feat(ai): enable automatic context compression

apps/cli/src/ai/agent-stream-runner.ts

@@ -12,6 +12,7 @@ import type {
 } from '@openchatlab/node-runtime'
 import {
   CHART_CAPABILITY_CORE_TOOLS,
+  DEFAULT_CONTEXT_COMPRESSION_CONFIG,
   SkillManager,
   buildPiModel,
   buildSkillMenuWithBuiltinChart,
@@ -93,7 +94,6 @@ export function createCliRunAgentStream(
       skillId,
       enableAutoSkill,
       chartAutoMode,
-      compressionConfig,
       ownerInfo,
       mentionedMembers,
       thinkingLevel,
@@ -110,7 +110,7 @@ export function createCliRunAgentStream(
     }
 
     const llmConfig = llmConfigStore.getDefaultAssistantConfig()
-    const maxToolResultPercent = compressionConfig?.maxToolResultPercent ?? 50
+    const maxToolResultPercent = DEFAULT_CONTEXT_COMPRESSION_CONFIG.maxToolResultPercent ?? 50
     const contextWindow = llmConfig ? (buildPiModel(llmConfig).contextWindow ?? 128000) : 128000
     const maxToolResultTokens = Math.floor(contextWindow * (maxToolResultPercent / 100))
 
@@ -196,15 +196,6 @@ export function createCliRunAgentStream(
       agentTools.push(activateSkillTool as any)
     }
 
-    const resolvedCompression = compressionConfig?.enabled
-      ? {
-          enabled: true as const,
-          tokenThresholdPercent: compressionConfig.tokenThresholdPercent ?? 75,
-          bufferSizePercent: compressionConfig.bufferSizePercent ?? 20,
-          maxToolResultPercent: compressionConfig.maxToolResultPercent,
-        }
-      : undefined
-
     let dataSnapshot: DataSnapshot | undefined
     if (db) {
       try {
@@ -223,7 +214,6 @@ export function createCliRunAgentStream(
       assistantSystemPrompt,
       skillMenu: resolvedSkillMenu,
       skillDef: resolvedSkillDef,
-      compressionConfig: resolvedCompression,
       tools: agentTools,
       llmConfig,
       aiChatManager,

apps/cli/src/ai/agent.ts

@@ -7,6 +7,7 @@
 
 import {
   DEFAULT_MAX_TOOL_ROUNDS,
+  DEFAULT_CONTEXT_COMPRESSION_CONFIG,
   buildPiModel,
   buildPlanGuidance,
   createAnalysisPlanner,
@@ -27,7 +28,6 @@ import {
   type PiMessage,
   type SimpleHistoryMessage,
   type AIChatManager,
-  type CompressionConfig,
   type AgentTool,
   type DataSnapshot,
   type AIServiceConfig,
@@ -49,7 +49,6 @@ export interface RunAgentOptions {
   assistantSystemPrompt?: string
   skillMenu?: string | null
   skillDef?: { name: string; prompt: string }
-  compressionConfig?: CompressionConfig
   tools?: AgentTool[]
   llmConfig: AIServiceConfig | null
   aiChatManager: AIChatManager
@@ -72,7 +71,6 @@ export async function runServerAgent(options: RunAgentOptions): Promise<void> {
     assistantSystemPrompt,
     skillMenu,
     skillDef,
-    compressionConfig,
     tools = [],
     llmConfig,
     aiChatManager,
@@ -123,7 +121,7 @@ export async function runServerAgent(options: RunAgentOptions): Promise<void> {
     systemPrompt,
   })
 
-  if (compressionConfig?.enabled && historyLeafMessageId === undefined) {
+  if (historyLeafMessageId === undefined) {
     const llmAdapter = createCompressionLlmAdapter({
       piModel,
       apiKey: llmConfig.apiKey,
@@ -131,7 +129,7 @@ export async function runServerAgent(options: RunAgentOptions): Promise<void> {
     })
     const compressionResult = await checkAndCompress(
       aiChatId,
-      compressionConfig,
+      DEFAULT_CONTEXT_COMPRESSION_CONFIG,
       systemPrompt,
       llmAdapter,
       aiChatManager,
@@ -148,7 +146,7 @@ export async function runServerAgent(options: RunAgentOptions): Promise<void> {
         },
       })
     }
-  } else if (compressionConfig?.enabled && historyLeafMessageId !== undefined) {
+  } else {
     aiLogger?.info?.('Compression', 'Skipping compression for edited branch request', {
       aiChatId,
       historyLeafMessageId,