diff --git a/src/infra/outbound/deliver.test.ts b/src/infra/outbound/deliver.test.ts
index b9c59f0e3..71acf883b 100644
--- a/src/infra/outbound/deliver.test.ts
+++ b/src/infra/outbound/deliver.test.ts
@@ -160,6 +160,30 @@ describe("deliverOutboundPayloads", () => {
     });
   });
 
+  it("clamps telegram text chunk size to protocol max even with higher config", async () => {
+    const sendTelegram = vi.fn().mockResolvedValue({ messageId: "m1", chatId: "c1" });
+    const cfg: OpenClawConfig = {
+      channels: { telegram: { botToken: "tok-1", textChunkLimit: 10_000 } },
+    };
+    const text = "<".repeat(3_000);
+    await withEnvAsync({ TELEGRAM_BOT_TOKEN: "" }, async () => {
+      await deliverOutboundPayloads({
+        cfg,
+        channel: "telegram",
+        to: "123",
+        payloads: [{ text }],
+        deps: { sendTelegram },
+      });
+    });
+
+    expect(sendTelegram.mock.calls.length).toBeGreaterThan(1);
+    const sentHtmlChunks = sendTelegram.mock.calls
+      .map((call) => call[1])
+      .filter((message): message is string => typeof message === "string");
+    expect(sentHtmlChunks.length).toBeGreaterThan(1);
+    expect(sentHtmlChunks.every((message) => message.length <= 4096)).toBe(true);
+  });
+
   it("keeps payload replyToId across all chunked telegram sends", async () => {
     const sendTelegram = vi.fn().mockResolvedValue({ messageId: "m1", chatId: "c1" });
     await withEnvAsync({ TELEGRAM_BOT_TOKEN: "" }, async () => {
diff --git a/src/infra/outbound/deliver.ts b/src/infra/outbound/deliver.ts
index 76ea0e787..9002245ab 100644
--- a/src/infra/outbound/deliver.ts
+++ b/src/infra/outbound/deliver.ts
@@ -40,6 +40,7 @@ export type { NormalizedOutboundPayload } from "./payloads.js";
 export { normalizeOutboundPayloads } from "./payloads.js";
 
 const log = createSubsystemLogger("outbound/deliver");
+const TELEGRAM_TEXT_LIMIT = 4096;
 
 type SendMatrixMessage = (
   to: string,
@@ -314,11 +315,15 @@ async function deliverOutboundPayloadsCore(
     silent: params.silent,
     mediaLocalRoots,
   });
-  const textLimit = handler.chunker
+  const configuredTextLimit = handler.chunker
     ? resolveTextChunkLimit(cfg, channel, accountId, {
         fallbackLimit: handler.textChunkLimit,
       })
     : undefined;
+  const textLimit =
+    channel === "telegram" && typeof configuredTextLimit === "number"
+      ? Math.min(configuredTextLimit, TELEGRAM_TEXT_LIMIT)
+      : configuredTextLimit;
   const chunkMode = handler.chunker ? resolveChunkMode(cfg, channel, accountId) : "length";
   const isSignalChannel = channel === "signal";
   const signalTableMode = isSignalChannel
diff --git a/src/telegram/format.ts b/src/telegram/format.ts
index f919a917f..acefd8f75 100644
--- a/src/telegram/format.ts
+++ b/src/telegram/format.ts
@@ -241,6 +241,58 @@ export function renderTelegramHtmlText(
   return markdownToTelegramHtml(text, { tableMode: options.tableMode });
 }
 
+function splitTelegramChunkByHtmlLimit(
+  chunk: MarkdownIR,
+  htmlLimit: number,
+  renderedHtmlLength: number,
+): MarkdownIR[] {
+  const currentTextLength = chunk.text.length;
+  if (currentTextLength <= 1) {
+    return [chunk];
+  }
+  const proportionalLimit = Math.floor(
+    (currentTextLength * htmlLimit) / Math.max(renderedHtmlLength, 1),
+  );
+  const candidateLimit = Math.min(currentTextLength - 1, proportionalLimit);
+  const splitLimit =
+    Number.isFinite(candidateLimit) && candidateLimit > 0
+      ? candidateLimit
+      : Math.max(1, Math.floor(currentTextLength / 2));
+  const split = chunkMarkdownIR(chunk, splitLimit);
+  if (split.length > 1) {
+    return split;
+  }
+  return chunkMarkdownIR(chunk, Math.max(1, Math.floor(currentTextLength / 2)));
+}
+
+function renderTelegramChunksWithinHtmlLimit(
+  ir: MarkdownIR,
+  limit: number,
+): TelegramFormattedChunk[] {
+  const normalizedLimit = Math.max(1, Math.floor(limit));
+  const pending = chunkMarkdownIR(ir, normalizedLimit);
+  const rendered: TelegramFormattedChunk[] = [];
+  while (pending.length > 0) {
+    const chunk = pending.shift();
+    if (!chunk) {
+      continue;
+    }
+    const html = wrapFileReferencesInHtml(renderTelegramHtml(chunk));
+    if (html.length <= normalizedLimit || chunk.text.length <= 1) {
+      rendered.push({ html, text: chunk.text });
+      continue;
+    }
+    const split = splitTelegramChunkByHtmlLimit(chunk, normalizedLimit, html.length);
+    if (split.length <= 1) {
+      // Worst-case safety: avoid retry loops, deliver the chunk as-is.
+      rendered.push({ html, text: chunk.text });
+      continue;
+    }
+    pending.unshift(...split);
+  }
+  return rendered;
+}
+
 export function markdownToTelegramChunks(
   markdown: string,
   limit: number,
@@ -253,11 +305,7 @@ export function markdownToTelegramChunks(
     blockquotePrefix: "",
     tableMode: options.tableMode,
   });
-  const chunks = chunkMarkdownIR(ir, limit);
-  return chunks.map((chunk) => ({
-    html: wrapFileReferencesInHtml(renderTelegramHtml(chunk)),
-    text: chunk.text,
-  }));
+  return renderTelegramChunksWithinHtmlLimit(ir, limit);
 }
 
 export function markdownToTelegramHtmlChunks(markdown: string, limit: number): string[] {
diff --git a/src/telegram/format.wrap-md.test.ts b/src/telegram/format.wrap-md.test.ts
index d059f950c..8d003eba3 100644
--- a/src/telegram/format.wrap-md.test.ts
+++ b/src/telegram/format.wrap-md.test.ts
@@ -158,6 +158,14 @@ describe("markdownToTelegramChunks - file reference wrapping", () => {
     expect(chunks[0].html).toContain("<code>README.md</code>");
     expect(chunks[0].html).toContain("<code>backup.sh</code>");
   });
+
+  it("keeps rendered html chunks within the provided limit", () => {
+    const input = "<".repeat(1500);
+    const chunks = markdownToTelegramChunks(input, 512);
+    expect(chunks.length).toBeGreaterThan(1);
+    expect(chunks.map((chunk) => chunk.text).join("")).toBe(input);
+    expect(chunks.every((chunk) => chunk.html.length <= 512)).toBe(true);
+  });
 });
 
 describe("edge cases", () => {