diff --git a/CHANGELOG.md b/CHANGELOG.md
index 7a051bd..9133e6b 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -6,6 +6,8 @@ All notable changes to the **OpenCode Go BYOK Provider** extension are documente
 
 ### Added
 
+- **`[Autocomplete]` Chat-prompt suggestions are now opt-in.** VS Code exposes the Copilot Chat prompt box as a virtual `chatSessionInput` document, and the `"**"` provider selector matched it — so ghost text appeared while writing prompts. The provider now only serves real editable code surfaces (`isCompletionDocument` — a code-editor scheme allowlist, so future interactive surfaces are excluded automatically) with an opt-in carve-out for the chat prompt via `opencodego.inlineSuggestionsChatInput`.
+
 - **`[Autocomplete]` Known limitation (follow-up planned):** inline-completion requests are not yet wired into the Go usage tracker's cost accounting (`tracker.record()` only runs in the chat provider path). The panel does track **Suggested / Approved** counts per day (see the usage dashboard entry), but the USD cost of completions is not attributed until a follow-up ships the transport summary from the completion engine. This is tracked as a documented TODO on #136/#138 rather than an oversight.
 
 - **`[Autocomplete]` Inline code suggestions (experimental, #49).** Ghost-text completions while typing, powered by the OpenCode gateway with thinking forced off. Opt-in via `opencodego.inlineSuggestions` (default `false`); model via `opencodego.inlineSuggestionsModel` (default `qwen3.5-plus`, whose `enable_thinking=false` mode is a genuine no-reasoning path — measured ~1.5s time-to-first-token with zero hidden reasoning). Requests are tiny (10 lines before the cursor + a short suffix), debounced 300ms, time out at 3s and abort on the next keystroke. The gateway exposes no FIM endpoint, so completions emulate fill-in-the-middle with FIM tokens over `/chat/completions`. New `src/autocomplete/` module (context, prompt, throttle, engine, provider, registration) with unit tests; `scripts/probe-completion-latency.ts` measures engine latency live. All timing/size knobs are user-tunable: `inlineSuggestionsDebounceMs`, `inlineSuggestionsTimeoutMs`, `inlineSuggestionsMaxTokens`, `inlineSuggestionsPrefixLines`, `inlineSuggestionsSuffixChars`.
diff --git a/package.json b/package.json
index b228d1c..0d4451a 100644
--- a/package.json
+++ b/package.json
@@ -300,6 +300,11 @@
           "default": "qwen3.5-plus",
           "markdownDescription": "Model used for inline code suggestions. Prefer the non-thinking Qwen options — measured: `qwen3.5-plus` with `enable_thinking=false` returns in ~1.5s with zero hidden reasoning, while reasoning models (e.g. `deepseek-v4-flash`) burn 100+ reasoning tokens even with thinking off."
         },
+        "opencodego.inlineSuggestionsChatInput": {
+          "type": "boolean",
+          "default": false,
+          "description": "Also offer inline suggestions inside the GitHub Copilot Chat prompt box while typing prompts. Off by default — completions are meant for code editors."
+        },
         "opencodego.inlineSuggestionsDebounceMs": {
           "type": "number",
           "default": 300,
diff --git a/src/autocomplete/context.ts b/src/autocomplete/context.ts
index 292671a..5db3c70 100644
--- a/src/autocomplete/context.ts
+++ b/src/autocomplete/context.ts
@@ -15,6 +15,35 @@ export {
   DEFAULT_INLINE_MAX_TOKENS as DEFAULT_MAX_TOKENS,
 } from "../config";
 
+/**
+ * VS Code exposes the Copilot Chat prompt box as a virtual editor document
+ * with one of these schemes (see `chatInputSchemes` in the VS Code source:
+ * `workbench/contrib/chat/common/constants.ts`). Inline-completion providers
+ * registered on `"**"` are asked for suggestions there too — ghost text must
+ * never appear while the user is typing a prompt.
+ */
+export const CHAT_INPUT_SCHEMES = new Set(["chatSessionInput", "sessions-chat"]);
+
+/** Whether a document is a chat/interactive prompt box (no completions there). */
+export function isChatInputDocument(uri: { scheme: string }): boolean {
+  return CHAT_INPUT_SCHEMES.has(uri.scheme);
+}
+
+/**
+ * Document schemes that are real editable code surfaces. Inline completions
+ * are only offered there. An ALLOWLIST (instead of only blocking known chat
+ * schemes) means any new interactive surface VS Code introduces in the
+ * future — chat prompt variants, webviews, output, custom editors — is
+ * excluded automatically without a manual update; genuinely new CODE
+ * surfaces are rare.
+ */
+export const CODE_EDITOR_SCHEMES = new Set(["file", "untitled", "git", "vscode-userdata", "vscode-notebook-cell"]);
+
+/** Whether a document is a normal editable code surface. */
+export function isCompletionDocument(uri: { scheme: string }): boolean {
+  return CODE_EDITOR_SCHEMES.has(uri.scheme);
+}
+
 export interface CompletionWindowOptions {
   prefixLines?: number;
   suffixChars?: number;
diff --git a/src/autocomplete/index.ts b/src/autocomplete/index.ts
index 13b9a9b..01a93ce 100644
--- a/src/autocomplete/index.ts
+++ b/src/autocomplete/index.ts
@@ -15,6 +15,7 @@ import {
   COMPLETION_USAGE_KEY,
   COMPLETION_USAGE_MAX_DAYS,
   CONFIG_SECTION,
+  DEFAULT_INLINE_SUGGESTIONS_CHAT_INPUT,
   DEFAULT_INLINE_DEBOUNCE_MS,
   DEFAULT_INLINE_MAX_TOKENS,
   DEFAULT_INLINE_MODEL,
@@ -28,6 +29,7 @@ import {
   INLINE_SUGGESTIONS_SETTING,
   INLINE_SUFFIX_CHARS_SETTING,
   INLINE_TIMEOUT_MS_SETTING,
+  SETTING_INLINE_SUGGESTIONS_CHAT_INPUT,
 } from "../config";
 import { toFiniteNumber } from "../utils";
 import { bumpCompletionUsage, matchesAcceptance, utcDayStart, type CompletionUsageDay } from "./usage";
@@ -146,6 +148,7 @@ export function registerInlineCompletions(context: vscode.ExtensionContext, deps
     resolveMaxTokens: () => readNumberSetting(INLINE_MAX_TOKENS_SETTING, DEFAULT_INLINE_MAX_TOKENS, 16, 1_024),
     resolvePrefixLines: () => readNumberSetting(INLINE_PREFIX_LINES_SETTING, DEFAULT_INLINE_PREFIX_LINES, 1, 100),
     resolveSuffixChars: () => readNumberSetting(INLINE_SUFFIX_CHARS_SETTING, DEFAULT_INLINE_SUFFIX_CHARS, 0, 5_000),
+    resolveChatInputEnabled: () => readSetting(SETTING_INLINE_SUGGESTIONS_CHAT_INPUT, DEFAULT_INLINE_SUGGESTIONS_CHAT_INPUT),
   });
 
   const registration = vscode.languages.registerInlineCompletionItemProvider({ pattern: "**" }, provider);
diff --git a/src/autocomplete/provider.ts b/src/autocomplete/provider.ts
index 82b6382..1119408 100644
--- a/src/autocomplete/provider.ts
+++ b/src/autocomplete/provider.ts
@@ -7,7 +7,7 @@
  */
 
 import * as vscode from "vscode";
-import { buildCompletionWindow } from "./context";
+import { buildCompletionWindow, isChatInputDocument, isCompletionDocument } from "./context";
 import { Debouncer } from "./throttle";
 import type { CompletionContext, CompletionEngine } from "./types";
 
@@ -21,6 +21,8 @@ export interface InlineCompletionProviderOptions {
   isEnabled: () => boolean;
   /** The model to use for suggestions (config-driven). */
   resolveModelId: () => string;
+  /** Whether suggestions are allowed inside the chat prompt box (opt-in). */
+  resolveChatInputEnabled: () => boolean;
   /** Debounce delay in ms before a request is sent (config-driven). */
   resolveDebounceMs: () => number;
   /** Max tokens a completion may produce (config-driven). */
@@ -48,6 +50,14 @@ export class OpenCodeInlineCompletionProvider implements vscode.InlineCompletion
       return Promise.resolve(undefined);
     }
 
+    // Only offer completions in real editable code surfaces. The Copilot
+    // Chat prompt box is a virtual chatSessionInput document — excluded by
+    // default (completions belong in code editors); users can opt in via
+    // opencodego.inlineSuggestionsChatInput.
+    if (!isCompletionDocument(document.uri) && !(isChatInputDocument(document.uri) && this.options.resolveChatInputEnabled())) {
+      return Promise.resolve(undefined);
+    }
+
     // Keep the debounce window live: a config change applies on the next
     // keystroke instead of requiring the provider to be recreated.
     const debounceMs = this.options.resolveDebounceMs();
diff --git a/src/config.ts b/src/config.ts
index f0ca714..f4680b7 100644
--- a/src/config.ts
+++ b/src/config.ts
@@ -56,6 +56,9 @@ export const SETTING_THINKING_MIMO = "thinking.mimo";
 
 export const INLINE_SUGGESTIONS_SETTING = "inlineSuggestions";
 export const INLINE_SUGGESTIONS_MODEL_SETTING = "inlineSuggestionsModel";
+/** Opt-in: also offer completions inside the Copilot Chat prompt box. */
+export const SETTING_INLINE_SUGGESTIONS_CHAT_INPUT = "inlineSuggestionsChatInput";
+export const DEFAULT_INLINE_SUGGESTIONS_CHAT_INPUT = false;
 export const INLINE_DEBOUNCE_MS_SETTING = "inlineSuggestionsDebounceMs";
 export const INLINE_TIMEOUT_MS_SETTING = "inlineSuggestionsTimeoutMs";
 export const INLINE_MAX_TOKENS_SETTING = "inlineSuggestionsMaxTokens";
diff --git a/src/test/autocomplete.test.ts b/src/test/autocomplete.test.ts
index 57f5e79..139e550 100644
--- a/src/test/autocomplete.test.ts
+++ b/src/test/autocomplete.test.ts
@@ -1,6 +1,12 @@
 import assert from "node:assert/strict";
 import { describe, it } from "node:test";
-import { buildCompletionWindow, DEFAULT_PREFIX_LINES, DEFAULT_SUFFIX_CHARS } from "../autocomplete/context";
+import {
+  buildCompletionWindow,
+  DEFAULT_PREFIX_LINES,
+  DEFAULT_SUFFIX_CHARS,
+  isChatInputDocument,
+  isCompletionDocument,
+} from "../autocomplete/context";
 import { buildCompletionPrompt, completionFamily, COMPLETION_SYSTEM_PROMPT } from "../autocomplete/prompt";
 import { cleanCompletion, extractChatCompletionText, parseSseData } from "../autocomplete/engine";
 import { Debouncer } from "../autocomplete/throttle";
@@ -103,6 +109,36 @@ describe("autocomplete — engine parsing", () => {
   });
 });
 
+describe("autocomplete — isChatInputDocument", () => {
+  it("rejects the Copilot Chat prompt box schemes", () => {
+    assert.ok(isChatInputDocument({ scheme: "chatSessionInput" }));
+    assert.ok(isChatInputDocument({ scheme: "sessions-chat" }));
+  });
+
+  it("accepts real code documents", () => {
+    assert.ok(!isChatInputDocument({ scheme: "file" }));
+    assert.ok(!isChatInputDocument({ scheme: "untitled" }));
+  });
+});
+
+describe("autocomplete — isCompletionDocument (code-editor allowlist)", () => {
+  it("accepts real editable code surfaces", () => {
+    assert.ok(isCompletionDocument({ scheme: "file" }));
+    assert.ok(isCompletionDocument({ scheme: "untitled" }));
+    assert.ok(isCompletionDocument({ scheme: "git" }));
+    assert.ok(isCompletionDocument({ scheme: "vscode-userdata" }));
+    assert.ok(isCompletionDocument({ scheme: "vscode-notebook-cell" }));
+  });
+
+  it("excludes chat prompts and other non-code surfaces", () => {
+    assert.ok(!isCompletionDocument({ scheme: "chatSessionInput" }));
+    assert.ok(!isCompletionDocument({ scheme: "sessions-chat" }));
+    assert.ok(!isCompletionDocument({ scheme: "output" }));
+    assert.ok(!isCompletionDocument({ scheme: "webviewPanel" }));
+    assert.ok(!isCompletionDocument({ scheme: "vscode-interactive-input" }));
+  });
+});
+
 describe("autocomplete — Debouncer", () => {
   it("honors a custom delay", async () => {
     const d = new Debouncer(120);