diff --git a/CHANGELOG.md b/CHANGELOG.md
index 59f5a036a..7645f0e30 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -39,6 +39,7 @@
 - Auth: default billing disable backoff to 5h (doubling, 24h cap) and surface disabled/cooldown profiles in `models list` + doctor. (#486) — thanks @steipete
 - Commands: harden slash command registry and list text-only commands in `/commands`.
 - Models/Auth: show per-agent auth candidates in `/model status`, and add `clawdbot models auth order {get,set,clear}` (per-agent auth rotation overrides). — thanks @steipete
+- Telegram: keep streamMode draft-only; avoid forcing block streaming. (#619) — thanks @rubyrunsstuff
 - Debugging: add raw model stream logging flags and document gateway watch mode.
 - Gateway: decode dns-sd escaped UTF-8 in discovery output and show scan progress immediately. — thanks @steipete
 - Agent: add claude-cli/opus-4.5 runner via Claude CLI with resume support (tools disabled).
diff --git a/src/agents/pi-embedded-block-chunker.ts b/src/agents/pi-embedded-block-chunker.ts
index 7254bbd9f..3bf95c14b 100644
--- a/src/agents/pi-embedded-block-chunker.ts
+++ b/src/agents/pi-embedded-block-chunker.ts
@@ -221,6 +221,10 @@ export class EmbeddedBlockChunker {
       if (sentenceIdx >= minChars) return { index: sentenceIdx };
     }
 
+    if (preference === "newline" && buffer.length < maxChars) {
+      return { index: -1 };
+    }
+
     for (let i = window.length - 1; i >= minChars; i--) {
       if (/\s/.test(window[i]) && isSafeFenceBreak(fenceSpans, i)) {
         return { index: i };
diff --git a/src/auto-reply/reply.block-streaming.test.ts b/src/auto-reply/reply.block-streaming.test.ts
index 8cb8ab767..9a13e65cc 100644
--- a/src/auto-reply/reply.block-streaming.test.ts
+++ b/src/auto-reply/reply.block-streaming.test.ts
@@ -273,4 +273,45 @@ describe("block streaming", () => {
       expect(sawAbort).toBe(true);
     });
   });
+
+  it("does not enable block streaming for telegram streamMode block", async () => {
+    await withTempHome(async (home) => {
+      const onBlockReply = vi.fn().mockResolvedValue(undefined);
+
+      const impl = async () => ({
+        payloads: [{ text: "final" }],
+        meta: {
+          durationMs: 5,
+          agentMeta: { sessionId: "s", provider: "p", model: "m" },
+        },
+      });
+      piEmbeddedMock.runEmbeddedPiAgent.mockImplementation(impl);
+
+      const res = await getReplyFromConfig(
+        {
+          Body: "ping",
+          From: "+1004",
+          To: "+2000",
+          MessageSid: "msg-126",
+          Provider: "telegram",
+        },
+        {
+          onBlockReply,
+        },
+        {
+          agents: {
+            defaults: {
+              model: "anthropic/claude-opus-4-5",
+              workspace: path.join(home, "clawd"),
+            },
+          },
+          telegram: { allowFrom: ["*"], streamMode: "block" },
+          session: { store: path.join(home, "sessions.json") },
+        },
+      );
+
+      expect(res?.text).toBe("final");
+      expect(onBlockReply).not.toHaveBeenCalled();
+    });
+  });
 });
diff --git a/src/auto-reply/reply/block-streaming.ts b/src/auto-reply/reply/block-streaming.ts
index 17f2cda43..da60e1ce9 100644
--- a/src/auto-reply/reply/block-streaming.ts
+++ b/src/auto-reply/reply/block-streaming.ts
@@ -58,9 +58,10 @@ export function resolveBlockStreamingChunking(
     Math.floor(chunkCfg?.maxChars ?? DEFAULT_BLOCK_STREAM_MAX),
   );
   const maxChars = Math.max(1, Math.min(maxRequested, textLimit));
+  const minFallback = DEFAULT_BLOCK_STREAM_MIN;
   const minRequested = Math.max(
     1,
-    Math.floor(chunkCfg?.minChars ?? DEFAULT_BLOCK_STREAM_MIN),
+    Math.floor(chunkCfg?.minChars ?? minFallback),
   );
   const minChars = Math.min(minRequested, maxChars);
   const breakPreference =
@@ -80,7 +81,7 @@ export function resolveBlockStreamingCoalescing(
     maxChars: number;
     breakPreference: "paragraph" | "newline" | "sentence";
   },
-): BlockStreamingCoalescing {
+): BlockStreamingCoalescing | undefined {
   const providerKey = normalizeChunkProvider(provider);
   const textLimit = resolveTextChunkLimit(cfg, providerKey, accountId);
   const normalizedAccountId = normalizeAccountId(accountId);
diff --git a/src/telegram/bot.ts b/src/telegram/bot.ts
index c5979ec76..aaba18307 100644
--- a/src/telegram/bot.ts
+++ b/src/telegram/bot.ts
@@ -826,6 +826,12 @@ export function createTelegramBot(opts: TelegramBotOptions) {
         onReplyStart: sendTyping,
       });
 
+    const disableBlockStreaming =
+      Boolean(draftStream) ||
+      (typeof telegramCfg.blockStreaming === "boolean"
+        ? !telegramCfg.blockStreaming
+        : undefined);
+
     const { queuedFinal } = await dispatchReplyFromConfig({
       ctx: ctxPayload,
       cfg,
@@ -841,11 +847,7 @@ export function createTelegramBot(opts: TelegramBotOptions) {
               if (payload.text) draftStream.update(payload.text);
             }
           : undefined,
-        disableBlockStreaming:
-          Boolean(draftStream) ||
-          (typeof telegramCfg.blockStreaming === "boolean"
-            ? !telegramCfg.blockStreaming
-            : undefined),
+        disableBlockStreaming,
       },
     });
     markDispatchIdle();