juliantanx · juliantanx · Jul 1, 2026 · Jul 1, 2026 · Jul 1, 2026
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -5,6 +5,14 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [Unreleased]
+
+### Added
+- **CodeBuddy IDE support** — detect and parse the Tencent CodeBuddy IDE (including the CN variant), whose per-message JSON logs live under `CodeBuddyExtension/Data/**/CodeBuddyIDE/**/history/<session>/<conversation>/messages/`. The existing `codebuddy` JSONL parser only covered `~/.codebuddy/projects`, so IDE usage went undetected. Token usage is read from each conversation's cumulative `statsSnapshot` (cache-miss input, cached input, output). Reported under the existing **CodeBuddy** tool via a new `codebuddy-ide` source; override the path with `AIUSAGE_CODEBUDDY_IDE_PATH`.
+
+### Fixed
+- **CodeBuddy CLI double-counted cached tokens** — the CLI logs usage under both `message.usage` (Anthropic-shaped field names but OpenAI-style semantics, where `input_tokens` already includes cached tokens) and `providerData.rawUsage`. The generic parser assumed Anthropic semantics and added `cache_read_input_tokens` on top of the cache-inclusive `input_tokens`, inflating input by the cached amount (~100× on cache-heavy turns). CodeBuddy now reads the clean `prompt_cache_hit/miss` decomposition from `rawUsage` (falling back to `message.usage` with cache reads subtracted).
+
 ## [1.5.7] - 2026-06-25
 
 ### Added

diff --git a/CHANGELOG.zh-CN.md b/CHANGELOG.zh-CN.md
@@ -5,6 +5,14 @@
 格式基于 [Keep a Changelog](https://keepachangelog.com/zh-CN/1.1.0/)，
 并遵循 [语义化版本控制](https://semver.org/lang/zh-CN/)。
 
+## [未发布]
+
+### 新增
+- **CodeBuddy IDE 支持** — 检测并解析腾讯 CodeBuddy IDE（含 CN 变体）。其逐条消息的 JSON 日志位于 `CodeBuddyExtension/Data/**/CodeBuddyIDE/**/history/<会话>/<对话>/messages/`。原有的 `codebuddy` JSONL 解析器仅覆盖 `~/.codebuddy/projects`，因此 IDE 用量此前无法被检测到。用量数据取自每个对话累计的 `statsSnapshot`（未命中缓存的输入、缓存输入、输出）。归入现有的 **CodeBuddy** 工具，新增 `codebuddy-ide` 数据源；可用 `AIUSAGE_CODEBUDDY_IDE_PATH` 覆盖路径。
+
+### 修复
+- **CodeBuddy CLI 缓存 token 重复计数** — CLI 同时把用量写在 `message.usage`（字段名是 Anthropic 风格，但语义是 OpenAI 风格——`input_tokens` 已包含缓存 token）和 `providerData.rawUsage` 里。通用解析器按 Anthropic 语义处理，在已含缓存的 `input_tokens` 之上又加了一遍 `cache_read_input_tokens`，导致输入被缓存量虚高（缓存密集的轮次可达约 100 倍）。现在 codebuddy 改为读取 `rawUsage` 中干净的 `prompt_cache_hit/miss` 分解（缺失时回退到 `message.usage` 并减去缓存读取部分）。
+
 ## [1.5.7] - 2026-06-25
 
 ### 新增

diff --git a/packages/cli/src/commands/parse-codebuddy.ts b/packages/cli/src/commands/parse-codebuddy.ts
@@ -0,0 +1,246 @@
+import { existsSync, readdirSync, readFileSync } from 'node:fs'
+import { join, basename, dirname } from 'node:path'
+import type { StatsRecord } from '@aiusage/core'
+import { generateRecordId, inferProvider, calculateCost } from '@aiusage/core'
+
+export interface CodeBuddyImportOptions {
+  dataDir: string          // .../CodeBuddyExtension/Data
+  device: string
+  deviceInstanceId: string
+  platform?: string
+  now: number
+  cursor?: number          // Unix ms; skip conversations last active at/before this
+  exchangeRate?: number
+}
+
+export interface CodeBuddyImportResult {
+  records: StatsRecord[]
+  nextCursor: number
+  errors: string[]
+}
+
+function num(value: unknown): number {
+  return typeof value === 'number' && Number.isFinite(value) ? value : 0
+}
+
+function toMs(iso: unknown): number {
+  if (typeof iso !== 'string') return 0
+  const ms = Date.parse(iso)
+  return Number.isFinite(ms) ? ms : 0
+}
+
+/**
+ * The CodeBuddy IDE stores each chat message as its own JSON file:
+ *   <profile>/CodeBuddyIDE/<workspace>/history/<session>/<conversation>/messages/*.json
+ * Only the final assistant message of each agent turn carries usage stats, under a
+ * stringified `extra` field. `statsSnapshot` is a running (conversation-cumulative)
+ * snapshot of input/cache tokens, so we take the latest snapshot per conversation and
+ * emit a single record for it to avoid double-counting across turns.
+ */
+interface MessageFile {
+  role: string
+  message?: string
+  extra?: string
+  createdAt?: string
+}
+
+interface ConversationUsage {
+  ts: number
+  model: string
+  inputTokens: number
+  outputTokens: number
+  cacheReadTokens: number
+  cacheWriteTokens: number
+  thinkingTokens: number
+  cwd?: string
+}
+
+const WORKSPACE_FOLDER_RE = /Workspace Folder:\s*(.+)/
+
+/** Extract the plain text of a user message (its `message` field is stringified JSON). */
+function userMessageText(raw: string): string {
+  try {
+    const inner = JSON.parse(raw) as { content?: Array<{ type?: string; text?: string }> }
+    if (Array.isArray(inner.content)) {
+      return inner.content
+        .filter((block) => block?.type === 'text' && typeof block.text === 'string')
+        .map((block) => block.text)
+        .join('\n')
+    }
+  } catch {
+    // Fall back to the raw string if it is not JSON.
+  }
+  return raw
+}
+
+/** Recursively collect every `messages` directory under a CodeBuddyIDE data root. */
+function findMessagesDirs(root: string, depth = 0): string[] {
+  if (depth > 12) return []
+  let entries
+  try {
+    entries = readdirSync(root, { withFileTypes: true })
+  } catch {
+    return []
+  }
+  const out: string[] = []
+  for (const entry of entries) {
+    if (!entry.isDirectory()) continue
+    const full = join(root, entry.name)
+    if (entry.name === 'messages') {
+      out.push(full)
+      continue
+    }
+    out.push(...findMessagesDirs(full, depth + 1))
+  }
+  return out
+}
+
+/** Read and summarize one conversation's messages directory into a single usage record. */
+function readConversation(messagesDir: string): ConversationUsage | null {
+  let files: string[]
+  try {
+    files = readdirSync(messagesDir).filter((f) => f.endsWith('.json'))
+  } catch {
+    return null
+  }
+  if (files.length === 0) return null
+
+  const parsed: MessageFile[] = []
+  for (const file of files) {
+    try {
+      const data = JSON.parse(readFileSync(join(messagesDir, file), 'utf-8')) as MessageFile
+      if (data && typeof data === 'object') parsed.push(data)
+    } catch {
+      // Skip unreadable/partial message files.
+    }
+  }
+  if (parsed.length === 0) return null
+
+  // Workspace folder (cwd) is embedded in the user prompt's <user_info> block.
+  let cwd: string | undefined
+  for (const msg of parsed) {
+    if (msg.role !== 'user' || typeof msg.message !== 'string') continue
+    const match = WORKSPACE_FOLDER_RE.exec(userMessageText(msg.message))
+    if (match) {
+      cwd = match[1].trim()
+      break
+    }
+  }
+
+  // Latest assistant message carrying a stats snapshot wins (cumulative snapshot).
+  let latest: { ts: number; extra: Record<string, unknown> } | null = null
+  for (const msg of parsed) {
+    if (msg.role !== 'assistant' || typeof msg.extra !== 'string') continue
+    let extra: Record<string, unknown>
+    try {
+      extra = JSON.parse(msg.extra) as Record<string, unknown>
+    } catch {
+      continue
+    }
+    const hasStats = extra.statsSnapshot != null || extra.lastStepInputTokens != null
+    if (!hasStats) continue
+    const ts = toMs(msg.createdAt)
+    if (!latest || ts >= latest.ts) latest = { ts, extra }
+  }
+  if (!latest) return null
+
+  const extra = latest.extra
+  const snapshot = (extra.statsSnapshot ?? {}) as Record<string, unknown>
+
+  // Prefer the cumulative snapshot; fall back to per-step fields when absent.
+  const cacheReadTokens = num(snapshot.cachedInputTokens ?? extra.lastStepCachedInputTokens)
+  const inputTokens = snapshot.cacheMissTokens != null
+    ? num(snapshot.cacheMissTokens)
+    : Math.max(0, num(extra.lastStepInputTokens) - cacheReadTokens)
+  const outputTokens = num(snapshot.lastOutputTokens ?? extra.lastStepOutputTokens)
+  const cacheWriteTokens = num(snapshot.cacheWriteTokens)
+  const thinkingTokens = num(snapshot.thinkingTokens)
+
+  const model = typeof extra.modelId === 'string' && extra.modelId.trim()
+    ? extra.modelId.trim()
+    : 'unknown'
+
+  return {
+    ts: latest.ts,
+    model,
+    inputTokens,
+    outputTokens,
+    cacheReadTokens,
+    cacheWriteTokens,
+    thinkingTokens,
+    cwd,
+  }
+}
+
+export function runParseCodeBuddy(options: CodeBuddyImportOptions): CodeBuddyImportResult {
+  const { dataDir, device, deviceInstanceId, platform, now, cursor, exchangeRate } = options
+  const records: StatsRecord[] = []
+  const errors: string[] = []
+  let nextCursor = cursor ?? 0
+
+  if (!existsSync(dataDir)) {
+    return { records, nextCursor, errors }
+  }
+
+  let messagesDirs: string[]
+  try {
+    messagesDirs = findMessagesDirs(dataDir)
+  } catch (e) {
+    return { records, nextCursor, errors: [String(e)] }
+  }
+
+  for (const messagesDir of messagesDirs) {
+    const conversationDir = dirname(messagesDir)
+    const sessionId = basename(conversationDir)
+    let usage: ConversationUsage | null
+    try {
+      usage = readConversation(messagesDir)
+    } catch (e) {
+      errors.push(`${messagesDir}: ${e instanceof Error ? e.message : e}`)
+      continue
+    }
+    if (!usage) continue
+    if (usage.inputTokens + usage.outputTokens + usage.cacheReadTokens === 0) continue
+
+    // Skip conversations already imported (last activity at or before the cursor).
+    if (cursor && usage.ts <= cursor) continue
+    if (usage.ts > nextCursor) nextCursor = usage.ts
+
+    const provider = inferProvider(usage.model)
+    const tokenArgs = {
+      inputTokens: usage.inputTokens,
+      outputTokens: usage.outputTokens,
+      cacheReadTokens: usage.cacheReadTokens,
+      cacheWriteTokens: usage.cacheWriteTokens,
+      thinkingTokens: usage.thinkingTokens,
+    }
+    const cost = calculateCost(usage.model, tokenArgs, exchangeRate)
+    const recordId = generateRecordId(deviceInstanceId, conversationDir, usage.ts)
+
+    records.push({
+      id: recordId,
+      ts: usage.ts,
+      ingestedAt: now,
+      updatedAt: now,
+      lineOffset: 0,
+      tool: 'codebuddy',
+      model: usage.model,
+      provider,
+      inputTokens: usage.inputTokens,
+      outputTokens: usage.outputTokens,
+      cacheReadTokens: usage.cacheReadTokens,
+      cacheWriteTokens: usage.cacheWriteTokens,
+      thinkingTokens: usage.thinkingTokens,
+      cost,
+      costSource: cost > 0 ? 'pricing' : 'unknown',
+      sessionId,
+      sourceFile: conversationDir,
+      cwd: usage.cwd,
+      device,
+      deviceInstanceId,
+      platform,
+    })
+  }
+
+  return { records, nextCursor, errors }
+}
diff --git a/packages/cli/src/commands/parse.ts b/packages/cli/src/commands/parse.ts
@@ -22,6 +22,7 @@ import { runParseZed } from './parse-zed.js'
 import { runParseKiro } from './parse-kiro.js'
 import { runParseZcode } from './parse-zcode.js'
 import { runParseTrae } from './parse-trae.js'
+import { runParseCodeBuddy } from './parse-codebuddy.js'
 import type { ProgressInfo } from '../progress.js'
 
 interface ParseResult {
@@ -1054,6 +1055,33 @@ export async function runParse(db: Database.Database, filterTool?: string, optio
     onProgress({ phase: 'Parsing SQLite', tool: 'trae', current: 1, total: 1, records: parsedCount, toolCalls: toolCallCount })
   }
 
+  // CodeBuddy IDE: per-message JSON files under CodeBuddyExtension/Data.
+  const codeBuddyIdeDir = getDbPath('codebuddy-ide') ?? ''
+  if ((!filterTool || filterTool === 'codebuddy') && existsSync(codeBuddyIdeDir)) {
+    try {
+      const cbCursor = wm.getCodeBuddyIdeCursor()
+      const result = runParseCodeBuddy({
+        dataDir: codeBuddyIdeDir,
+        device,
+        deviceInstanceId,
+        platform: devicePlatform,
+        now: Date.now(),
+        cursor: cbCursor,
+        exchangeRate,
+      })
+      for (const record of result.records) insertRecord(db, record)
+      parsedCount += result.records.length
+      errors.push(...result.errors)
+      if (result.nextCursor > cbCursor) {
+        wm.setCodeBuddyIdeCursor(result.nextCursor)
+        wm.save()
+      }
+      onProgress({ phase: 'Parsing logs', tool: 'codebuddy', current: 1, total: 1, records: parsedCount, toolCalls: toolCallCount })
+    } catch (e) {
+      errors.push(`${codeBuddyIdeDir}: ${e instanceof Error ? e.message : e}`)
+    }
+  }
+
   // Fix historical records that were parsed before init created state.json.
   // If the current device UUID is known, backfill any records with 'unknown' device_instance_id.
   if (deviceInstanceId !== 'unknown') {

diff --git a/packages/cli/src/discovery.ts b/packages/cli/src/discovery.ts
@@ -352,6 +352,59 @@ function probeCodeBuddy(ctx: ProbeContext): string | null {
   return existsSync(dir) ? dir : null
 }
 
+/** CodeBuddy IDE (Tencent) stores per-message JSON under CodeBuddyExtension/Data. */
+function codeBuddyIdeRoots(ctx: ProbeContext): string[] {
+  const roots: string[] = []
+  if (platform() === 'darwin') {
+    roots.push(join(ctx.home, 'Library', 'Application Support', 'CodeBuddyExtension', 'Data'))
+  } else if (platform() === 'win32') {
+    const appData = ctx.env.APPDATA ?? join(ctx.home, 'AppData', 'Roaming')
+    roots.push(join(appData, 'CodeBuddyExtension', 'Data'))
+  } else {
+    const config = ctx.env.XDG_CONFIG_HOME ?? join(ctx.home, '.config')
+    roots.push(join(config, 'CodeBuddyExtension', 'Data'))
+  }
+  return unique(roots)
+}
+
+function probeCodeBuddyIde(ctx: ProbeContext): string | null {
+  const override = envOverride('codebuddy-ide', ctx.env)
+  if (override) return override
+  const legacy = ctx.legacySources?.['codebuddy-ide']
+  if (legacy) return legacy
+  for (const root of codeBuddyIdeRoots(ctx)) {
+    if (existsSync(root)) return root
+  }
+  return null
+}
+
+/** Count CodeBuddy IDE conversations that contain at least one message file. */
+function countCodeBuddyIdeConversations(dir: string): number {
+  let count = 0
+  const walk = (root: string, depth: number): void => {
+    if (depth > 12) return
+    let entries
+    try {
+      entries = readdirSync(root, { withFileTypes: true })
+    } catch {
+      return
+    }
+    for (const entry of entries) {
+      if (!entry.isDirectory()) continue
+      const full = join(root, entry.name)
+      if (entry.name === 'messages') {
+        try {
+          if (readdirSync(full).some((f) => f.endsWith('.json'))) count++
+        } catch {}
+        continue
+      }
+      walk(full, depth + 1)
+    }
+  }
+  walk(dir, 0)
+  return count
+}
+
 function probeKiro(ctx: ProbeContext): string | null {
   const override = envOverride('kiro', ctx.env)
   if (override) return override
@@ -607,6 +660,7 @@ const TOOL_REGISTRY: readonly ToolEntry[] = [
   { tool: 'gemini', sourceKey: 'gemini', label: 'Gemini CLI', probe: probeGemini },
   { tool: 'kimi', sourceKey: 'kimi', label: 'Kimi Code', probe: probeKimi },
   { tool: 'codebuddy', sourceKey: 'codebuddy', label: 'CodeBuddy', probe: probeCodeBuddy },
+  { tool: 'codebuddy', sourceKey: 'codebuddy-ide', label: 'CodeBuddy (IDE)', probe: probeCodeBuddyIde },
   { tool: 'kiro', sourceKey: 'kiro', label: 'Kiro', probe: probeKiro },
   { tool: 'grok', sourceKey: 'grok', label: 'Grok Build', probe: probeGrok },
   { tool: 'antigravity', sourceKey: 'antigravity', label: 'Antigravity', probe: probeAntigravity },
@@ -654,7 +708,9 @@ export function discoverTools(env: NodeJS.ProcessEnv = process.env): DetectedToo
       try {
         const stat = statSync(detectedPath)
         if (stat.isDirectory()) {
-          if (entry.sourceKey === 'roocode' || entry.sourceKey === 'kilocode') {
+          if (entry.sourceKey === 'codebuddy-ide') {
+            fileCount += countCodeBuddyIdeConversations(detectedPath)
+          } else if (entry.sourceKey === 'roocode' || entry.sourceKey === 'kilocode') {
             fileCount += findJsonFiles(detectedPath).filter((p) => basename(p) === 'ui_messages.json').length
           } else if (entry.sourceKey === 'kelivo') {
             fileCount += findJsonFiles(detectedPath).filter((p) => basename(p) === 'chats.json').length