diff --git a/.changeset/stream-read-large-files.md b/.changeset/stream-read-large-files.md
new file mode 100644
index 0000000000..185bfcc605
--- /dev/null
+++ b/.changeset/stream-read-large-files.md
@@ -0,0 +1,5 @@
+---
+"@moonshot-ai/kimi-code": patch
+---
+
+Avoid loading entire large text files into memory when reading them.
diff --git a/packages/kaos/src/local.ts b/packages/kaos/src/local.ts
index 71fa16b527..01937729c6 100644
--- a/packages/kaos/src/local.ts
+++ b/packages/kaos/src/local.ts
@@ -408,17 +408,35 @@ export class LocalKaos implements Kaos {
     const resolved = this._resolvePath(path);
     const encoding = options?.encoding ?? 'utf-8';
     const errors = options?.errors ?? 'strict';
-    const buf = await readFile(resolved);
-    const content = decodeTextWithErrors(buf, encoding, errors);
-    const lines = content.split('\n');
-    for (let i = 0; i < lines.length; i++) {
-      const line = lines[i];
-      if (line === undefined) continue;
-      if (i < lines.length - 1) {
-        yield line + '\n';
-      } else if (line !== '') {
-        yield line;
+    const fh = await open(resolved, 'r');
+    try {
+      const chunkSize = 64 * 1024;
+      const buf = Buffer.alloc(chunkSize);
+      let pending = Buffer.alloc(0);
+      while (true) {
+        const { bytesRead } = await fh.read(buf, 0, buf.length, null);
+        if (bytesRead === 0) break;
+        const data =
+          pending.length > 0
+            ? Buffer.concat([pending, buf.subarray(0, bytesRead)])
+            : buf.subarray(0, bytesRead);
+        let lineStart = 0;
+        for (let i = 0; i < data.length; i += 1) {
+          // Split on the LF byte (0x0a). LF/CR are ASCII and never appear
+          // inside a multibyte UTF-8/UTF-16 sequence, so cutting here is
+          // always on a character boundary and safe to decode per line.
+          if (data[i] === 0x0a) {
+            yield decodeTextWithErrors(data.subarray(lineStart, i + 1), encoding, errors);
+            lineStart = i + 1;
+          }
+        }
+        pending = lineStart < data.length ? Buffer.from(data.subarray(lineStart)) : Buffer.alloc(0);
       }
+      if (pending.length > 0) {
+        yield decodeTextWithErrors(pending, encoding, errors);
+      }
+    } finally {
+      await fh.close();
     }
   }
 
diff --git a/packages/kaos/test/local.test.ts b/packages/kaos/test/local.test.ts
index dd8b88b6c6..b0a8dd9193 100644
--- a/packages/kaos/test/local.test.ts
+++ b/packages/kaos/test/local.test.ts
@@ -214,6 +214,106 @@ describe('LocalKaos', () => {
     });
   });
 
+  describe('readLines streaming', () => {
+    it('preserves content exactly across representative line endings', async () => {
+      const fixtures: Array<[string, string]> = [
+        ['multiline', 'line1\nline2\nline3\n'],
+        ['no trailing newline', 'line1\nline2'],
+        ['single line', 'only'],
+        ['single newline', '\n'],
+        ['empty', ''],
+        ['crlf', 'a\r\nb\r\n'],
+        ['lone cr', 'a\rB\n'],
+      ];
+      for (const [name, content] of fixtures) {
+        const filePath = join(tempDir, `${name}.txt`);
+        await kaos.writeText(filePath, content);
+        const lines: string[] = [];
+        for await (const line of kaos.readLines(filePath)) {
+          lines.push(line);
+        }
+        expect(lines.join('')).toBe(content);
+      }
+    });
+
+    it('reads a large multi-chunk file line by line', async () => {
+      const filePath = join(tempDir, 'large.txt');
+      const lineCount = 5000;
+      const content =
+        Array.from({ length: lineCount }, (_, i) => `line ${String(i)} ${'x'.repeat(40)}`).join(
+          '\n',
+        ) + '\n';
+      await kaos.writeText(filePath, content);
+
+      const lines: string[] = [];
+      for await (const line of kaos.readLines(filePath)) {
+        lines.push(line);
+      }
+      expect(lines.length).toBe(lineCount);
+      expect(lines.join('')).toBe(content);
+    });
+
+    it('preserves a multibyte character straddling the 64KiB chunk boundary', async () => {
+      const filePath = join(tempDir, 'boundary.txt');
+      const emoji = '😀';
+      const content = `${'a'.repeat(65535)}${emoji}\n`;
+      await kaos.writeText(filePath, content);
+
+      const lines: string[] = [];
+      for await (const line of kaos.readLines(filePath)) {
+        lines.push(line);
+      }
+      expect(lines).toEqual([content]);
+    });
+
+    describe('errors parameter', () => {
+      // "中" + invalid 0xff + "文" + "\n"
+      const invalidBytes = Buffer.concat([
+        Buffer.from([0xe4, 0xb8, 0xad]),
+        Buffer.from([0xff]),
+        Buffer.from([0xe6, 0x96, 0x87]),
+        Buffer.from([0x0a]),
+      ]);
+
+      it('throws on invalid utf-8 with errors="strict"', async () => {
+        const filePath = join(tempDir, 'invalid-strict.txt');
+        await kaos.writeBytes(filePath, invalidBytes);
+        await expect(
+          (async () => {
+            const lines: string[] = [];
+            for await (const line of kaos.readLines(filePath)) {
+              lines.push(line);
+            }
+            return lines;
+          })(),
+        ).rejects.toThrow();
+      });
+
+      it('replaces invalid bytes with errors="replace"', async () => {
+        const filePath = join(tempDir, 'invalid-replace.txt');
+        await kaos.writeBytes(filePath, invalidBytes);
+        const lines: string[] = [];
+        for await (const line of kaos.readLines(filePath, { errors: 'replace' })) {
+          lines.push(line);
+        }
+        const output = lines.join('');
+        expect(output).toContain('\uFFFD');
+        expect(output).toContain('中');
+        expect(output).toContain('文');
+      });
+
+      it('drops invalid bytes with errors="ignore"', async () => {
+        const filePath = join(tempDir, 'invalid-ignore.txt');
+        await kaos.writeBytes(filePath, invalidBytes);
+        const lines: string[] = [];
+        for await (const line of kaos.readLines(filePath, { errors: 'ignore' })) {
+          lines.push(line);
+        }
+        expect(lines.join('')).toBe('中文\n');
+      });
+    });
+  });
+
   describe('LF preservation', () => {
     it('should not convert LF to CRLF', async () => {
       const filePath = join(tempDir, 'lf.txt');