diff --git a/app/src/main/java/com/rainy/token/ui/heatmap/HeatmapViewModel.kt b/app/src/main/java/com/rainy/token/ui/heatmap/HeatmapViewModel.kt
new file mode 100644
index 0000000..b203c69
--- /dev/null
+++ b/app/src/main/java/com/rainy/token/ui/heatmap/HeatmapViewModel.kt
@@ -0,0 +1,322 @@
+package com.rainy.token.ui.heatmap
+
+import androidx.lifecycle.ViewModel
+import androidx.lifecycle.viewModelScope
+import com.rainy.token.data.local.UsageCache
+import com.rainy.token.data.local.UsageRecord
+import com.rainy.token.data.repository.CredentialRepository
+import com.rainy.token.domain.model.Credential
+import com.rainy.token.domain.service.ServiceType
+import dagger.hilt.android.lifecycle.HiltViewModel
+import kotlinx.coroutines.Dispatchers
+import kotlinx.coroutines.flow.MutableStateFlow
+import kotlinx.coroutines.flow.StateFlow
+import kotlinx.coroutines.flow.asStateFlow
+import kotlinx.coroutines.flow.update
+import kotlinx.coroutines.launch
+import kotlinx.coroutines.withContext
+import javax.inject.Inject
+import javax.inject.Provider
+
+// ── 公共数据类 ──────────────────────────────────────────────
+
+/** 热力图视图模式 */
+enum class HeatmapViewMode { DAILY, WEEKLY, CUMULATIVE }
+
+/** 每日 / 累计视图的单格数据 */
+data class HeatmapDayData(
+    val dayTs: Long,   // UTC 日期时间戳（即天分桶的时间戳）
+    val tokens: Long,  // 当天 token 数（DAILY）或累计 token 数（CUMULATIVE）
+    val level: Int,    // 0-5 颜色等级
+)
+
+/** 每周视图的单格数据 */
+data class HeatmapWeekData(
+    val weekStartTs: Long, // 该周起始日时间戳
+    val tokens: Long,      // 该周 7 天累计 token 数
+    val level: Int,        // 0-5 颜色等级
+)
+
+/** UI 状态 */
+data class HeatmapUiState(
+    val loading: Boolean = true,
+    val viewMode: HeatmapViewMode = HeatmapViewMode.DAILY,
+    val dailyData: List<HeatmapDayData> = emptyList(),
+    val weeklyData: List<HeatmapWeekData> = emptyList(),
+    val cumulativeData: List<HeatmapDayData> = emptyList(),
+    val colorLevels: IntArray = IntArray(6),  // [0, p25, p50, p75, p95, max]
+) {
+    override fun equals(other: Any?): Boolean {
+        if (this === other) return true
+        if (other !is HeatmapUiState) return false
+        return loading == other.loading &&
+            viewMode == other.viewMode &&
+            dailyData == other.dailyData &&
+            weeklyData == other.weeklyData &&
+            cumulativeData == other.cumulativeData &&
+            colorLevels.contentEquals(other.colorLevels)
+    }
+
+    override fun hashCode(): Int {
+        var result = loading.hashCode()
+        result = 31 * result + viewMode.hashCode()
+        result = 31 * result + dailyData.hashCode()
+        result = 31 * result + weeklyData.hashCode()
+        result = 31 * result + cumulativeData.hashCode()
+        result = 31 * result + colorLevels.contentHashCode()
+        return result
+    }
+}
+
+/**
+ * 分位阈值与颜色等级映射。
+ *
+ * thresholds[0..5] 存 Long 阈值：
+ * - Level 0: value == 0
+ * - Level 1: 0 < value ≤ P25
+ * - Level 2: P25 < value ≤ P50
+ * - Level 3: P50 < value ≤ P75
+ * - Level 4: P75 < value ≤ P95
+ * - Level 5: value > P95
+ *
+ * 边界情况（非零天数 < 2 或所有非零值相同）时，所有非零值设为 Level 3。
+ */
+internal data class QuantileLevels(val thresholds: LongArray) {
+    fun getColorLevel(value: Long): Int {
+        if (value <= 0L) return 0
+        for (level in 1..5) {
+            if (value <= thresholds[level]) return level
+        }
+        return 5
+    }
+
+    fun toIntArray(): IntArray = IntArray(6) { idx ->
+        val v = thresholds[idx]
+        if (v > Int.MAX_VALUE) Int.MAX_VALUE else v.toInt()
+    }
+
+    override fun equals(other: Any?): Boolean {
+        if (this === other) return true
+        if (other !is QuantileLevels) return false
+        return thresholds.contentEquals(other.thresholds)
+    }
+
+    override fun hashCode(): Int = thresholds.contentHashCode()
+}
+
+// ── ViewModel ──────────────────────────────────────────────
+
+@HiltViewModel
+class HeatmapViewModel @Inject constructor(
+    private val cacheProvider: Provider<UsageCache>,
+    private val credentialRepository: CredentialRepository,
+) : ViewModel() {
+
+    private val _uiState = MutableStateFlow(HeatmapUiState())
+    val uiState: StateFlow<HeatmapUiState> = _uiState.asStateFlow()
+
+    private var loaded = false
+    private var allViewsResult: AllViewsResult? = null
+
+    /**
+     * Composable 层在 LaunchedEffect 中调用一次即可。
+     * 三种视图数据全部在此次加载中计算好，切换视图不需要重新加载。
+     */
+    fun load() {
+        if (loaded) return
+        loaded = true
+        viewModelScope.launch { loadInternal() }
+    }
+
+    /** 切换视图模式（仅改 State，不重新加载数据） */
+    fun setViewMode(mode: HeatmapViewMode) {
+        val result = allViewsResult
+        val levels = result?.let {
+            when (mode) {
+                HeatmapViewMode.DAILY -> it.dailyLevels
+                HeatmapViewMode.WEEKLY -> it.weeklyLevels
+                HeatmapViewMode.CUMULATIVE -> it.cumulativeLevels
+            }
+        }
+        _uiState.update { it.copy(viewMode = mode, colorLevels = levels?.toIntArray() ?: it.colorLevels) }
+    }
+
+    // ── 内部逻辑 ────────────────────────────────────────────
+
+    private suspend fun workspaceId(): String? {
+        val c = credentialRepository.get(ServiceType.OPENCODE_GO)
+        return (c as? Credential.SessionCredential)?.workspaceId?.takeIf { it.isNotBlank() }
+    }
+
+    private suspend fun loadInternal() {
+        try {
+            val wid = workspaceId() ?: run {
+                _uiState.update { it.copy(loading = false) }
+                return
+            }
+            val cache = cacheProvider.get()
+
+            // 所有重操作放在 Dispatchers.Default 上
+            val result = withContext(Dispatchers.Default) {
+                val records = cache.getRecords(wid)
+                computeAllViews(records)
+            }
+
+            allViewsResult = result
+
+            _uiState.update {
+                it.copy(
+                    loading = false,
+                    dailyData = result.dailyData,
+                    weeklyData = result.weeklyData,
+                    cumulativeData = result.cumulativeData,
+                    colorLevels = result.dailyLevels.toIntArray(),
+                )
+            }
+        } catch (e: Exception) {
+            _uiState.update { it.copy(loading = false) }
+        }
+    }
+
+    // ── 聚合计算 ────────────────────────────────────────────
+
+    companion object {
+        private const val DAY_MS = 86_400_000L
+        private const val RANGE_DAYS = 365
+
+        /**
+         * 从全部记录计算三种视图数据 + 颜色等级阈值。
+         * 返回 (dailyData, weeklyData, cumulativeData, colorLevels)
+         */
+        internal fun computeAllViews(records: List<UsageRecord>): AllViewsResult {
+            val now = System.currentTimeMillis()
+
+            // 1. 按天分桶聚合 token 数
+            val todayTs = now / DAY_MS * DAY_MS
+            val rangeStartTs = todayTs - (RANGE_DAYS - 1) * DAY_MS
+
+            // 聚合每一天的 token
+            val dayTokenMap = HashMap<Long, Long>()
+            for (r in records) {
+                val dayTs = r.timeCreated / DAY_MS * DAY_MS
+                if (dayTs < rangeStartTs || dayTs > todayTs) continue
+                val tokens = r.inputTokens + r.cacheReadTokens + r.reasoningTokens + r.outputTokens
+                dayTokenMap.merge(dayTs, tokens, Long::plus)
+            }
+
+            // 2. 构建完整的每日数据列表（365 天，从 rangeStartTs 到 todayTs）
+            val dailyRaw = ArrayList<HeatmapDayData>(RANGE_DAYS)
+            for (i in 0 until RANGE_DAYS) {
+                val ts = rangeStartTs + i * DAY_MS
+                val tokens = dayTokenMap[ts] ?: 0L
+                dailyRaw.add(HeatmapDayData(dayTs = ts, tokens = tokens, level = 0))
+            }
+
+            // 3. 计算每日视图颜色等级
+            val dailyLevels = computeQuantileLevels(dailyRaw.map { it.tokens })
+            val dailyData = dailyRaw.map { it.copy(level = dailyLevels.getColorLevel(it.tokens)) }
+
+            // 4. 每周视图：从最早日数据开始，每 7 天一组
+            val (weeklyData, weeklyLevels) = buildWeeklyData(dailyRaw)
+
+            // 5. 累计视图：从第一天到当天的累计总和
+            var cumulativeSum = 0L
+            val cumulativeRaw = ArrayList<HeatmapDayData>(RANGE_DAYS)
+            for (d in dailyRaw) {
+                cumulativeSum += d.tokens
+                cumulativeRaw.add(HeatmapDayData(dayTs = d.dayTs, tokens = cumulativeSum, level = 0))
+            }
+            val cumulativeLevels = computeQuantileLevels(cumulativeRaw.map { it.tokens })
+            val cumulativeData = cumulativeRaw.map { it.copy(level = cumulativeLevels.getColorLevel(it.tokens)) }
+
+            // 保存三种视图各自的分位阈值，供切换视图时更新图例
+            return AllViewsResult(dailyData, weeklyData, cumulativeData, dailyLevels, weeklyLevels, cumulativeLevels)
+        }
+
+        /**
+         * 构建每周视图数据。
+         * 从 rangeStartTs 对应的那天开始，每 7 天为一组。
+         */
+        private fun buildWeeklyData(dailyRaw: List<HeatmapDayData>): Pair<List<HeatmapWeekData>, QuantileLevels> {
+            val weeks = ArrayList<HeatmapWeekData>()
+            var i = 0
+            while (i < 52 * 7 && i < dailyRaw.size) {
+                val weekEnd = minOf(i + 7, dailyRaw.size)
+                var weekTokens = 0L
+                for (j in i until weekEnd) {
+                    weekTokens += dailyRaw[j].tokens
+                }
+                weeks.add(HeatmapWeekData(
+                    weekStartTs = dailyRaw[i].dayTs,
+                    tokens = weekTokens,
+                    level = 0,
+                ))
+                i += 7
+            }
+
+            // 计算每周视图颜色等级
+            val weekLevels = computeQuantileLevels(weeks.map { it.tokens })
+            return weeks.map { it.copy(level = weekLevels.getColorLevel(it.tokens)) } to weekLevels
+        }
+
+        // ── 分位数颜色等级计算 ──────────────────────────────
+
+        /**
+         * 使用 nearest-rank 分位数法计算 P25/P50/P75/P95 阈值。
+         *
+         * 边界情况：
+         * - 非零天数 < 2 → 全部设为 Level 3
+         * - 所有非零值相同 → 全部设为 Level 3
+         *
+         * 返回 [QuantileLevels]，内部 thresholds 为 LongArray(6)：
+         * index 0 = 0（Level 0），index 1-5 分别存 P25/P50/P75/P95/MaxValue。
+         */
+        internal fun computeQuantileLevels(values: List<Long>): QuantileLevels {
+            // 收集所有非零值
+            val nonZero = values.filter { it > 0L }
+
+            // 边界：非零值 < 2 或全部相同 → 全部设为 Level 3
+            if (nonZero.size < 2 || nonZero.toSet().size == 1) {
+                // thresholds[3] = Long.MAX_VALUE 使得 0 < value ≤ MAX → Level 3
+                val t = LongArray(6)
+                t[3] = Long.MAX_VALUE
+                return QuantileLevels(t)
+            }
+
+            val sorted = nonZero.sorted()
+            val n = sorted.size
+
+            val p25 = nearestRank(sorted, n, 0.25)
+            val p50 = nearestRank(sorted, n, 0.50)
+            val p75 = nearestRank(sorted, n, 0.75)
+            val p95 = nearestRank(sorted, n, 0.95)
+
+            // LongArray(6): [0, p25, p50, p75, p95, MaxValue]
+            val thresholds = LongArray(6)
+            thresholds[0] = 0L
+            thresholds[1] = p25
+            thresholds[2] = p50
+            thresholds[3] = p75
+            thresholds[4] = p95
+            thresholds[5] = Long.MAX_VALUE
+            return QuantileLevels(thresholds)
+        }
+
+        private fun nearestRank(sorted: List<Long>, n: Int, percentile: Double): Long {
+            val rank = (Math.ceil(n * percentile).toInt() - 1).coerceIn(0, n - 1)
+            return sorted[rank]
+        }
+    }
+}
+
+// ── 辅助类型 ──────────────────────────────────────────────
+
+/** computeAllViews 返回值 */
+internal data class AllViewsResult(
+    val dailyData: List<HeatmapDayData>,
+    val weeklyData: List<HeatmapWeekData>,
+    val cumulativeData: List<HeatmapDayData>,
+    val dailyLevels: QuantileLevels,
+    val weeklyLevels: QuantileLevels,
+    val cumulativeLevels: QuantileLevels,
+)
\ No newline at end of file