuRam() {
   const [l1] = loadavg2();
   const cores = cpus2().length;
   const cpuModel = cpus2()[0]?.model ?? "";
-  const totalMem = totalmem2();
+  const totalMem = totalmem3();
   const usedMem = totalMem - freemem2();
   return {
     cpuUtil: Math.min(100, Math.round(l1 / cores * 100)),
@@ -48410,14 +48532,18 @@ function startTask(task, config, repoRoot, voice, stream, taskStores, bruteForce
       }
       backend = new CascadeBackend(config.model, [primaryEndpoint, ...fallbackEndpoints], {
         onSwitch: (_from, to, reason) => {
-          process.stderr.write(`
+          const msg = `
 [cascade] Failover \u2192 ${to.label ?? to.url}: ${reason}
-`);
+`;
+          if (!bufferIfOverlay("stderr", msg))
+            process.stderr.write(msg);
         },
         onProbeResult: (ep, success) => {
           if (success) {
-            process.stderr.write(`[cascade] Primary recovered: ${ep.label ?? ep.url}
-`);
+            const msg = `[cascade] Primary recovered: ${ep.label ?? ep.url}
+`;
+            if (!bufferIfOverlay("stderr", msg))
+              process.stderr.write(msg);
           }
         }
       });
@@ -48657,8 +48783,22 @@ ${entry.fullContent}`
   let streamStartMs = 0;
   let streamTextBuffer = "";
   const contentWrite = (fn) => {
-    if (overlayActive)
+    if (isOverlayActive()) {
+      const origWrite = process.stdout.write;
+      const chunks = [];
+      process.stdout.write = ((chunk) => {
+        chunks.push(chunk);
+        return true;
+      });
+      try {
+        fn();
+      } finally {
+        process.stdout.write = origWrite;
+      }
+      for (const chunk of chunks)
+        bufferIfOverlay("stdout", chunk);
       return;
+    }
     if (statusBar?.isActive) {
       statusBar.beginContentWrite();
       fn();
@@ -49410,8 +49550,22 @@ Rationale: ${proposal.rationale}${provenanceNote}`;
     }
   }
   function writeContent(fn) {
-    if (overlayActive)
+    if (isOverlayActive()) {
+      const origWrite = process.stdout.write;
+      const chunks = [];
+      process.stdout.write = ((chunk) => {
+        chunks.push(chunk);
+        return true;
+      });
+      try {
+        fn();
+      } finally {
+        process.stdout.write = origWrite;
+      }
+      for (const chunk of chunks)
+        bufferIfOverlay("stdout", chunk);
       return;
+    }
     if (statusBar.isActive) {
       statusBar.beginContentWrite();
       try {
@@ -49749,7 +49903,10 @@ Rationale: ${proposal.rationale}${provenanceNote}`;
       return voiceEngine.lastClonedFilename;
     },
     setOverlayActive(active) {
-      overlayActive = active;
+      if (active)
+        enterOverlay();
+      else
+        leaveOverlay();
     },
     streamToggle() {
       streamEnabled = !streamEnabled;
@@ -51439,7 +51596,7 @@ async function runWithTUI(task, config, repoPath) {
     process.exit(1);
   }
 }
-var overlayActive, taskManager;
+var taskManager;
 var init_interactive = __esm({
   "packages/cli/dist/tui/interactive.js"() {
     "use strict";
@@ -51475,7 +51632,7 @@ var init_interactive = __esm({
     init_telegram_bridge();
     init_status_bar();
     init_dist6();
-    overlayActive = false;
+    init_overlay_lock();
     taskManager = new BackgroundTaskManager();
   }
 });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "open-agents-ai",
-  "version": "0.103.99",
+  "version": "0.104.2",
   "description": "AI coding agent powered by open-source models (Ollama/vLLM) — interactive TUI with agentic tool-calling loop",
   "type": "module",
   "main": "./dist/index.js",