Add `/compress` command to force a compression of the context (#986)

Related to https://b.corp.google.com/issues/423605555 - I figured this might be a simpler solution to start with, while still also being useful on its own even if we do implement that.

7 files changed, 207 insertions, 18 deletions

diff --git a/packages/cli/src/ui/App.tsx b/packages/cli/src/ui/App.tsx
index 7d8ef463..40935f93 100644
--- a/packages/cli/src/ui/App.tsx
+++ b/packages/cli/src/ui/App.tsx
@@ -170,7 +170,11 @@ const App = ({ config, settings, startupWarnings = [] }: AppProps) => {
     }
   }, [config, addItem]);
 
-  const { handleSlashCommand, slashCommands } = useSlashCommandProcessor(
+  const {
+    handleSlashCommand,
+    slashCommands,
+    pendingHistoryItems: pendingSlashCommandHistoryItems,
+  } = useSlashCommandProcessor(
     config,
     history,
     addItem,
@@ -186,6 +190,7 @@ const App = ({ config, settings, startupWarnings = [] }: AppProps) => {
     showToolDescriptions,
     setQuittingMessages,
   );
+  const pendingHistoryItems = [...pendingSlashCommandHistoryItems];
 
   const { rows: terminalHeight, columns: terminalWidth } = useTerminalSize();
   const { stdin, setRawMode } = useStdin();
@@ -286,18 +291,23 @@ const App = ({ config, settings, startupWarnings = [] }: AppProps) => {
     return editorType as EditorType;
   }, [settings, openEditorDialog]);
 
-  const { streamingState, submitQuery, initError, pendingHistoryItems } =
-    useGeminiStream(
-      config.getGeminiClient(),
-      history,
-      addItem,
-      setShowHelp,
-      config,
-      setDebugMessage,
-      handleSlashCommand,
-      shellModeActive,
-      getPreferredEditor,
-    );
+  const {
+    streamingState,
+    submitQuery,
+    initError,
+    pendingHistoryItems: pendingGeminiHistoryItems,
+  } = useGeminiStream(
+    config.getGeminiClient(),
+    history,
+    addItem,
+    setShowHelp,
+    config,
+    setDebugMessage,
+    handleSlashCommand,
+    shellModeActive,
+    getPreferredEditor,
+  );
+  pendingHistoryItems.push(...pendingGeminiHistoryItems);
   const { elapsedTime, currentLoadingPhrase } =
     useLoadingIndicator(streamingState);
   const showAutoAcceptIndicator = useAutoAcceptIndicator({ config });
diff --git a/packages/cli/src/ui/components/HistoryItemDisplay.tsx b/packages/cli/src/ui/components/HistoryItemDisplay.tsx
index fc1b128d..d99ad503 100644
--- a/packages/cli/src/ui/components/HistoryItemDisplay.tsx
+++ b/packages/cli/src/ui/components/HistoryItemDisplay.tsx
@@ -13,6 +13,7 @@ import { InfoMessage } from './messages/InfoMessage.js';
 import { ErrorMessage } from './messages/ErrorMessage.js';
 import { ToolGroupMessage } from './messages/ToolGroupMessage.js';
 import { GeminiMessageContent } from './messages/GeminiMessageContent.js';
+import { CompressionMessage } from './messages/CompressionMessage.js';
 import { Box } from 'ink';
 import { AboutBox } from './AboutBox.js';
 import { StatsDisplay } from './StatsDisplay.js';
@@ -81,5 +82,8 @@ export const HistoryItemDisplay: React.FC<HistoryItemDisplayProps> = ({
         isFocused={isFocused}
       />
     )}
+    {item.type === 'compression' && (
+      <CompressionMessage compression={item.compression} />
+    )}
   </Box>
 );
diff --git a/packages/cli/src/ui/components/messages/CompressionMessage.tsx b/packages/cli/src/ui/components/messages/CompressionMessage.tsx
new file mode 100644
index 00000000..aaa56149
--- /dev/null
+++ b/packages/cli/src/ui/components/messages/CompressionMessage.tsx
@@ -0,0 +1,48 @@
+/**
+ * @license
+ * Copyright 2025 Google LLC
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+import React from 'react';
+import { Box, Text } from 'ink';
+import { CompressionProps } from '../../types.js';
+import Spinner from 'ink-spinner';
+import { Colors } from '../../colors.js';
+
+export interface CompressionDisplayProps {
+  compression: CompressionProps;
+}
+
+/*
+ * Compression messages appear when the /compress command is ran, and show a loading spinner
+ * while compression is in progress, followed up by some compression stats.
+ */
+export const CompressionMessage: React.FC<CompressionDisplayProps> = ({
+  compression,
+}) => {
+  const text = compression.isPending
+    ? 'Compressing chat history'
+    : `Chat history compressed from ${compression.originalTokenCount} to ${compression.newTokenCount} tokens.`;
+
+  return (
+    <Box flexDirection="row">
+      <Box marginRight={1}>
+        {compression.isPending ? (
+          <Spinner type="dots" />
+        ) : (
+          <Text color={Colors.AccentPurple}>✦</Text>
+        )}
+      </Box>
+      <Box>
+        <Text
+          color={
+            compression.isPending ? Colors.AccentPurple : Colors.AccentGreen
+          }
+        >
+          {text}
+        </Text>
+      </Box>
+    </Box>
+  );
+};
diff --git a/packages/cli/src/ui/hooks/slashCommandProcessor.test.ts b/packages/cli/src/ui/hooks/slashCommandProcessor.test.ts
index c2873bd6..73669651 100644
--- a/packages/cli/src/ui/hooks/slashCommandProcessor.test.ts
+++ b/packages/cli/src/ui/hooks/slashCommandProcessor.test.ts
@@ -62,6 +62,7 @@ import {
   getMCPServerStatus,
   MCPDiscoveryState,
   getMCPDiscoveryState,
+  GeminiClient,
 } from '@gemini-cli/core';
 import { useSessionStats } from '../contexts/SessionContext.js';
 
@@ -100,6 +101,8 @@ describe('useSlashCommandProcessor', () => {
   let mockOpenEditorDialog: ReturnType<typeof vi.fn>;
   let mockPerformMemoryRefresh: ReturnType<typeof vi.fn>;
   let mockSetQuittingMessages: ReturnType<typeof vi.fn>;
+  let mockTryCompressChat: ReturnType<typeof vi.fn>;
+  let mockGeminiClient: GeminiClient;
   let mockConfig: Config;
   let mockCorgiMode: ReturnType<typeof vi.fn>;
   const mockUseSessionStats = useSessionStats as Mock;
@@ -115,8 +118,13 @@ describe('useSlashCommandProcessor', () => {
     mockOpenEditorDialog = vi.fn();
     mockPerformMemoryRefresh = vi.fn().mockResolvedValue(undefined);
     mockSetQuittingMessages = vi.fn();
+    mockTryCompressChat = vi.fn();
+    mockGeminiClient = {
+      tryCompressChat: mockTryCompressChat,
+    } as unknown as GeminiClient;
     mockConfig = {
       getDebugMode: vi.fn(() => false),
+      getGeminiClient: () => mockGeminiClient,
       getSandbox: vi.fn(() => 'test-sandbox'),
       getModel: vi.fn(() => 'test-model'),
       getProjectRoot: vi.fn(() => '/test/dir'),
@@ -944,4 +952,35 @@ Add any other context about the problem here.
       expect(commandResult).toBe(true);
     });
   });
+
+  describe('/compress command', () => {
+    it('should call tryCompressChat(true)', async () => {
+      const { handleSlashCommand } = getProcessor();
+      mockTryCompressChat.mockImplementationOnce(async (force?: boolean) => {
+        // TODO: Check that we have a pending compression item in the history.
+        expect(force).toBe(true);
+        return {
+          originalTokenCount: 100,
+          newTokenCount: 50,
+        };
+      });
+
+      await act(async () => {
+        handleSlashCommand('/compress');
+      });
+      expect(mockGeminiClient.tryCompressChat).toHaveBeenCalledWith(true);
+      expect(mockAddItem).toHaveBeenNthCalledWith(
+        2,
+        expect.objectContaining({
+          type: MessageType.COMPRESSION,
+          compression: {
+            isPending: false,
+            originalTokenCount: 100,
+            newTokenCount: 50,
+          },
+        }),
+        expect.any(Number),
+      );
+    });
+  });
 });
diff --git a/packages/cli/src/ui/hooks/slashCommandProcessor.ts b/packages/cli/src/ui/hooks/slashCommandProcessor.ts
index 861d7bd9..97374e4f 100644
--- a/packages/cli/src/ui/hooks/slashCommandProcessor.ts
+++ b/packages/cli/src/ui/hooks/slashCommandProcessor.ts
@@ -9,6 +9,7 @@ import { type PartListUnion } from '@google/genai';
 import open from 'open';
 import process from 'node:process';
 import { UseHistoryManagerReturn } from './useHistoryManager.js';
+import { useStateAndRef } from './useStateAndRef.js';
 import {
   Config,
   GitService,
@@ -80,6 +81,13 @@ export const useSlashCommandProcessor = (
     return new GitService(config.getProjectRoot());
   }, [config]);
 
+  const pendingHistoryItems: HistoryItemWithoutId[] = [];
+  const [pendingCompressionItemRef, setPendingCompressionItem] =
+    useStateAndRef<HistoryItemWithoutId | null>(null);
+  if (pendingCompressionItemRef.current != null) {
+    pendingHistoryItems.push(pendingCompressionItemRef.current);
+  }
+
   const addMessage = useCallback(
     (message: Message) => {
       // Convert Message to HistoryItemWithoutId
@@ -105,6 +113,11 @@ export const useSlashCommandProcessor = (
           stats: message.stats,
           duration: message.duration,
         };
+      } else if (message.type === MessageType.COMPRESSION) {
+        historyItemContent = {
+          type: 'compression',
+          compression: message.compression,
+        };
       } else {
         historyItemContent = {
           type: message.type as
@@ -641,6 +654,57 @@ Add any other context about the problem here.
           }, 100);
         },
       },
+      {
+        name: 'compress',
+        altName: 'summarize',
+        description: 'Compresses the context by replacing it with a summary.',
+        action: async (_mainCommand, _subCommand, _args) => {
+          if (pendingCompressionItemRef.current !== null) {
+            addMessage({
+              type: MessageType.ERROR,
+              content:
+                'Already compressing, wait for previous request to complete',
+              timestamp: new Date(),
+            });
+            return;
+          }
+          setPendingCompressionItem({
+            type: MessageType.COMPRESSION,
+            compression: {
+              isPending: true,
+            },
+          });
+          try {
+            const compressed = await config!
+              .getGeminiClient()!
+              .tryCompressChat(true);
+            if (compressed) {
+              addMessage({
+                type: MessageType.COMPRESSION,
+                compression: {
+                  isPending: false,
+                  originalTokenCount: compressed.originalTokenCount,
+                  newTokenCount: compressed.newTokenCount,
+                },
+                timestamp: new Date(),
+              });
+            } else {
+              addMessage({
+                type: MessageType.ERROR,
+                content: 'Failed to compress chat history.',
+                timestamp: new Date(),
+              });
+            }
+          } catch (e) {
+            addMessage({
+              type: MessageType.ERROR,
+              content: `Failed to compress chat history: ${e instanceof Error ? e.message : String(e)}`,
+              timestamp: new Date(),
+            });
+          }
+          setPendingCompressionItem(null);
+        },
+      },
     ];
 
     if (config?.getCheckpointEnabled()) {
@@ -767,6 +831,8 @@ Add any other context about the problem here.
     loadHistory,
     addItem,
     setQuittingMessages,
+    pendingCompressionItemRef,
+    setPendingCompressionItem,
   ]);
 
   const handleSlashCommand = useCallback(
@@ -830,5 +896,5 @@ Add any other context about the problem here.
     [addItem, slashCommands, addMessage],
   );
 
-  return { handleSlashCommand, slashCommands };
+  return { handleSlashCommand, slashCommands, pendingHistoryItems };
 };
diff --git a/packages/cli/src/ui/hooks/useGeminiStream.ts b/packages/cli/src/ui/hooks/useGeminiStream.ts
index 920ec490..bff38a2b 100644
--- a/packages/cli/src/ui/hooks/useGeminiStream.ts
+++ b/packages/cli/src/ui/hooks/useGeminiStream.ts
@@ -13,6 +13,7 @@ import {
   ServerGeminiStreamEvent as GeminiEvent,
   ServerGeminiContentEvent as ContentEvent,
   ServerGeminiErrorEvent as ErrorEvent,
+  ServerGeminiChatCompressedEvent,
   getErrorMessage,
   isNodeError,
   MessageSenderType,
@@ -368,11 +369,14 @@ export const useGeminiStream = (
   );
 
   const handleChatCompressionEvent = useCallback(
-    () =>
+    (eventValue: ServerGeminiChatCompressedEvent['value']) =>
       addItem(
         {
           type: 'info',
-          text: `IMPORTANT: this conversation approached the input token limit for ${config.getModel()}. We'll send a compressed context to the model for any future messages.`,
+          text:
+            `IMPORTANT: This conversation approached the input token limit for ${config.getModel()}. ` +
+            `A compressed context will be sent for future messages (compressed from: ` +
+            `${eventValue.originalTokenCount} to ${eventValue.newTokenCount} tokens).`,
         },
         Date.now(),
       ),
@@ -406,7 +410,7 @@ export const useGeminiStream = (
             handleErrorEvent(event.value, userMessageTimestamp);
             break;
           case ServerGeminiEventType.ChatCompressed:
-            handleChatCompressionEvent();
+            handleChatCompressionEvent(event.value);
             break;
           case ServerGeminiEventType.UsageMetadata:
             addUsage(event.value);
diff --git a/packages/cli/src/ui/types.ts b/packages/cli/src/ui/types.ts
index 728b3476..3c0ec616 100644
--- a/packages/cli/src/ui/types.ts
+++ b/packages/cli/src/ui/types.ts
@@ -53,6 +53,12 @@ export interface IndividualToolCallDisplay {
   renderOutputAsMarkdown?: boolean;
 }
 
+export interface CompressionProps {
+  isPending: boolean;
+  originalTokenCount?: number;
+  newTokenCount?: number;
+}
+
 export interface HistoryItemBase {
   text?: string; // Text content for user/gemini/info/error messages
 }
@@ -113,6 +119,11 @@ export type HistoryItemUserShell = HistoryItemBase & {
   text: string;
 };
 
+export type HistoryItemCompression = HistoryItemBase & {
+  type: 'compression';
+  compression: CompressionProps;
+};
+
 // Using Omit<HistoryItem, 'id'> seems to have some issues with typescript's
 // type inference e.g. historyItem.type === 'tool_group' isn't auto-inferring that
 // 'tools' in historyItem.
@@ -127,7 +138,8 @@ export type HistoryItemWithoutId =
   | HistoryItemAbout
   | HistoryItemToolGroup
   | HistoryItemStats
-  | HistoryItemQuit;
+  | HistoryItemQuit
+  | HistoryItemCompression;
 
 export type HistoryItem = HistoryItemWithoutId & { id: number };
 
@@ -140,6 +152,7 @@ export enum MessageType {
   STATS = 'stats',
   QUIT = 'quit',
   GEMINI = 'gemini',
+  COMPRESSION = 'compression',
 }
 
 // Simplified message structure for internal feedback
@@ -172,6 +185,11 @@ export type Message =
       stats: CumulativeStats;
       duration: string;
       content?: string;
+    }
+  | {
+      type: MessageType.COMPRESSION;
+      compression: CompressionProps;
+      timestamp: Date;
     };
 
 export interface ConsoleMessageItem {