diff --git a/app/schemas/com.m57.hermescontrol.data.local.HermesDatabase/7.json b/app/schemas/com.m57.hermescontrol.data.local.HermesDatabase/7.json
new file mode 100644
index 00000000..3f6766e7
--- /dev/null
+++ b/app/schemas/com.m57.hermescontrol.data.local.HermesDatabase/7.json
@@ -0,0 +1,110 @@
+{
+  "formatVersion": 1,
+  "database": {
+    "version": 7,
+    "identityHash": "3deb0f3f202de9316222a02a0b5ae80e",
+    "entities": [
+      {
+        "tableName": "chat_messages",
+        "createSql": "CREATE TABLE IF NOT EXISTS `${TABLE_NAME}` (`id` TEXT NOT NULL, `session_id` TEXT NOT NULL, `role` TEXT NOT NULL, `content` TEXT NOT NULL, `reasoning_text` TEXT NOT NULL, `timestamp` INTEGER NOT NULL, `tool_name` TEXT, `tool_call_id` TEXT NOT NULL, `tool_status` TEXT, `is_streaming` INTEGER NOT NULL, `display_kind` TEXT, `token_count` INTEGER, `tps` REAL, PRIMARY KEY(`id`))",
+        "fields": [
+          {
+            "fieldPath": "id",
+            "columnName": "id",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "sessionId",
+            "columnName": "session_id",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "role",
+            "columnName": "role",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "content",
+            "columnName": "content",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "reasoningText",
+            "columnName": "reasoning_text",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "timestamp",
+            "columnName": "timestamp",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "toolName",
+            "columnName": "tool_name",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "toolCallId",
+            "columnName": "tool_call_id",
+            "affinity": "TEXT",
+            "notNull": true
+          },
+          {
+            "fieldPath": "toolStatus",
+            "columnName": "tool_status",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "isStreaming",
+            "columnName": "is_streaming",
+            "affinity": "INTEGER",
+            "notNull": true
+          },
+          {
+            "fieldPath": "displayKind",
+            "columnName": "display_kind",
+            "affinity": "TEXT"
+          },
+          {
+            "fieldPath": "tokenCount",
+            "columnName": "token_count",
+            "affinity": "INTEGER"
+          },
+          {
+            "fieldPath": "tps",
+            "columnName": "tps",
+            "affinity": "REAL"
+          }
+        ],
+        "primaryKey": {
+          "autoGenerate": false,
+          "columnNames": [
+            "id"
+          ]
+        },
+        "indices": [
+          {
+            "name": "index_chat_messages_session_id_timestamp",
+            "unique": false,
+            "columnNames": [
+              "session_id",
+              "timestamp"
+            ],
+            "orders": [],
+            "createSql": "CREATE INDEX IF NOT EXISTS `index_chat_messages_session_id_timestamp` ON `${TABLE_NAME}` (`session_id`, `timestamp`)"
+          }
+        ]
+      }
+    ],
+    "setupQueries": [
+      "CREATE TABLE IF NOT EXISTS room_master_table (id INTEGER PRIMARY KEY,identity_hash TEXT)",
+      "INSERT OR REPLACE INTO room_master_table (id,identity_hash) VALUES(42, '3deb0f3f202de9316222a02a0b5ae80e')"
+    ]
+  }
+}
\ No newline at end of file
diff --git a/app/src/main/java/com/m57/hermescontrol/data/local/ChatMessageEntity.kt b/app/src/main/java/com/m57/hermescontrol/data/local/ChatMessageEntity.kt
index dcfebf18..548e45b5 100644
--- a/app/src/main/java/com/m57/hermescontrol/data/local/ChatMessageEntity.kt
+++ b/app/src/main/java/com/m57/hermescontrol/data/local/ChatMessageEntity.kt
@@ -42,4 +42,8 @@ data class ChatMessageEntity(
     val isStreaming: Boolean = false,
     @ColumnInfo(name = "display_kind")
     val displayKind: String? = null,
+    @ColumnInfo(name = "token_count")
+    val tokenCount: Int? = null,
+    @ColumnInfo(name = "tps")
+    val tps: Double? = null,
 )
diff --git a/app/src/main/java/com/m57/hermescontrol/data/local/ChatMessageMapper.kt b/app/src/main/java/com/m57/hermescontrol/data/local/ChatMessageMapper.kt
index a7f40f68..2600557b 100644
--- a/app/src/main/java/com/m57/hermescontrol/data/local/ChatMessageMapper.kt
+++ b/app/src/main/java/com/m57/hermescontrol/data/local/ChatMessageMapper.kt
@@ -35,6 +35,8 @@ fun ChatMessageEntity.toUiModel(): ChatMessage =
                 else -> null
             },
         displayKind = displayKind,
+        tokenCount = tokenCount,
+        tps = tps,
     )
 
 fun ChatMessage.toEntity(sessionId: String): ChatMessageEntity =
@@ -50,4 +52,6 @@ fun ChatMessage.toEntity(sessionId: String): ChatMessageEntity =
         toolStatus = toolStatus?.name,
         isStreaming = isStreaming,
         displayKind = displayKind,
+        tokenCount = tokenCount,
+        tps = tps,
     )
diff --git a/app/src/main/java/com/m57/hermescontrol/data/local/HermesDatabase.kt b/app/src/main/java/com/m57/hermescontrol/data/local/HermesDatabase.kt
index 48b4b922..ab12e8b5 100644
--- a/app/src/main/java/com/m57/hermescontrol/data/local/HermesDatabase.kt
+++ b/app/src/main/java/com/m57/hermescontrol/data/local/HermesDatabase.kt
@@ -12,7 +12,7 @@ import java.io.File
 
 @Database(
     entities = [ChatMessageEntity::class],
-    version = 6,
+    version = 7,
     exportSchema = true,
 )
 abstract class HermesDatabase : RoomDatabase() {
@@ -67,6 +67,18 @@ abstract class HermesDatabase : RoomDatabase() {
                 }
             }
 
+        val MIGRATION_6_7: Migration =
+            object : Migration(6, 7) {
+                override suspend fun migrate(connection: SQLiteConnection) {
+                    connection.execSQL(
+                        "ALTER TABLE `chat_messages` ADD COLUMN `token_count` INTEGER",
+                    )
+                    connection.execSQL(
+                        "ALTER TABLE `chat_messages` ADD COLUMN `tps` REAL",
+                    )
+                }
+            }
+
         fun get(context: Context): HermesDatabase =
             instance ?: synchronized(this) {
                 // SQLCipher can't open plaintext SQLite databases — if an old
@@ -97,6 +109,7 @@ abstract class HermesDatabase : RoomDatabase() {
                         MIGRATION_3_4,
                         MIGRATION_4_5,
                         MIGRATION_5_6,
+                        MIGRATION_6_7,
                     ).fallbackToDestructiveMigration(false)
                     .build()
                     .also { instance = it }
diff --git a/app/src/main/java/com/m57/hermescontrol/data/model/SessionMessagesResponse.kt b/app/src/main/java/com/m57/hermescontrol/data/model/SessionMessagesResponse.kt
index 7f2ca663..c05002a3 100644
--- a/app/src/main/java/com/m57/hermescontrol/data/model/SessionMessagesResponse.kt
+++ b/app/src/main/java/com/m57/hermescontrol/data/model/SessionMessagesResponse.kt
@@ -4,6 +4,7 @@ import kotlinx.serialization.Serializable
 import kotlinx.serialization.json.JsonElement
 import kotlinx.serialization.json.JsonPrimitive
 import kotlinx.serialization.json.doubleOrNull
+import kotlinx.serialization.json.intOrNull
 import kotlinx.serialization.json.longOrNull
 
 @Serializable
@@ -50,10 +51,18 @@ data class SessionMessage(
     val display_kind: String? = null,
     /** Display-only metadata for the marker (e.g. delegation result counts). */
     val display_metadata: JsonElement? = null,
+    /** Token count recorded by the backend. */
+    val token_count: JsonElement? = null,
 ) {
     val timestampText: String?
         get() = (timestamp as? JsonPrimitive)?.content
 
+    val tokenCount: Int?
+        get() {
+            val prim = token_count as? JsonPrimitive ?: return null
+            return prim.intOrNull ?: prim.content.toIntOrNull()
+        }
+
     val timestampEpochMs: Long?
         get() {
             val prim = timestamp as? JsonPrimitive ?: return null
diff --git a/app/src/main/java/com/m57/hermescontrol/data/model/UsageSnapshotResponse.kt b/app/src/main/java/com/m57/hermescontrol/data/model/UsageSnapshotResponse.kt
index 4250dc1f..9897320f 100644
--- a/app/src/main/java/com/m57/hermescontrol/data/model/UsageSnapshotResponse.kt
+++ b/app/src/main/java/com/m57/hermescontrol/data/model/UsageSnapshotResponse.kt
@@ -20,6 +20,8 @@ data class UsageSnapshotResponse(
     val contextMax: Long? = null,
     /** Cumulative session total tokens. */
     val totalTokens: Long? = null,
+    /** Rolling tokens-per-second generation speed. */
+    val avgTps: Double? = null,
 )
 
 /**
@@ -38,5 +40,6 @@ fun parseUsageSnapshot(result: Any?): UsageSnapshotResponse? {
         contextUsed = (usageMap["context_used"] as? Number)?.toLong(),
         contextMax = (usageMap["context_max"] as? Number)?.toLong(),
         totalTokens = (usageMap["total"] as? Number)?.toLong(),
+        avgTps = (usageMap["avg_tps"] as? Number)?.toDouble(),
     )
 }
diff --git a/app/src/main/java/com/m57/hermescontrol/data/ws/EventParser.kt b/app/src/main/java/com/m57/hermescontrol/data/ws/EventParser.kt
index 4a19bd98..f2d680b3 100644
--- a/app/src/main/java/com/m57/hermescontrol/data/ws/EventParser.kt
+++ b/app/src/main/java/com/m57/hermescontrol/data/ws/EventParser.kt
@@ -87,7 +87,7 @@ object EventParser {
             "message.complete" -> {
                 val text = payload?.get("text") as? String ?: ""
                 val reasoning = payload?.get("reasoning") as? String
-                WsEvent.MessageComplete(text, sessionId, reasoning)
+                WsEvent.MessageComplete(text, sessionId, reasoning, rawPayload = payload)
             }
 
             "message.done" -> {
diff --git a/app/src/main/java/com/m57/hermescontrol/data/ws/WsEvent.kt b/app/src/main/java/com/m57/hermescontrol/data/ws/WsEvent.kt
index 21d61b6a..3282eef9 100644
--- a/app/src/main/java/com/m57/hermescontrol/data/ws/WsEvent.kt
+++ b/app/src/main/java/com/m57/hermescontrol/data/ws/WsEvent.kt
@@ -61,6 +61,8 @@ sealed class WsEvent {
         val reasoning: String? = null,
         /** Stored session id captured before background disconnect clears the active mapping. */
         val storedSessionId: String? = null,
+        /** Full raw payload map (including usage/avg_tps) emitted with message.complete. */
+        val rawPayload: Map<String, Any?>? = null,
     ) : WsEvent()
 
     data class MessageDone(
diff --git a/app/src/main/java/com/m57/hermescontrol/ui/bots/group/GroupChatModels.kt b/app/src/main/java/com/m57/hermescontrol/ui/bots/group/GroupChatModels.kt
index 95600291..fcada2ea 100644
--- a/app/src/main/java/com/m57/hermescontrol/ui/bots/group/GroupChatModels.kt
+++ b/app/src/main/java/com/m57/hermescontrol/ui/bots/group/GroupChatModels.kt
@@ -34,6 +34,8 @@ data class GroupChatMessage(
     val isSystem: Boolean = false,
     val isPass: Boolean = false,
     val toolCalls: List<GroupChatToolCall> = emptyList(),
+    val tokenCount: Int? = null,
+    val tps: Double? = null,
 )
 
 const val STOPPED_SYSTEM_TEXT = "Discussion stopped"
diff --git a/app/src/main/java/com/m57/hermescontrol/ui/bots/group/components/GroupMessageCard.kt b/app/src/main/java/com/m57/hermescontrol/ui/bots/group/components/GroupMessageCard.kt
index bc403a26..719adf32 100644
--- a/app/src/main/java/com/m57/hermescontrol/ui/bots/group/components/GroupMessageCard.kt
+++ b/app/src/main/java/com/m57/hermescontrol/ui/bots/group/components/GroupMessageCard.kt
@@ -49,6 +49,7 @@ import com.m57.hermescontrol.ui.bots.group.CAPPED_SYSTEM_TEXT
 import com.m57.hermescontrol.ui.bots.group.GroupChatMessage
 import com.m57.hermescontrol.ui.bots.group.STOPPED_SYSTEM_TEXT
 import com.m57.hermescontrol.ui.chat.MarkdownText
+import com.m57.hermescontrol.ui.chat.TokenEstimator
 import com.m57.hermescontrol.ui.chat.formatTimestamp
 import com.m57.hermescontrol.ui.common.BotAvatar
 import com.m57.hermescontrol.util.BidiUtils
@@ -195,6 +196,26 @@ fun GroupMessageCard(
                                 color = MaterialTheme.colorScheme.onPrimary.copy(alpha = 0.7f),
                                 style = MaterialTheme.typography.labelSmall,
                             )
+                            val userTokens =
+                                message.tokenCount ?: TokenEstimator.estimate(message.text).takeIf { it > 0 }
+                            if (userTokens != null && userTokens > 0) {
+                                Spacer(modifier = Modifier.width(4.dp))
+                                Text(
+                                    text = "•",
+                                    style = MaterialTheme.typography.labelSmall,
+                                    color = MaterialTheme.colorScheme.onPrimary.copy(alpha = 0.5f),
+                                )
+                                Spacer(modifier = Modifier.width(4.dp))
+                                Text(
+                                    text =
+                                        stringResource(
+                                            R.string.chat_msg_tokens,
+                                            TokenEstimator.formatTokenCount(userTokens),
+                                        ),
+                                    color = MaterialTheme.colorScheme.onPrimary.copy(alpha = 0.7f),
+                                    style = MaterialTheme.typography.labelSmall,
+                                )
+                            }
                         }
                     }
                 }
@@ -304,6 +325,44 @@ fun GroupMessageCard(
                                 color = MaterialTheme.colorScheme.onSurfaceVariant.copy(alpha = 0.7f),
                                 style = MaterialTheme.typography.labelSmall,
                             )
+                            val botTokens =
+                                message.tokenCount ?: TokenEstimator.estimate(message.text).takeIf { it > 0 }
+                            if (botTokens != null && botTokens > 0) {
+                                Spacer(modifier = Modifier.width(4.dp))
+                                Text(
+                                    text = "•",
+                                    style = MaterialTheme.typography.labelSmall,
+                                    color = MaterialTheme.colorScheme.onSurfaceVariant.copy(alpha = 0.4f),
+                                )
+                                Spacer(modifier = Modifier.width(4.dp))
+                                Text(
+                                    text =
+                                        stringResource(
+                                            R.string.chat_msg_tokens,
+                                            TokenEstimator.formatTokenCount(botTokens),
+                                        ),
+                                    color = MaterialTheme.colorScheme.onSurfaceVariant.copy(alpha = 0.7f),
+                                    style = MaterialTheme.typography.labelSmall,
+                                )
+                            }
+                            if (message.tps != null && message.tps > 0.0) {
+                                Spacer(modifier = Modifier.width(4.dp))
+                                Text(
+                                    text = "•",
+                                    style = MaterialTheme.typography.labelSmall,
+                                    color = MaterialTheme.colorScheme.onSurfaceVariant.copy(alpha = 0.4f),
+                                )
+                                Spacer(modifier = Modifier.width(4.dp))
+                                Text(
+                                    text =
+                                        stringResource(
+                                            R.string.chat_msg_tps,
+                                            TokenEstimator.formatTps(message.tps),
+                                        ),
+                                    color = MaterialTheme.colorScheme.onSurfaceVariant.copy(alpha = 0.7f),
+                                    style = MaterialTheme.typography.labelSmall,
+                                )
+                            }
                         }
                     }
                 }
diff --git a/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatBubble.kt b/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatBubble.kt
index 779788d2..6668826a 100644
--- a/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatBubble.kt
+++ b/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatBubble.kt
@@ -257,6 +257,44 @@ fun ChatBubble(
                                         color = userBubbleTextColor.copy(alpha = 0.6f),
                                         style = MaterialTheme.typography.labelSmall,
                                     )
+                                    if (message.tokenCount != null && message.tokenCount > 0) {
+                                        Spacer(modifier = Modifier.width(4.dp))
+                                        Text(
+                                            text = "•",
+                                            style = MaterialTheme.typography.labelSmall,
+                                            color = userBubbleTextColor.copy(alpha = 0.4f),
+                                        )
+                                        Spacer(modifier = Modifier.width(4.dp))
+                                        Text(
+                                            text =
+                                                stringResource(
+                                                    R.string.chat_msg_tokens,
+                                                    TokenEstimator.formatTokenCount(message.tokenCount),
+                                                ),
+                                            color = userBubbleTextColor.copy(alpha = 0.7f),
+                                            style = MaterialTheme.typography.labelSmall,
+                                            modifier = Modifier.testTag("bubble_token_count"),
+                                        )
+                                    }
+                                    if (message.tps != null && message.tps > 0.0) {
+                                        Spacer(modifier = Modifier.width(4.dp))
+                                        Text(
+                                            text = "•",
+                                            style = MaterialTheme.typography.labelSmall,
+                                            color = userBubbleTextColor.copy(alpha = 0.4f),
+                                        )
+                                        Spacer(modifier = Modifier.width(4.dp))
+                                        Text(
+                                            text =
+                                                stringResource(
+                                                    R.string.chat_msg_tps,
+                                                    TokenEstimator.formatTps(message.tps),
+                                                ),
+                                            color = userBubbleTextColor.copy(alpha = 0.7f),
+                                            style = MaterialTheme.typography.labelSmall,
+                                            modifier = Modifier.testTag("bubble_tps"),
+                                        )
+                                    }
                                 }
                             }
                         }
diff --git a/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatMessage.kt b/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatMessage.kt
index 55b49a1f..b7afe3d6 100644
--- a/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatMessage.kt
+++ b/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatMessage.kt
@@ -79,6 +79,10 @@ data class ChatMessage(
      * When present, rendered at the bottom beside the copy button.
      */
     val finishTimestamp: Long? = null,
+    /** Token count used by this message (persisted to SQLite). */
+    val tokenCount: Int? = null,
+    /** Tokens per second generation speed (persisted to SQLite). */
+    val tps: Double? = null,
 )
 
 /**
diff --git a/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatServerMessageMapper.kt b/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatServerMessageMapper.kt
index 82eeffd5..2ef7269b 100644
--- a/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatServerMessageMapper.kt
+++ b/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatServerMessageMapper.kt
@@ -171,6 +171,7 @@ internal fun mapServerMessages(
             }
         }
 
+        val tokenCount = msg.tokenCount ?: TokenEstimator.estimate(finalContent).takeIf { it > 0 }
         mapped.add(
             ChatMessage(
                 id = restId,
@@ -182,6 +183,7 @@ internal fun mapServerMessages(
                 timestamp = timestamp,
                 isStreaming = false,
                 displayKind = msg.display_kind,
+                tokenCount = tokenCount,
             ),
         )
     }
diff --git a/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatViewModel.kt b/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatViewModel.kt
index 3077478c..a671f7b0 100644
--- a/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatViewModel.kt
+++ b/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatViewModel.kt
@@ -146,6 +146,8 @@ data class ChatUiState(
     // until the first successful session.usage fetch) — drives the
     // "compressed ×N" badge on the context chip.
     val compressionCount: Int? = null,
+    /** Rolling output tokens/sec over the last ~10 calls. */
+    val latestTps: Double? = null,
     // Attachment state
     val pendingAttachments: List<Attachment> = emptyList(),
     /** One-shot composer recovery after an attachment is rejected before send. */
@@ -1298,6 +1300,7 @@ class ChatViewModel(
                 role = MessageRole.USER,
                 content = text,
                 attachments = if (attachments.isNotEmpty()) attachments else null,
+                tokenCount = TokenEstimator.estimate(text).takeIf { it > 0 },
             )
 
         // Update UI immediately
@@ -1342,6 +1345,7 @@ class ChatViewModel(
                 role = MessageRole.USER,
                 content = text,
                 attachments = if (attachments.isNotEmpty()) attachments else null,
+                tokenCount = TokenEstimator.estimate(text).takeIf { it > 0 },
             )
 
         // Upload attachments then submit prompt
@@ -1624,7 +1628,12 @@ class ChatViewModel(
 
         val displayContent =
             if (result is SlashResult.QueuePrompt) result.displayContent else command
-        val userMsg = ChatMessage(role = MessageRole.USER, content = displayContent)
+        val userMsg =
+            ChatMessage(
+                role = MessageRole.USER,
+                content = displayContent,
+                tokenCount = TokenEstimator.estimate(displayContent).takeIf { it > 0 },
+            )
         val sessionId = _uiState.value.currentSessionId
 
         _uiState.update { it.copy(messages = it.messages + userMsg) }
diff --git a/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatWsEventReducer.kt b/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatWsEventReducer.kt
index d8db9c15..4c2ad8eb 100644
--- a/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatWsEventReducer.kt
+++ b/app/src/main/java/com/m57/hermescontrol/ui/chat/ChatWsEventReducer.kt
@@ -221,6 +221,8 @@ object ChatWsEventReducer {
                     streamingState.streamingMessage.copy(
                         isStreaming = false,
                         finishTimestamp = System.currentTimeMillis(),
+                        tokenCount = TokenEstimator.estimate(streamingState.streamingMessage.content).takeIf { it > 0 },
+                        tps = state.latestTps?.takeIf { it > 0.0 },
                     )
                 orphan = finalized
                 state.copy(
@@ -379,17 +381,25 @@ object ChatWsEventReducer {
                 streamingState = streamingState.copy(streamingMessage = null),
             )
         }
+        val rawUsage = event.rawPayload?.get("usage") as? Map<*, *>
+        val tpsFromPayload = (rawUsage?.get("avg_tps") as? Number)?.toDouble()
+        val tps = (tpsFromPayload ?: state.latestTps)?.takeIf { it > 0.0 }
+        val tokenCount = TokenEstimator.estimate(text).takeIf { it > 0 }
         val msg =
             streaming?.copy(
                 content = text,
                 isStreaming = false,
                 reasoningText = reasoning,
                 finishTimestamp = System.currentTimeMillis(),
+                tokenCount = tokenCount,
+                tps = tps,
             ) ?: ChatMessage(
                 role = MessageRole.ASSISTANT,
                 content = text,
                 reasoningText = reasoning,
                 finishTimestamp = System.currentTimeMillis(),
+                tokenCount = tokenCount,
+                tps = tps,
             )
         val effects = mutableListOf<ReducerEffect>()
         val sid = state.currentSessionId
@@ -435,11 +445,14 @@ object ChatWsEventReducer {
             } else {
                 streaming.reasoningText
             }
+        val tokenCount = TokenEstimator.estimate(streaming.content).takeIf { it > 0 }
         val msg =
             streaming.copy(
                 isStreaming = false,
                 reasoningText = reasoning,
                 finishTimestamp = System.currentTimeMillis(),
+                tokenCount = streaming.tokenCount ?: tokenCount,
+                tps = streaming.tps ?: state.latestTps?.takeIf { it > 0.0 },
             )
         val effects = mutableListOf<ReducerEffect>()
         val sid = state.currentSessionId
@@ -504,6 +517,8 @@ object ChatWsEventReducer {
                         isStreaming = false,
                         reasoningText = reasoning,
                         finishTimestamp = System.currentTimeMillis(),
+                        tokenCount = TokenEstimator.estimate(streamingState.streamingMessage.content).takeIf { it > 0 },
+                        tps = state.latestTps?.takeIf { it > 0.0 },
                     )
                 orphanToPersist = finalized
                 state.copy(
@@ -962,6 +977,9 @@ object ChatWsEventReducer {
         if (snapshot.contextMax != null && snapshot.contextMax > 0L) {
             updatedState = updatedState.copy(fullContextTokens = snapshot.contextMax)
         }
+        if (snapshot.avgTps != null && snapshot.avgTps > 0.0) {
+            updatedState = updatedState.copy(latestTps = snapshot.avgTps)
+        }
         return ReducerResult(state = updatedState, streamingState = streamingState)
     }
 }
diff --git a/app/src/main/java/com/m57/hermescontrol/ui/chat/TokenEstimator.kt b/app/src/main/java/com/m57/hermescontrol/ui/chat/TokenEstimator.kt
new file mode 100644
index 00000000..884edcc1
--- /dev/null
+++ b/app/src/main/java/com/m57/hermescontrol/ui/chat/TokenEstimator.kt
@@ -0,0 +1,55 @@
+package com.m57.hermescontrol.ui.chat
+
+import java.util.Locale
+
+/**
+ * Token count estimation and formatting utilities.
+ *
+ * The estimation algorithm matches the backend's `estimate_tokens_rough`
+ * in `agent/model_metadata.py`:
+ * - CJK ideographs, Hangul, Kana count as ~1 token each.
+ * - Non-CJK characters count as ceil(UTF-8 bytes / 4).
+ */
+object TokenEstimator {
+    fun estimate(text: String?): Int {
+        if (text.isNullOrEmpty()) return 0
+        var dense = 0
+        var nonDenseByteCount = 0
+        for (i in 0 until text.length) {
+            val ch = text[i]
+            if (isCjkDense(ch)) {
+                dense++
+            } else {
+                val code = ch.code
+                nonDenseByteCount +=
+                    when {
+                        code <= 0x7F -> 1
+                        code <= 0x7FF -> 2
+                        code <= 0xFFFF -> 3
+                        else -> 4
+                    }
+            }
+        }
+        return dense + ((nonDenseByteCount + 3) / 4)
+    }
+
+    private fun isCjkDense(ch: Char): Boolean {
+        val code = ch.code
+        return (code in 0x4E00..0x9FFF) ||
+            (code in 0x3400..0x4DBF) ||
+            (code in 0xAC00..0xD7AF) ||
+            (code in 0x1100..0x11FF) ||
+            (code in 0x3040..0x309F) ||
+            (code in 0x30A0..0x30FF)
+    }
+
+    fun formatTokenCount(tokens: Int): String =
+        when {
+            tokens >= 1_000_000 -> String.format(Locale.US, "%.1fM", tokens / 1_000_000.0)
+            tokens >= 10_000 -> String.format(Locale.US, "%.1fk", tokens / 1_000.0)
+            tokens >= 1_000 -> String.format(Locale.US, "%,d", tokens)
+            else -> tokens.toString()
+        }
+
+    fun formatTps(tps: Double): String = Math.round(tps).toString()
+}
diff --git a/app/src/main/java/com/m57/hermescontrol/ui/chat/fullbleed/FullBleedAgentMessage.kt b/app/src/main/java/com/m57/hermescontrol/ui/chat/fullbleed/FullBleedAgentMessage.kt
index 8acdba59..f34a0fa5 100644
--- a/app/src/main/java/com/m57/hermescontrol/ui/chat/fullbleed/FullBleedAgentMessage.kt
+++ b/app/src/main/java/com/m57/hermescontrol/ui/chat/fullbleed/FullBleedAgentMessage.kt
@@ -40,6 +40,7 @@ import com.m57.hermescontrol.ui.chat.ChatMessage
 import com.m57.hermescontrol.ui.chat.ImageViewerModel
 import com.m57.hermescontrol.ui.chat.InlineAttachment
 import com.m57.hermescontrol.ui.chat.MarkdownText
+import com.m57.hermescontrol.ui.chat.TokenEstimator
 import com.m57.hermescontrol.ui.chat.components.ReasoningCard
 import kotlinx.coroutines.delay
 import kotlinx.coroutines.launch
@@ -171,6 +172,44 @@ internal fun FullBleedAgentMessage(
                     color = MaterialTheme.colorScheme.onSurfaceVariant.copy(alpha = 0.7f),
                     modifier = Modifier.testTag("fullbleed_finish_time"),
                 )
+                if (message.tokenCount != null && message.tokenCount > 0) {
+                    Spacer(modifier = Modifier.width(6.dp))
+                    Text(
+                        text = "•",
+                        style = MaterialTheme.typography.labelSmall,
+                        color = MaterialTheme.colorScheme.onSurfaceVariant.copy(alpha = 0.5f),
+                    )
+                    Spacer(modifier = Modifier.width(6.dp))
+                    Text(
+                        text =
+                            stringResource(
+                                R.string.chat_msg_tokens,
+                                TokenEstimator.formatTokenCount(message.tokenCount),
+                            ),
+                        style = MaterialTheme.typography.labelSmall,
+                        color = MaterialTheme.colorScheme.onSurfaceVariant.copy(alpha = 0.7f),
+                        modifier = Modifier.testTag("fullbleed_token_count"),
+                    )
+                }
+                if (message.tps != null && message.tps > 0.0) {
+                    Spacer(modifier = Modifier.width(6.dp))
+                    Text(
+                        text = "•",
+                        style = MaterialTheme.typography.labelSmall,
+                        color = MaterialTheme.colorScheme.onSurfaceVariant.copy(alpha = 0.5f),
+                    )
+                    Spacer(modifier = Modifier.width(6.dp))
+                    Text(
+                        text =
+                            stringResource(
+                                R.string.chat_msg_tps,
+                                TokenEstimator.formatTps(message.tps),
+                            ),
+                        style = MaterialTheme.typography.labelSmall,
+                        color = MaterialTheme.colorScheme.onSurfaceVariant.copy(alpha = 0.7f),
+                        modifier = Modifier.testTag("fullbleed_tps"),
+                    )
+                }
             }
         }
     }
diff --git a/app/src/main/res/values-ar/strings.xml b/app/src/main/res/values-ar/strings.xml
index 1ba136d5..66e586fc 100644
--- a/app/src/main/res/values-ar/strings.xml
+++ b/app/src/main/res/values-ar/strings.xml
@@ -168,6 +168,8 @@
 <string name="content_desc_scroll_to_bottom">التمرير إلى الأسفل</string>
 <string name="content_desc_copy">نسخ</string>
 <string name="content_desc_edit">تعديل</string>
+<string name="chat_msg_tokens">%1$s رمز</string>
+<string name="chat_msg_tps">%1$s رمز/ث</string>
 <string name="chat_attach_remove_desc">إزالة المرفق</string>
 <string name="chat_camera_error">تعذر التقاط الصورة</string>
 <string name="settings_sec_connection">الاتصال</string>
diff --git a/app/src/main/res/values-ko/strings.xml b/app/src/main/res/values-ko/strings.xml
index e48b33cc..074382d3 100644
--- a/app/src/main/res/values-ko/strings.xml
+++ b/app/src/main/res/values-ko/strings.xml
@@ -165,6 +165,8 @@
     <string name="content_desc_scroll_to_bottom">맨 아래로</string>
     <string name="content_desc_copy">복사</string>
     <string name="content_desc_edit">편집</string>
+    <string name="chat_msg_tokens">%1$s 토큰</string>
+    <string name="chat_msg_tps">%1$s t/s</string>
 
     <!-- Attachment / File Picker -->
     <string name="chat_attach_remove_desc">첨부 제거</string>
diff --git a/app/src/main/res/values-zh/strings.xml b/app/src/main/res/values-zh/strings.xml
index fcbfd2ec..4f941cc8 100644
--- a/app/src/main/res/values-zh/strings.xml
+++ b/app/src/main/res/values-zh/strings.xml
@@ -168,6 +168,8 @@
     <string name="content_desc_scroll_to_bottom">滚动到底部</string>
     <string name="content_desc_copy">复制</string>
     <string name="content_desc_edit">编辑</string>
+    <string name="chat_msg_tokens">%1$s tok</string>
+    <string name="chat_msg_tps">%1$s t/s</string>
     <string name="chat_attach_remove_desc">移除附件</string>
     <string name="chat_camera_error">拍照失败</string>
     <string name="settings_sec_connection">连接</string>
diff --git a/app/src/main/res/values/strings.xml b/app/src/main/res/values/strings.xml
index 5ca033e7..c5872bb3 100644
--- a/app/src/main/res/values/strings.xml
+++ b/app/src/main/res/values/strings.xml
@@ -186,6 +186,8 @@
     <string name="content_desc_scroll_to_bottom">Scroll to bottom</string>
     <string name="content_desc_copy">Copy</string>
     <string name="content_desc_edit">Edit</string>
+    <string name="chat_msg_tokens">%1$s tok</string>
+    <string name="chat_msg_tps">%1$s t/s</string>
 
     <!-- Attachment / File Picker -->
     <string name="chat_attach_remove_desc">Remove attachment</string>
diff --git a/app/src/test/java/com/m57/hermescontrol/data/local/ChatMessageMapperTest.kt b/app/src/test/java/com/m57/hermescontrol/data/local/ChatMessageMapperTest.kt
index 1498858f..bf254f79 100644
--- a/app/src/test/java/com/m57/hermescontrol/data/local/ChatMessageMapperTest.kt
+++ b/app/src/test/java/com/m57/hermescontrol/data/local/ChatMessageMapperTest.kt
@@ -23,6 +23,8 @@ class ChatMessageMapperTest {
                 toolName = "search",
                 toolStatus = "RUNNING",
                 isStreaming = true,
+                tokenCount = 42,
+                tps = 55.5,
             )
 
         val ui = entity.toUiModel()
@@ -35,6 +37,8 @@ class ChatMessageMapperTest {
         assertEquals("search", ui.toolName)
         assertEquals(ToolStatus.RUNNING, ui.toolStatus)
         assertTrue(ui.isStreaming)
+        assertEquals(42, ui.tokenCount)
+        assertEquals(55.5, ui.tps ?: 0.0, 0.001)
     }
 
     @Test
@@ -49,6 +53,8 @@ class ChatMessageMapperTest {
                 toolName = null,
                 toolStatus = null,
                 isStreaming = false,
+                tokenCount = 100,
+                tps = 24.8,
             )
 
         val entity = ui.toEntity("session-b")
@@ -62,6 +68,8 @@ class ChatMessageMapperTest {
         assertNull(entity.toolName)
         assertNull(entity.toolStatus)
         assertFalse(entity.isStreaming)
+        assertEquals(100, entity.tokenCount)
+        assertEquals(24.8, entity.tps ?: 0.0, 0.001)
     }
 
     @Test
diff --git a/app/src/test/java/com/m57/hermescontrol/data/local/HermesDatabaseMigrationsTest.kt b/app/src/test/java/com/m57/hermescontrol/data/local/HermesDatabaseMigrationsTest.kt
index 35453938..63a290fa 100644
--- a/app/src/test/java/com/m57/hermescontrol/data/local/HermesDatabaseMigrationsTest.kt
+++ b/app/src/test/java/com/m57/hermescontrol/data/local/HermesDatabaseMigrationsTest.kt
@@ -70,4 +70,19 @@ class HermesDatabaseMigrationsTest {
                 )
             }
         }
+
+    @Test
+    fun migration6to7_executesTokenCountAndTpsColumnAdditionOnSQLiteConnection() =
+        runBlocking {
+            val connection = mockk<SQLiteConnection>(relaxed = true)
+            HermesDatabase.MIGRATION_6_7.migrate(connection)
+            verify(exactly = 1) {
+                connection.execSQL(
+                    "ALTER TABLE `chat_messages` ADD COLUMN `token_count` INTEGER",
+                )
+                connection.execSQL(
+                    "ALTER TABLE `chat_messages` ADD COLUMN `tps` REAL",
+                )
+            }
+        }
 }
diff --git a/app/src/test/java/com/m57/hermescontrol/ui/chat/TokenEstimatorTest.kt b/app/src/test/java/com/m57/hermescontrol/ui/chat/TokenEstimatorTest.kt
new file mode 100644
index 00000000..eea5f418
--- /dev/null
+++ b/app/src/test/java/com/m57/hermescontrol/ui/chat/TokenEstimatorTest.kt
@@ -0,0 +1,43 @@
+package com.m57.hermescontrol.ui.chat
+
+import org.junit.Assert.assertEquals
+import org.junit.Test
+
+class TokenEstimatorTest {
+    @Test
+    fun estimate_emptyOrNull_returnsZero() {
+        assertEquals(0, TokenEstimator.estimate(null))
+        assertEquals(0, TokenEstimator.estimate(""))
+    }
+
+    @Test
+    fun estimate_asciiText_usesByteDiv4() {
+        // "test" is 4 bytes -> (4 + 3) / 4 = 1 token
+        assertEquals(1, TokenEstimator.estimate("test"))
+        // 8 chars -> (8 + 3) / 4 = 2 tokens
+        assertEquals(2, TokenEstimator.estimate("testtest"))
+    }
+
+    @Test
+    fun estimate_cjkText_countsDenseCharacters() {
+        // 3 CJK ideographs -> 3 tokens
+        assertEquals(3, TokenEstimator.estimate("你好世界".substring(0, 3)))
+        // 4 Korean Hangul -> 4 tokens
+        assertEquals(4, TokenEstimator.estimate("안녕하세요".substring(0, 4)))
+    }
+
+    @Test
+    fun formatTokenCount_formatsCompactUnits() {
+        assertEquals("42", TokenEstimator.formatTokenCount(42))
+        assertEquals("999", TokenEstimator.formatTokenCount(999))
+        assertEquals("1,500", TokenEstimator.formatTokenCount(1500))
+        assertEquals("12.5k", TokenEstimator.formatTokenCount(12500))
+        assertEquals("1.2M", TokenEstimator.formatTokenCount(1200000))
+    }
+
+    @Test
+    fun formatTps_roundsProperly() {
+        assertEquals("42", TokenEstimator.formatTps(41.6))
+        assertEquals("50", TokenEstimator.formatTps(50.2))
+    }
+}