diff --git a/chat.pb.go b/chat.pb.go
index dfd71ef..2555a37 100644
--- a/chat.pb.go
+++ b/chat.pb.go
@@ -217,7 +217,7 @@ type Message struct {
 	ReplyToMessageId *string                `protobuf:"bytes,8,opt,name=reply_to_message_id,json=replyToMessageId,proto3,oneof" json:"reply_to_message_id,omitempty"` // parent message when this is a reply
 	RootMessageId    string                 `protobuf:"bytes,9,opt,name=root_message_id,json=rootMessageId,proto3" json:"root_message_id,omitempty"`                  // root of the reply chain (== id for non-replies)
 	ReplyDepth       int32                  `protobuf:"varint,10,opt,name=reply_depth,json=replyDepth,proto3" json:"reply_depth,omitempty"`                           // 0 for non-replies, parent depth + 1 for replies; hard max 20
-	CreatedAt        *timestamppb.Timestamp `protobuf:"bytes,11,opt,name=created_at,json=createdAt,proto3" json:"created_at,omitempty"`
+	CreatedAt        *timestamppb.Timestamp `protobuf:"bytes,11,opt,name=created_at,json=createdAt,proto3" json:"created_at,omitempty"`                               // original creation time; age = now - created_at
 	unknownFields    protoimpl.UnknownFields
 	sizeCache        protoimpl.SizeCache
 }
@@ -689,6 +689,76 @@ func (x *PostMessageRequest) GetIdempotencyKey() string {
 	return ""
 }
 
+type AgentPostMessageRequest struct {
+	state            protoimpl.MessageState `protogen:"open.v1"`
+	TopicId          string                 `protobuf:"bytes,1,opt,name=topic_id,json=topicId,proto3" json:"topic_id,omitempty"`
+	Content          string                 `protobuf:"bytes,2,opt,name=content,proto3" json:"content,omitempty"`
+	ReplyToMessageId *string                `protobuf:"bytes,3,opt,name=reply_to_message_id,json=replyToMessageId,proto3,oneof" json:"reply_to_message_id,omitempty"`
+	// Mandatory: retrying with the same key returns the original message
+	// (duplicated=true) instead of creating a duplicate.
+	IdempotencyKey string `protobuf:"bytes,4,opt,name=idempotency_key,json=idempotencyKey,proto3" json:"idempotency_key,omitempty"`
+	unknownFields  protoimpl.UnknownFields
+	sizeCache      protoimpl.SizeCache
+}
+
+func (x *AgentPostMessageRequest) Reset() {
+	*x = AgentPostMessageRequest{}
+	mi := &file_chat_proto_msgTypes[8]
+	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+	ms.StoreMessageInfo(mi)
+}
+
+func (x *AgentPostMessageRequest) String() string {
+	return protoimpl.X.MessageStringOf(x)
+}
+
+func (*AgentPostMessageRequest) ProtoMessage() {}
+
+func (x *AgentPostMessageRequest) ProtoReflect() protoreflect.Message {
+	mi := &file_chat_proto_msgTypes[8]
+	if x != nil {
+		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+		if ms.LoadMessageInfo() == nil {
+			ms.StoreMessageInfo(mi)
+		}
+		return ms
+	}
+	return mi.MessageOf(x)
+}
+
+// Deprecated: Use AgentPostMessageRequest.ProtoReflect.Descriptor instead.
+func (*AgentPostMessageRequest) Descriptor() ([]byte, []int) {
+	return file_chat_proto_rawDescGZIP(), []int{8}
+}
+
+func (x *AgentPostMessageRequest) GetTopicId() string {
+	if x != nil {
+		return x.TopicId
+	}
+	return ""
+}
+
+func (x *AgentPostMessageRequest) GetContent() string {
+	if x != nil {
+		return x.Content
+	}
+	return ""
+}
+
+func (x *AgentPostMessageRequest) GetReplyToMessageId() string {
+	if x != nil && x.ReplyToMessageId != nil {
+		return *x.ReplyToMessageId
+	}
+	return ""
+}
+
+func (x *AgentPostMessageRequest) GetIdempotencyKey() string {
+	if x != nil {
+		return x.IdempotencyKey
+	}
+	return ""
+}
+
 type PostMessageResponse struct {
 	state         protoimpl.MessageState `protogen:"open.v1"`
 	Message       *Message               `protobuf:"bytes,1,opt,name=message,proto3" json:"message,omitempty"`
@@ -699,7 +769,7 @@ type PostMessageResponse struct {
 
 func (x *PostMessageResponse) Reset() {
 	*x = PostMessageResponse{}
-	mi := &file_chat_proto_msgTypes[8]
+	mi := &file_chat_proto_msgTypes[9]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -711,7 +781,7 @@ func (x *PostMessageResponse) String() string {
 func (*PostMessageResponse) ProtoMessage() {}
 
 func (x *PostMessageResponse) ProtoReflect() protoreflect.Message {
-	mi := &file_chat_proto_msgTypes[8]
+	mi := &file_chat_proto_msgTypes[9]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -724,7 +794,7 @@ func (x *PostMessageResponse) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use PostMessageResponse.ProtoReflect.Descriptor instead.
 func (*PostMessageResponse) Descriptor() ([]byte, []int) {
-	return file_chat_proto_rawDescGZIP(), []int{8}
+	return file_chat_proto_rawDescGZIP(), []int{9}
 }
 
 func (x *PostMessageResponse) GetMessage() *Message {
@@ -750,7 +820,7 @@ type GetMessageRequest struct {
 
 func (x *GetMessageRequest) Reset() {
 	*x = GetMessageRequest{}
-	mi := &file_chat_proto_msgTypes[9]
+	mi := &file_chat_proto_msgTypes[10]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -762,7 +832,7 @@ func (x *GetMessageRequest) String() string {
 func (*GetMessageRequest) ProtoMessage() {}
 
 func (x *GetMessageRequest) ProtoReflect() protoreflect.Message {
-	mi := &file_chat_proto_msgTypes[9]
+	mi := &file_chat_proto_msgTypes[10]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -775,7 +845,7 @@ func (x *GetMessageRequest) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use GetMessageRequest.ProtoReflect.Descriptor instead.
 func (*GetMessageRequest) Descriptor() ([]byte, []int) {
-	return file_chat_proto_rawDescGZIP(), []int{9}
+	return file_chat_proto_rawDescGZIP(), []int{10}
 }
 
 func (x *GetMessageRequest) GetMessageId() string {
@@ -802,7 +872,7 @@ type HistoryRequest struct {
 
 func (x *HistoryRequest) Reset() {
 	*x = HistoryRequest{}
-	mi := &file_chat_proto_msgTypes[10]
+	mi := &file_chat_proto_msgTypes[11]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -814,7 +884,7 @@ func (x *HistoryRequest) String() string {
 func (*HistoryRequest) ProtoMessage() {}
 
 func (x *HistoryRequest) ProtoReflect() protoreflect.Message {
-	mi := &file_chat_proto_msgTypes[10]
+	mi := &file_chat_proto_msgTypes[11]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -827,7 +897,7 @@ func (x *HistoryRequest) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use HistoryRequest.ProtoReflect.Descriptor instead.
 func (*HistoryRequest) Descriptor() ([]byte, []int) {
-	return file_chat_proto_rawDescGZIP(), []int{10}
+	return file_chat_proto_rawDescGZIP(), []int{11}
 }
 
 func (x *HistoryRequest) GetTopicId() string {
@@ -869,7 +939,7 @@ type HistoryResponse struct {
 
 func (x *HistoryResponse) Reset() {
 	*x = HistoryResponse{}
-	mi := &file_chat_proto_msgTypes[11]
+	mi := &file_chat_proto_msgTypes[12]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -881,7 +951,7 @@ func (x *HistoryResponse) String() string {
 func (*HistoryResponse) ProtoMessage() {}
 
 func (x *HistoryResponse) ProtoReflect() protoreflect.Message {
-	mi := &file_chat_proto_msgTypes[11]
+	mi := &file_chat_proto_msgTypes[12]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -894,7 +964,7 @@ func (x *HistoryResponse) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use HistoryResponse.ProtoReflect.Descriptor instead.
 func (*HistoryResponse) Descriptor() ([]byte, []int) {
-	return file_chat_proto_rawDescGZIP(), []int{11}
+	return file_chat_proto_rawDescGZIP(), []int{12}
 }
 
 func (x *HistoryResponse) GetMessages() []*Message {
@@ -913,15 +983,14 @@ func (x *HistoryResponse) GetHasMore() bool {
 
 type SubscribeRequest struct {
 	state         protoimpl.MessageState `protogen:"open.v1"`
-	ConsumerId    string                 `protobuf:"bytes,1,opt,name=consumer_id,json=consumerId,proto3" json:"consumer_id,omitempty"` // stable identity of the consumer/agent
-	TopicId       string                 `protobuf:"bytes,2,opt,name=topic_id,json=topicId,proto3" json:"topic_id,omitempty"`
+	TopicId       string                 `protobuf:"bytes,1,opt,name=topic_id,json=topicId,proto3" json:"topic_id,omitempty"`
 	unknownFields protoimpl.UnknownFields
 	sizeCache     protoimpl.SizeCache
 }
 
 func (x *SubscribeRequest) Reset() {
 	*x = SubscribeRequest{}
-	mi := &file_chat_proto_msgTypes[12]
+	mi := &file_chat_proto_msgTypes[13]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -933,7 +1002,7 @@ func (x *SubscribeRequest) String() string {
 func (*SubscribeRequest) ProtoMessage() {}
 
 func (x *SubscribeRequest) ProtoReflect() protoreflect.Message {
-	mi := &file_chat_proto_msgTypes[12]
+	mi := &file_chat_proto_msgTypes[13]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -946,14 +1015,7 @@ func (x *SubscribeRequest) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use SubscribeRequest.ProtoReflect.Descriptor instead.
 func (*SubscribeRequest) Descriptor() ([]byte, []int) {
-	return file_chat_proto_rawDescGZIP(), []int{12}
-}
-
-func (x *SubscribeRequest) GetConsumerId() string {
-	if x != nil {
-		return x.ConsumerId
-	}
-	return ""
+	return file_chat_proto_rawDescGZIP(), []int{13}
 }
 
 func (x *SubscribeRequest) GetTopicId() string {
@@ -963,18 +1025,90 @@ func (x *SubscribeRequest) GetTopicId() string {
 	return ""
 }
 
+// DeliveryMessage wraps one delivered message with delivery metadata.
+// message.created_at is always the original creation time; replay + the
+// timestamp let the SDK decide how to treat a delivery.
+type DeliveryMessage struct {
+	state   protoimpl.MessageState `protogen:"open.v1"`
+	Message *Message               `protobuf:"bytes,1,opt,name=message,proto3" json:"message,omitempty"`
+	// true when the message was committed BEFORE this subscription started
+	// (backlog catch-up after connect/reconnect); false for messages committed
+	// while this subscription is live.
+	Replay bool `protobuf:"varint,2,opt,name=replay,proto3" json:"replay,omitempty"`
+	// Opaque id of the current subscription, generated by the server per
+	// Subscribe stream. MUST be echoed in Acknowledge; a stale id (from a
+	// replaced/closed stream) is rejected.
+	SubscriptionId string `protobuf:"bytes,3,opt,name=subscription_id,json=subscriptionId,proto3" json:"subscription_id,omitempty"`
+	unknownFields  protoimpl.UnknownFields
+	sizeCache      protoimpl.SizeCache
+}
+
+func (x *DeliveryMessage) Reset() {
+	*x = DeliveryMessage{}
+	mi := &file_chat_proto_msgTypes[14]
+	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+	ms.StoreMessageInfo(mi)
+}
+
+func (x *DeliveryMessage) String() string {
+	return protoimpl.X.MessageStringOf(x)
+}
+
+func (*DeliveryMessage) ProtoMessage() {}
+
+func (x *DeliveryMessage) ProtoReflect() protoreflect.Message {
+	mi := &file_chat_proto_msgTypes[14]
+	if x != nil {
+		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
+		if ms.LoadMessageInfo() == nil {
+			ms.StoreMessageInfo(mi)
+		}
+		return ms
+	}
+	return mi.MessageOf(x)
+}
+
+// Deprecated: Use DeliveryMessage.ProtoReflect.Descriptor instead.
+func (*DeliveryMessage) Descriptor() ([]byte, []int) {
+	return file_chat_proto_rawDescGZIP(), []int{14}
+}
+
+func (x *DeliveryMessage) GetMessage() *Message {
+	if x != nil {
+		return x.Message
+	}
+	return nil
+}
+
+func (x *DeliveryMessage) GetReplay() bool {
+	if x != nil {
+		return x.Replay
+	}
+	return false
+}
+
+func (x *DeliveryMessage) GetSubscriptionId() string {
+	if x != nil {
+		return x.SubscriptionId
+	}
+	return ""
+}
+
 type AcknowledgeRequest struct {
-	state           protoimpl.MessageState `protogen:"open.v1"`
-	ConsumerId      string                 `protobuf:"bytes,1,opt,name=consumer_id,json=consumerId,proto3" json:"consumer_id,omitempty"`
-	TopicId         string                 `protobuf:"bytes,2,opt,name=topic_id,json=topicId,proto3" json:"topic_id,omitempty"`
-	LastAckSequence int64                  `protobuf:"varint,3,opt,name=last_ack_sequence,json=lastAckSequence,proto3" json:"last_ack_sequence,omitempty"` // highest sequence processed by the consumer
-	unknownFields   protoimpl.UnknownFields
-	sizeCache       protoimpl.SizeCache
+	state   protoimpl.MessageState `protogen:"open.v1"`
+	TopicId string                 `protobuf:"bytes,1,opt,name=topic_id,json=topicId,proto3" json:"topic_id,omitempty"`
+	// Highest sequence fully processed by the consumer (cumulative ack).
+	LastAckSequence int64 `protobuf:"varint,2,opt,name=last_ack_sequence,json=lastAckSequence,proto3" json:"last_ack_sequence,omitempty"`
+	// The subscription_id of the stream that delivered the message. Must be
+	// the CURRENT active subscription of this consumer in this topic.
+	SubscriptionId string `protobuf:"bytes,3,opt,name=subscription_id,json=subscriptionId,proto3" json:"subscription_id,omitempty"`
+	unknownFields  protoimpl.UnknownFields
+	sizeCache      protoimpl.SizeCache
 }
 
 func (x *AcknowledgeRequest) Reset() {
 	*x = AcknowledgeRequest{}
-	mi := &file_chat_proto_msgTypes[13]
+	mi := &file_chat_proto_msgTypes[15]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -986,7 +1120,7 @@ func (x *AcknowledgeRequest) String() string {
 func (*AcknowledgeRequest) ProtoMessage() {}
 
 func (x *AcknowledgeRequest) ProtoReflect() protoreflect.Message {
-	mi := &file_chat_proto_msgTypes[13]
+	mi := &file_chat_proto_msgTypes[15]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -999,14 +1133,7 @@ func (x *AcknowledgeRequest) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use AcknowledgeRequest.ProtoReflect.Descriptor instead.
 func (*AcknowledgeRequest) Descriptor() ([]byte, []int) {
-	return file_chat_proto_rawDescGZIP(), []int{13}
-}
-
-func (x *AcknowledgeRequest) GetConsumerId() string {
-	if x != nil {
-		return x.ConsumerId
-	}
-	return ""
+	return file_chat_proto_rawDescGZIP(), []int{15}
 }
 
 func (x *AcknowledgeRequest) GetTopicId() string {
@@ -1023,6 +1150,13 @@ func (x *AcknowledgeRequest) GetLastAckSequence() int64 {
 	return 0
 }
 
+func (x *AcknowledgeRequest) GetSubscriptionId() string {
+	if x != nil {
+		return x.SubscriptionId
+	}
+	return ""
+}
+
 type AcknowledgeResponse struct {
 	state           protoimpl.MessageState `protogen:"open.v1"`
 	LastAckSequence int64                  `protobuf:"varint,1,opt,name=last_ack_sequence,json=lastAckSequence,proto3" json:"last_ack_sequence,omitempty"` // the cursor as stored (never lower than before)
@@ -1032,7 +1166,7 @@ type AcknowledgeResponse struct {
 
 func (x *AcknowledgeResponse) Reset() {
 	*x = AcknowledgeResponse{}
-	mi := &file_chat_proto_msgTypes[14]
+	mi := &file_chat_proto_msgTypes[16]
 	ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 	ms.StoreMessageInfo(mi)
 }
@@ -1044,7 +1178,7 @@ func (x *AcknowledgeResponse) String() string {
 func (*AcknowledgeResponse) ProtoMessage() {}
 
 func (x *AcknowledgeResponse) ProtoReflect() protoreflect.Message {
-	mi := &file_chat_proto_msgTypes[14]
+	mi := &file_chat_proto_msgTypes[16]
 	if x != nil {
 		ms := protoimpl.X.MessageStateOf(protoimpl.Pointer(x))
 		if ms.LoadMessageInfo() == nil {
@@ -1057,7 +1191,7 @@ func (x *AcknowledgeResponse) ProtoReflect() protoreflect.Message {
 
 // Deprecated: Use AcknowledgeResponse.ProtoReflect.Descriptor instead.
 func (*AcknowledgeResponse) Descriptor() ([]byte, []int) {
-	return file_chat_proto_rawDescGZIP(), []int{14}
+	return file_chat_proto_rawDescGZIP(), []int{16}
 }
 
 func (x *AcknowledgeResponse) GetLastAckSequence() int64 {
@@ -1119,6 +1253,12 @@ const file_chat_proto_rawDesc = "" +
 	"\acontent\x18\x05 \x01(\tR\acontent\x122\n" +
 	"\x13reply_to_message_id\x18\x06 \x01(\tH\x00R\x10replyToMessageId\x88\x01\x01\x12'\n" +
 	"\x0fidempotency_key\x18\a \x01(\tR\x0eidempotencyKeyB\x16\n" +
+	"\x14_reply_to_message_id\"\xc3\x01\n" +
+	"\x17AgentPostMessageRequest\x12\x19\n" +
+	"\btopic_id\x18\x01 \x01(\tR\atopicId\x12\x18\n" +
+	"\acontent\x18\x02 \x01(\tR\acontent\x122\n" +
+	"\x13reply_to_message_id\x18\x03 \x01(\tH\x00R\x10replyToMessageId\x88\x01\x01\x12'\n" +
+	"\x0fidempotency_key\x18\x04 \x01(\tR\x0eidempotencyKeyB\x16\n" +
 	"\x14_reply_to_message_id\"^\n" +
 	"\x13PostMessageResponse\x12'\n" +
 	"\amessage\x18\x01 \x01(\v2\r.chat.MessageR\amessage\x12\x1e\n" +
@@ -1135,16 +1275,17 @@ const file_chat_proto_rawDesc = "" +
 	"\x05limit\x18\x04 \x01(\x05R\x05limit\"W\n" +
 	"\x0fHistoryResponse\x12)\n" +
 	"\bmessages\x18\x01 \x03(\v2\r.chat.MessageR\bmessages\x12\x19\n" +
-	"\bhas_more\x18\x02 \x01(\bR\ahasMore\"N\n" +
-	"\x10SubscribeRequest\x12\x1f\n" +
-	"\vconsumer_id\x18\x01 \x01(\tR\n" +
-	"consumerId\x12\x19\n" +
-	"\btopic_id\x18\x02 \x01(\tR\atopicId\"|\n" +
-	"\x12AcknowledgeRequest\x12\x1f\n" +
-	"\vconsumer_id\x18\x01 \x01(\tR\n" +
-	"consumerId\x12\x19\n" +
-	"\btopic_id\x18\x02 \x01(\tR\atopicId\x12*\n" +
-	"\x11last_ack_sequence\x18\x03 \x01(\x03R\x0flastAckSequence\"A\n" +
+	"\bhas_more\x18\x02 \x01(\bR\ahasMore\"-\n" +
+	"\x10SubscribeRequest\x12\x19\n" +
+	"\btopic_id\x18\x01 \x01(\tR\atopicId\"{\n" +
+	"\x0fDeliveryMessage\x12'\n" +
+	"\amessage\x18\x01 \x01(\v2\r.chat.MessageR\amessage\x12\x16\n" +
+	"\x06replay\x18\x02 \x01(\bR\x06replay\x12'\n" +
+	"\x0fsubscription_id\x18\x03 \x01(\tR\x0esubscriptionId\"\x84\x01\n" +
+	"\x12AcknowledgeRequest\x12\x19\n" +
+	"\btopic_id\x18\x01 \x01(\tR\atopicId\x12*\n" +
+	"\x11last_ack_sequence\x18\x02 \x01(\x03R\x0flastAckSequence\x12'\n" +
+	"\x0fsubscription_id\x18\x03 \x01(\tR\x0esubscriptionId\"A\n" +
 	"\x13AcknowledgeResponse\x12*\n" +
 	"\x11last_ack_sequence\x18\x01 \x01(\x03R\x0flastAckSequence*\x8c\x01\n" +
 	"\n" +
@@ -1159,18 +1300,19 @@ const file_chat_proto_rawDesc = "" +
 	"\x12MESSAGE_SOURCE_WEB\x10\x01\x12\x1b\n" +
 	"\x17MESSAGE_SOURCE_TELEGRAM\x10\x02\x12\x18\n" +
 	"\x14MESSAGE_SOURCE_AGENT\x10\x03\x12\x19\n" +
-	"\x15MESSAGE_SOURCE_SYSTEM\x10\x042\x96\x04\n" +
+	"\x15MESSAGE_SOURCE_SYSTEM\x10\x042\xec\x04\n" +
 	"\vChatService\x12?\n" +
 	"\n" +
 	"ListTopics\x12\x17.chat.ListTopicsRequest\x1a\x18.chat.ListTopicsResponse\x12.\n" +
 	"\bGetTopic\x12\x15.chat.GetTopicRequest\x1a\v.chat.Topic\x124\n" +
 	"\vCreateTopic\x12\x18.chat.CreateTopicRequest\x1a\v.chat.Topic\x124\n" +
 	"\vUpdateTopic\x12\x18.chat.UpdateTopicRequest\x1a\v.chat.Topic\x12B\n" +
-	"\vPostMessage\x12\x18.chat.PostMessageRequest\x1a\x19.chat.PostMessageResponse\x124\n" +
+	"\vPostMessage\x12\x18.chat.PostMessageRequest\x1a\x19.chat.PostMessageResponse\x12L\n" +
+	"\x10AgentPostMessage\x12\x1d.chat.AgentPostMessageRequest\x1a\x19.chat.PostMessageResponse\x124\n" +
 	"\n" +
 	"GetMessage\x12\x17.chat.GetMessageRequest\x1a\r.chat.Message\x126\n" +
-	"\aHistory\x12\x14.chat.HistoryRequest\x1a\x15.chat.HistoryResponse\x124\n" +
-	"\tSubscribe\x12\x16.chat.SubscribeRequest\x1a\r.chat.Message0\x01\x12B\n" +
+	"\aHistory\x12\x14.chat.HistoryRequest\x1a\x15.chat.HistoryResponse\x12<\n" +
+	"\tSubscribe\x12\x16.chat.SubscribeRequest\x1a\x15.chat.DeliveryMessage0\x01\x12B\n" +
 	"\vAcknowledge\x12\x18.chat.AcknowledgeRequest\x1a\x19.chat.AcknowledgeResponseB,Z*git.be.kaien.ru/l-ai/l-ai-chat-proto/protob\x06proto3"
 
 var (
@@ -1186,60 +1328,65 @@ func file_chat_proto_rawDescGZIP() []byte {
 }
 
 var file_chat_proto_enumTypes = make([]protoimpl.EnumInfo, 2)
-var file_chat_proto_msgTypes = make([]protoimpl.MessageInfo, 15)
+var file_chat_proto_msgTypes = make([]protoimpl.MessageInfo, 17)
 var file_chat_proto_goTypes = []any{
-	(AuthorType)(0),               // 0: chat.AuthorType
-	(MessageSource)(0),            // 1: chat.MessageSource
-	(*Topic)(nil),                 // 2: chat.Topic
-	(*Message)(nil),               // 3: chat.Message
-	(*ListTopicsRequest)(nil),     // 4: chat.ListTopicsRequest
-	(*ListTopicsResponse)(nil),    // 5: chat.ListTopicsResponse
-	(*GetTopicRequest)(nil),       // 6: chat.GetTopicRequest
-	(*CreateTopicRequest)(nil),    // 7: chat.CreateTopicRequest
-	(*UpdateTopicRequest)(nil),    // 8: chat.UpdateTopicRequest
-	(*PostMessageRequest)(nil),    // 9: chat.PostMessageRequest
-	(*PostMessageResponse)(nil),   // 10: chat.PostMessageResponse
-	(*GetMessageRequest)(nil),     // 11: chat.GetMessageRequest
-	(*HistoryRequest)(nil),        // 12: chat.HistoryRequest
-	(*HistoryResponse)(nil),       // 13: chat.HistoryResponse
-	(*SubscribeRequest)(nil),      // 14: chat.SubscribeRequest
-	(*AcknowledgeRequest)(nil),    // 15: chat.AcknowledgeRequest
-	(*AcknowledgeResponse)(nil),   // 16: chat.AcknowledgeResponse
-	(*timestamppb.Timestamp)(nil), // 17: google.protobuf.Timestamp
+	(AuthorType)(0),                 // 0: chat.AuthorType
+	(MessageSource)(0),              // 1: chat.MessageSource
+	(*Topic)(nil),                   // 2: chat.Topic
+	(*Message)(nil),                 // 3: chat.Message
+	(*ListTopicsRequest)(nil),       // 4: chat.ListTopicsRequest
+	(*ListTopicsResponse)(nil),      // 5: chat.ListTopicsResponse
+	(*GetTopicRequest)(nil),         // 6: chat.GetTopicRequest
+	(*CreateTopicRequest)(nil),      // 7: chat.CreateTopicRequest
+	(*UpdateTopicRequest)(nil),      // 8: chat.UpdateTopicRequest
+	(*PostMessageRequest)(nil),      // 9: chat.PostMessageRequest
+	(*AgentPostMessageRequest)(nil), // 10: chat.AgentPostMessageRequest
+	(*PostMessageResponse)(nil),     // 11: chat.PostMessageResponse
+	(*GetMessageRequest)(nil),       // 12: chat.GetMessageRequest
+	(*HistoryRequest)(nil),          // 13: chat.HistoryRequest
+	(*HistoryResponse)(nil),         // 14: chat.HistoryResponse
+	(*SubscribeRequest)(nil),        // 15: chat.SubscribeRequest
+	(*DeliveryMessage)(nil),         // 16: chat.DeliveryMessage
+	(*AcknowledgeRequest)(nil),      // 17: chat.AcknowledgeRequest
+	(*AcknowledgeResponse)(nil),     // 18: chat.AcknowledgeResponse
+	(*timestamppb.Timestamp)(nil),   // 19: google.protobuf.Timestamp
 }
 var file_chat_proto_depIdxs = []int32{
-	17, // 0: chat.Topic.created_at:type_name -> google.protobuf.Timestamp
+	19, // 0: chat.Topic.created_at:type_name -> google.protobuf.Timestamp
 	0,  // 1: chat.Message.author_type:type_name -> chat.AuthorType
 	1,  // 2: chat.Message.source:type_name -> chat.MessageSource
-	17, // 3: chat.Message.created_at:type_name -> google.protobuf.Timestamp
+	19, // 3: chat.Message.created_at:type_name -> google.protobuf.Timestamp
 	2,  // 4: chat.ListTopicsResponse.topics:type_name -> chat.Topic
 	0,  // 5: chat.PostMessageRequest.author_type:type_name -> chat.AuthorType
 	1,  // 6: chat.PostMessageRequest.source:type_name -> chat.MessageSource
 	3,  // 7: chat.PostMessageResponse.message:type_name -> chat.Message
 	3,  // 8: chat.HistoryResponse.messages:type_name -> chat.Message
-	4,  // 9: chat.ChatService.ListTopics:input_type -> chat.ListTopicsRequest
-	6,  // 10: chat.ChatService.GetTopic:input_type -> chat.GetTopicRequest
-	7,  // 11: chat.ChatService.CreateTopic:input_type -> chat.CreateTopicRequest
-	8,  // 12: chat.ChatService.UpdateTopic:input_type -> chat.UpdateTopicRequest
-	9,  // 13: chat.ChatService.PostMessage:input_type -> chat.PostMessageRequest
-	11, // 14: chat.ChatService.GetMessage:input_type -> chat.GetMessageRequest
-	12, // 15: chat.ChatService.History:input_type -> chat.HistoryRequest
-	14, // 16: chat.ChatService.Subscribe:input_type -> chat.SubscribeRequest
-	15, // 17: chat.ChatService.Acknowledge:input_type -> chat.AcknowledgeRequest
-	5,  // 18: chat.ChatService.ListTopics:output_type -> chat.ListTopicsResponse
-	2,  // 19: chat.ChatService.GetTopic:output_type -> chat.Topic
-	2,  // 20: chat.ChatService.CreateTopic:output_type -> chat.Topic
-	2,  // 21: chat.ChatService.UpdateTopic:output_type -> chat.Topic
-	10, // 22: chat.ChatService.PostMessage:output_type -> chat.PostMessageResponse
-	3,  // 23: chat.ChatService.GetMessage:output_type -> chat.Message
-	13, // 24: chat.ChatService.History:output_type -> chat.HistoryResponse
-	3,  // 25: chat.ChatService.Subscribe:output_type -> chat.Message
-	16, // 26: chat.ChatService.Acknowledge:output_type -> chat.AcknowledgeResponse
-	18, // [18:27] is the sub-list for method output_type
-	9,  // [9:18] is the sub-list for method input_type
-	9,  // [9:9] is the sub-list for extension type_name
-	9,  // [9:9] is the sub-list for extension extendee
-	0,  // [0:9] is the sub-list for field type_name
+	3,  // 9: chat.DeliveryMessage.message:type_name -> chat.Message
+	4,  // 10: chat.ChatService.ListTopics:input_type -> chat.ListTopicsRequest
+	6,  // 11: chat.ChatService.GetTopic:input_type -> chat.GetTopicRequest
+	7,  // 12: chat.ChatService.CreateTopic:input_type -> chat.CreateTopicRequest
+	8,  // 13: chat.ChatService.UpdateTopic:input_type -> chat.UpdateTopicRequest
+	9,  // 14: chat.ChatService.PostMessage:input_type -> chat.PostMessageRequest
+	10, // 15: chat.ChatService.AgentPostMessage:input_type -> chat.AgentPostMessageRequest
+	12, // 16: chat.ChatService.GetMessage:input_type -> chat.GetMessageRequest
+	13, // 17: chat.ChatService.History:input_type -> chat.HistoryRequest
+	15, // 18: chat.ChatService.Subscribe:input_type -> chat.SubscribeRequest
+	17, // 19: chat.ChatService.Acknowledge:input_type -> chat.AcknowledgeRequest
+	5,  // 20: chat.ChatService.ListTopics:output_type -> chat.ListTopicsResponse
+	2,  // 21: chat.ChatService.GetTopic:output_type -> chat.Topic
+	2,  // 22: chat.ChatService.CreateTopic:output_type -> chat.Topic
+	2,  // 23: chat.ChatService.UpdateTopic:output_type -> chat.Topic
+	11, // 24: chat.ChatService.PostMessage:output_type -> chat.PostMessageResponse
+	11, // 25: chat.ChatService.AgentPostMessage:output_type -> chat.PostMessageResponse
+	3,  // 26: chat.ChatService.GetMessage:output_type -> chat.Message
+	14, // 27: chat.ChatService.History:output_type -> chat.HistoryResponse
+	16, // 28: chat.ChatService.Subscribe:output_type -> chat.DeliveryMessage
+	18, // 29: chat.ChatService.Acknowledge:output_type -> chat.AcknowledgeResponse
+	20, // [20:30] is the sub-list for method output_type
+	10, // [10:20] is the sub-list for method input_type
+	10, // [10:10] is the sub-list for extension type_name
+	10, // [10:10] is the sub-list for extension extendee
+	0,  // [0:10] is the sub-list for field type_name
 }
 
 func init() { file_chat_proto_init() }
@@ -1253,13 +1400,14 @@ func file_chat_proto_init() {
 		(*GetTopicRequest_Slug)(nil),
 	}
 	file_chat_proto_msgTypes[7].OneofWrappers = []any{}
+	file_chat_proto_msgTypes[8].OneofWrappers = []any{}
 	type x struct{}
 	out := protoimpl.TypeBuilder{
 		File: protoimpl.DescBuilder{
 			GoPackagePath: reflect.TypeOf(x{}).PkgPath(),
 			RawDescriptor: unsafe.Slice(unsafe.StringData(file_chat_proto_rawDesc), len(file_chat_proto_rawDesc)),
 			NumEnums:      2,
-			NumMessages:   15,
+			NumMessages:   17,
 			NumExtensions: 0,
 			NumServices:   1,
 		},
diff --git a/chat.proto b/chat.proto
index a3b14b9..d2eb8ed 100644
--- a/chat.proto
+++ b/chat.proto
@@ -24,11 +24,22 @@ service ChatService {
   // there is no delete API.
   rpc UpdateTopic(UpdateTopicRequest) returns (Topic);
 
-  // PostMessage appends a message to a topic. Repeated calls with the same
-  // (author_id, idempotency_key) must not create a second message; the
-  // original message is returned with duplicated=true instead.
+  // PostMessage appends a message to a topic. NOTE: this RPC is fail-closed
+  // until the human-authentication stage: there is no trusted way to fill
+  // author_id/author_type/source yet, so any call currently returns
+  // Unimplemented. Agents MUST use AgentPostMessage (the server derives the
+  // identity from the service token). The internal canonical-message path is
+  // re-exposed after human auth lands.
   rpc PostMessage(PostMessageRequest) returns (PostMessageResponse);
 
+  // AgentPostMessage posts a message as the authenticated agent. The server
+  // derives every server-owned field: author_id (deterministic UUID of the
+  // authenticated identity), author_type=AGENT, source=AGENT, sequence,
+  // root_message_id, reply_depth, created_at. Clients cannot spoof them.
+  // idempotency_key is mandatory: retries with the same key return the
+  // original message (duplicated=true).
+  rpc AgentPostMessage(AgentPostMessageRequest) returns (PostMessageResponse);
+
   // GetMessage returns a single message by id.
   rpc GetMessage(GetMessageRequest) returns (Message);
 
@@ -37,13 +48,26 @@ service ChatService {
   // after_sequence for oldest-first forward paging; at most one may be set.
   rpc History(HistoryRequest) returns (HistoryResponse);
 
-  // Subscribe streams new messages of a topic to a consumer. The consumer is
-  // expected to Acknowledge the messages it processed (at-least-once).
-  // NOTE: stream delivery is implemented in a later stage.
-  rpc Subscribe(SubscribeRequest) returns (stream Message);
+  // Subscribe opens the delivery stream of a topic for the authenticated
+  // consumer. Delivery is durable at-least-once: PostgreSQL (messages +
+  // per-(consumer, topic) delivery cursor) is the source of correctness;
+  // the stream is a live projection. Exactly one active subscription is
+  // admitted per (consumer, topic): a new Subscribe for the same pair
+  // replaces the previous one, which terminates with Aborted. Every
+  // DeliveryMessage carries the opaque subscription_id the consumer must
+  // echo in Acknowledge. Messages are delivered in ascending sequence
+  // order; gaps between sequences of one topic are normal (other topics
+  // consume global sequence values).
+  rpc Subscribe(SubscribeRequest) returns (stream DeliveryMessage);
 
-  // Acknowledge advances the durable delivery cursor of a consumer in a
-  // topic. Cursors are monotonic: the stored value never moves backwards.
+  // Acknowledge advances the durable delivery cursor of the CURRENT active
+  // subscription of the authenticated consumer in a topic. Cumulative:
+  // last_ack_sequence = highest sequence the consumer has fully processed.
+  // The ACK is validated against the active subscription: the sequence must
+  // not exceed the highest sequence this subscription actually delivered,
+  // and a stale/old subscription_id is rejected. The cursor is monotonic —
+  // it never moves backwards. Cursor changes ONLY here; delivery itself
+  // never persists cursor state.
   rpc Acknowledge(AcknowledgeRequest) returns (AcknowledgeResponse);
 }
 
@@ -86,7 +110,7 @@ message Message {
   optional string reply_to_message_id = 8; // parent message when this is a reply
   string root_message_id = 9; // root of the reply chain (== id for non-replies)
   int32 reply_depth = 10; // 0 for non-replies, parent depth + 1 for replies; hard max 20
-  google.protobuf.Timestamp created_at = 11;
+  google.protobuf.Timestamp created_at = 11; // original creation time; age = now - created_at
 }
 
 message ListTopicsRequest {}
@@ -124,6 +148,15 @@ message PostMessageRequest {
   string idempotency_key = 7;
 }
 
+message AgentPostMessageRequest {
+  string topic_id = 1;
+  string content = 2;
+  optional string reply_to_message_id = 3;
+  // Mandatory: retrying with the same key returns the original message
+  // (duplicated=true) instead of creating a duplicate.
+  string idempotency_key = 4;
+}
+
 message PostMessageResponse {
   Message message = 1;
   bool duplicated = 2; // true when an idempotent replay returned an existing message
@@ -152,14 +185,34 @@ message HistoryResponse {
 }
 
 message SubscribeRequest {
-  string consumer_id = 1; // stable identity of the consumer/agent
-  string topic_id = 2;
+  string topic_id = 1;
+  // The consumer identity is NOT a request field: it is derived from the
+  // authenticated service credential, so one agent cannot subscribe as
+  // another.
+}
+
+// DeliveryMessage wraps one delivered message with delivery metadata.
+// message.created_at is always the original creation time; replay + the
+// timestamp let the SDK decide how to treat a delivery.
+message DeliveryMessage {
+  Message message = 1;
+  // true when the message was committed BEFORE this subscription started
+  // (backlog catch-up after connect/reconnect); false for messages committed
+  // while this subscription is live.
+  bool replay = 2;
+  // Opaque id of the current subscription, generated by the server per
+  // Subscribe stream. MUST be echoed in Acknowledge; a stale id (from a
+  // replaced/closed stream) is rejected.
+  string subscription_id = 3;
 }
 
 message AcknowledgeRequest {
-  string consumer_id = 1;
-  string topic_id = 2;
-  int64 last_ack_sequence = 3; // highest sequence processed by the consumer
+  string topic_id = 1;
+  // Highest sequence fully processed by the consumer (cumulative ack).
+  int64 last_ack_sequence = 2;
+  // The subscription_id of the stream that delivered the message. Must be
+  // the CURRENT active subscription of this consumer in this topic.
+  string subscription_id = 3;
 }
 
 message AcknowledgeResponse {
diff --git a/chat_grpc.pb.go b/chat_grpc.pb.go
index 2c9f706..05f4d94 100644
--- a/chat_grpc.pb.go
+++ b/chat_grpc.pb.go
@@ -19,15 +19,16 @@ import (
 const _ = grpc.SupportPackageIsVersion9
 
 const (
-	ChatService_ListTopics_FullMethodName  = "/chat.ChatService/ListTopics"
-	ChatService_GetTopic_FullMethodName    = "/chat.ChatService/GetTopic"
-	ChatService_CreateTopic_FullMethodName = "/chat.ChatService/CreateTopic"
-	ChatService_UpdateTopic_FullMethodName = "/chat.ChatService/UpdateTopic"
-	ChatService_PostMessage_FullMethodName = "/chat.ChatService/PostMessage"
-	ChatService_GetMessage_FullMethodName  = "/chat.ChatService/GetMessage"
-	ChatService_History_FullMethodName     = "/chat.ChatService/History"
-	ChatService_Subscribe_FullMethodName   = "/chat.ChatService/Subscribe"
-	ChatService_Acknowledge_FullMethodName = "/chat.ChatService/Acknowledge"
+	ChatService_ListTopics_FullMethodName       = "/chat.ChatService/ListTopics"
+	ChatService_GetTopic_FullMethodName         = "/chat.ChatService/GetTopic"
+	ChatService_CreateTopic_FullMethodName      = "/chat.ChatService/CreateTopic"
+	ChatService_UpdateTopic_FullMethodName      = "/chat.ChatService/UpdateTopic"
+	ChatService_PostMessage_FullMethodName      = "/chat.ChatService/PostMessage"
+	ChatService_AgentPostMessage_FullMethodName = "/chat.ChatService/AgentPostMessage"
+	ChatService_GetMessage_FullMethodName       = "/chat.ChatService/GetMessage"
+	ChatService_History_FullMethodName          = "/chat.ChatService/History"
+	ChatService_Subscribe_FullMethodName        = "/chat.ChatService/Subscribe"
+	ChatService_Acknowledge_FullMethodName      = "/chat.ChatService/Acknowledge"
 )
 
 // ChatServiceClient is the client API for ChatService service.
@@ -48,22 +49,45 @@ type ChatServiceClient interface {
 	// identifier and cannot be changed. Topics are persistent in the MVP:
 	// there is no delete API.
 	UpdateTopic(ctx context.Context, in *UpdateTopicRequest, opts ...grpc.CallOption) (*Topic, error)
-	// PostMessage appends a message to a topic. Repeated calls with the same
-	// (author_id, idempotency_key) must not create a second message; the
-	// original message is returned with duplicated=true instead.
+	// PostMessage appends a message to a topic. NOTE: this RPC is fail-closed
+	// until the human-authentication stage: there is no trusted way to fill
+	// author_id/author_type/source yet, so any call currently returns
+	// Unimplemented. Agents MUST use AgentPostMessage (the server derives the
+	// identity from the service token). The internal canonical-message path is
+	// re-exposed after human auth lands.
 	PostMessage(ctx context.Context, in *PostMessageRequest, opts ...grpc.CallOption) (*PostMessageResponse, error)
+	// AgentPostMessage posts a message as the authenticated agent. The server
+	// derives every server-owned field: author_id (deterministic UUID of the
+	// authenticated identity), author_type=AGENT, source=AGENT, sequence,
+	// root_message_id, reply_depth, created_at. Clients cannot spoof them.
+	// idempotency_key is mandatory: retries with the same key return the
+	// original message (duplicated=true).
+	AgentPostMessage(ctx context.Context, in *AgentPostMessageRequest, opts ...grpc.CallOption) (*PostMessageResponse, error)
 	// GetMessage returns a single message by id.
 	GetMessage(ctx context.Context, in *GetMessageRequest, opts ...grpc.CallOption) (*Message, error)
 	// History returns messages of a topic, paginated by sequence (no offset
 	// pagination). Use before_sequence for newest-first backward paging or
 	// after_sequence for oldest-first forward paging; at most one may be set.
 	History(ctx context.Context, in *HistoryRequest, opts ...grpc.CallOption) (*HistoryResponse, error)
-	// Subscribe streams new messages of a topic to a consumer. The consumer is
-	// expected to Acknowledge the messages it processed (at-least-once).
-	// NOTE: stream delivery is implemented in a later stage.
-	Subscribe(ctx context.Context, in *SubscribeRequest, opts ...grpc.CallOption) (grpc.ServerStreamingClient[Message], error)
-	// Acknowledge advances the durable delivery cursor of a consumer in a
-	// topic. Cursors are monotonic: the stored value never moves backwards.
+	// Subscribe opens the delivery stream of a topic for the authenticated
+	// consumer. Delivery is durable at-least-once: PostgreSQL (messages +
+	// per-(consumer, topic) delivery cursor) is the source of correctness;
+	// the stream is a live projection. Exactly one active subscription is
+	// admitted per (consumer, topic): a new Subscribe for the same pair
+	// replaces the previous one, which terminates with Aborted. Every
+	// DeliveryMessage carries the opaque subscription_id the consumer must
+	// echo in Acknowledge. Messages are delivered in ascending sequence
+	// order; gaps between sequences of one topic are normal (other topics
+	// consume global sequence values).
+	Subscribe(ctx context.Context, in *SubscribeRequest, opts ...grpc.CallOption) (grpc.ServerStreamingClient[DeliveryMessage], error)
+	// Acknowledge advances the durable delivery cursor of the CURRENT active
+	// subscription of the authenticated consumer in a topic. Cumulative:
+	// last_ack_sequence = highest sequence the consumer has fully processed.
+	// The ACK is validated against the active subscription: the sequence must
+	// not exceed the highest sequence this subscription actually delivered,
+	// and a stale/old subscription_id is rejected. The cursor is monotonic —
+	// it never moves backwards. Cursor changes ONLY here; delivery itself
+	// never persists cursor state.
 	Acknowledge(ctx context.Context, in *AcknowledgeRequest, opts ...grpc.CallOption) (*AcknowledgeResponse, error)
 }
 
@@ -125,6 +149,16 @@ func (c *chatServiceClient) PostMessage(ctx context.Context, in *PostMessageRequ
 	return out, nil
 }
 
+func (c *chatServiceClient) AgentPostMessage(ctx context.Context, in *AgentPostMessageRequest, opts ...grpc.CallOption) (*PostMessageResponse, error) {
+	cOpts := append([]grpc.CallOption{grpc.StaticMethod()}, opts...)
+	out := new(PostMessageResponse)
+	err := c.cc.Invoke(ctx, ChatService_AgentPostMessage_FullMethodName, in, out, cOpts...)
+	if err != nil {
+		return nil, err
+	}
+	return out, nil
+}
+
 func (c *chatServiceClient) GetMessage(ctx context.Context, in *GetMessageRequest, opts ...grpc.CallOption) (*Message, error) {
 	cOpts := append([]grpc.CallOption{grpc.StaticMethod()}, opts...)
 	out := new(Message)
@@ -145,13 +179,13 @@ func (c *chatServiceClient) History(ctx context.Context, in *HistoryRequest, opt
 	return out, nil
 }
 
-func (c *chatServiceClient) Subscribe(ctx context.Context, in *SubscribeRequest, opts ...grpc.CallOption) (grpc.ServerStreamingClient[Message], error) {
+func (c *chatServiceClient) Subscribe(ctx context.Context, in *SubscribeRequest, opts ...grpc.CallOption) (grpc.ServerStreamingClient[DeliveryMessage], error) {
 	cOpts := append([]grpc.CallOption{grpc.StaticMethod()}, opts...)
 	stream, err := c.cc.NewStream(ctx, &ChatService_ServiceDesc.Streams[0], ChatService_Subscribe_FullMethodName, cOpts...)
 	if err != nil {
 		return nil, err
 	}
-	x := &grpc.GenericClientStream[SubscribeRequest, Message]{ClientStream: stream}
+	x := &grpc.GenericClientStream[SubscribeRequest, DeliveryMessage]{ClientStream: stream}
 	if err := x.ClientStream.SendMsg(in); err != nil {
 		return nil, err
 	}
@@ -162,7 +196,7 @@ func (c *chatServiceClient) Subscribe(ctx context.Context, in *SubscribeRequest,
 }
 
 // This type alias is provided for backwards compatibility with existing code that references the prior non-generic stream type by name.
-type ChatService_SubscribeClient = grpc.ServerStreamingClient[Message]
+type ChatService_SubscribeClient = grpc.ServerStreamingClient[DeliveryMessage]
 
 func (c *chatServiceClient) Acknowledge(ctx context.Context, in *AcknowledgeRequest, opts ...grpc.CallOption) (*AcknowledgeResponse, error) {
 	cOpts := append([]grpc.CallOption{grpc.StaticMethod()}, opts...)
@@ -192,22 +226,45 @@ type ChatServiceServer interface {
 	// identifier and cannot be changed. Topics are persistent in the MVP:
 	// there is no delete API.
 	UpdateTopic(context.Context, *UpdateTopicRequest) (*Topic, error)
-	// PostMessage appends a message to a topic. Repeated calls with the same
-	// (author_id, idempotency_key) must not create a second message; the
-	// original message is returned with duplicated=true instead.
+	// PostMessage appends a message to a topic. NOTE: this RPC is fail-closed
+	// until the human-authentication stage: there is no trusted way to fill
+	// author_id/author_type/source yet, so any call currently returns
+	// Unimplemented. Agents MUST use AgentPostMessage (the server derives the
+	// identity from the service token). The internal canonical-message path is
+	// re-exposed after human auth lands.
 	PostMessage(context.Context, *PostMessageRequest) (*PostMessageResponse, error)
+	// AgentPostMessage posts a message as the authenticated agent. The server
+	// derives every server-owned field: author_id (deterministic UUID of the
+	// authenticated identity), author_type=AGENT, source=AGENT, sequence,
+	// root_message_id, reply_depth, created_at. Clients cannot spoof them.
+	// idempotency_key is mandatory: retries with the same key return the
+	// original message (duplicated=true).
+	AgentPostMessage(context.Context, *AgentPostMessageRequest) (*PostMessageResponse, error)
 	// GetMessage returns a single message by id.
 	GetMessage(context.Context, *GetMessageRequest) (*Message, error)
 	// History returns messages of a topic, paginated by sequence (no offset
 	// pagination). Use before_sequence for newest-first backward paging or
 	// after_sequence for oldest-first forward paging; at most one may be set.
 	History(context.Context, *HistoryRequest) (*HistoryResponse, error)
-	// Subscribe streams new messages of a topic to a consumer. The consumer is
-	// expected to Acknowledge the messages it processed (at-least-once).
-	// NOTE: stream delivery is implemented in a later stage.
-	Subscribe(*SubscribeRequest, grpc.ServerStreamingServer[Message]) error
-	// Acknowledge advances the durable delivery cursor of a consumer in a
-	// topic. Cursors are monotonic: the stored value never moves backwards.
+	// Subscribe opens the delivery stream of a topic for the authenticated
+	// consumer. Delivery is durable at-least-once: PostgreSQL (messages +
+	// per-(consumer, topic) delivery cursor) is the source of correctness;
+	// the stream is a live projection. Exactly one active subscription is
+	// admitted per (consumer, topic): a new Subscribe for the same pair
+	// replaces the previous one, which terminates with Aborted. Every
+	// DeliveryMessage carries the opaque subscription_id the consumer must
+	// echo in Acknowledge. Messages are delivered in ascending sequence
+	// order; gaps between sequences of one topic are normal (other topics
+	// consume global sequence values).
+	Subscribe(*SubscribeRequest, grpc.ServerStreamingServer[DeliveryMessage]) error
+	// Acknowledge advances the durable delivery cursor of the CURRENT active
+	// subscription of the authenticated consumer in a topic. Cumulative:
+	// last_ack_sequence = highest sequence the consumer has fully processed.
+	// The ACK is validated against the active subscription: the sequence must
+	// not exceed the highest sequence this subscription actually delivered,
+	// and a stale/old subscription_id is rejected. The cursor is monotonic —
+	// it never moves backwards. Cursor changes ONLY here; delivery itself
+	// never persists cursor state.
 	Acknowledge(context.Context, *AcknowledgeRequest) (*AcknowledgeResponse, error)
 	mustEmbedUnimplementedChatServiceServer()
 }
@@ -234,13 +291,16 @@ func (UnimplementedChatServiceServer) UpdateTopic(context.Context, *UpdateTopicR
 func (UnimplementedChatServiceServer) PostMessage(context.Context, *PostMessageRequest) (*PostMessageResponse, error) {
 	return nil, status.Error(codes.Unimplemented, "method PostMessage not implemented")
 }
+func (UnimplementedChatServiceServer) AgentPostMessage(context.Context, *AgentPostMessageRequest) (*PostMessageResponse, error) {
+	return nil, status.Error(codes.Unimplemented, "method AgentPostMessage not implemented")
+}
 func (UnimplementedChatServiceServer) GetMessage(context.Context, *GetMessageRequest) (*Message, error) {
 	return nil, status.Error(codes.Unimplemented, "method GetMessage not implemented")
 }
 func (UnimplementedChatServiceServer) History(context.Context, *HistoryRequest) (*HistoryResponse, error) {
 	return nil, status.Error(codes.Unimplemented, "method History not implemented")
 }
-func (UnimplementedChatServiceServer) Subscribe(*SubscribeRequest, grpc.ServerStreamingServer[Message]) error {
+func (UnimplementedChatServiceServer) Subscribe(*SubscribeRequest, grpc.ServerStreamingServer[DeliveryMessage]) error {
 	return status.Error(codes.Unimplemented, "method Subscribe not implemented")
 }
 func (UnimplementedChatServiceServer) Acknowledge(context.Context, *AcknowledgeRequest) (*AcknowledgeResponse, error) {
@@ -357,6 +417,24 @@ func _ChatService_PostMessage_Handler(srv interface{}, ctx context.Context, dec
 	return interceptor(ctx, in, info, handler)
 }
 
+func _ChatService_AgentPostMessage_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
+	in := new(AgentPostMessageRequest)
+	if err := dec(in); err != nil {
+		return nil, err
+	}
+	if interceptor == nil {
+		return srv.(ChatServiceServer).AgentPostMessage(ctx, in)
+	}
+	info := &grpc.UnaryServerInfo{
+		Server:     srv,
+		FullMethod: ChatService_AgentPostMessage_FullMethodName,
+	}
+	handler := func(ctx context.Context, req interface{}) (interface{}, error) {
+		return srv.(ChatServiceServer).AgentPostMessage(ctx, req.(*AgentPostMessageRequest))
+	}
+	return interceptor(ctx, in, info, handler)
+}
+
 func _ChatService_GetMessage_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
 	in := new(GetMessageRequest)
 	if err := dec(in); err != nil {
@@ -398,11 +476,11 @@ func _ChatService_Subscribe_Handler(srv interface{}, stream grpc.ServerStream) e
 	if err := stream.RecvMsg(m); err != nil {
 		return err
 	}
-	return srv.(ChatServiceServer).Subscribe(m, &grpc.GenericServerStream[SubscribeRequest, Message]{ServerStream: stream})
+	return srv.(ChatServiceServer).Subscribe(m, &grpc.GenericServerStream[SubscribeRequest, DeliveryMessage]{ServerStream: stream})
 }
 
 // This type alias is provided for backwards compatibility with existing code that references the prior non-generic stream type by name.
-type ChatService_SubscribeServer = grpc.ServerStreamingServer[Message]
+type ChatService_SubscribeServer = grpc.ServerStreamingServer[DeliveryMessage]
 
 func _ChatService_Acknowledge_Handler(srv interface{}, ctx context.Context, dec func(interface{}) error, interceptor grpc.UnaryServerInterceptor) (interface{}, error) {
 	in := new(AcknowledgeRequest)
@@ -449,6 +527,10 @@ var ChatService_ServiceDesc = grpc.ServiceDesc{
 			MethodName: "PostMessage",
 			Handler:    _ChatService_PostMessage_Handler,
 		},
+		{
+			MethodName: "AgentPostMessage",
+			Handler:    _ChatService_AgentPostMessage_Handler,
+		},
 		{
 			MethodName: "GetMessage",
 			Handler:    _ChatService_GetMessage_Handler,