feat: max context tokens

db/migrations/0001_init.up.sql

@@ -117,6 +117,7 @@ CREATE TABLE IF NOT EXISTS bots (
   is_active BOOLEAN NOT NULL DEFAULT true,
   status TEXT NOT NULL DEFAULT 'ready',
   max_context_load_time INTEGER NOT NULL DEFAULT 1440,
+  max_context_tokens INTEGER NOT NULL DEFAULT 0,
   language TEXT NOT NULL DEFAULT 'auto',
   allow_guest BOOLEAN NOT NULL DEFAULT false,
   chat_model_id UUID REFERENCES models(id) ON DELETE SET NULL,
@@ -242,6 +243,7 @@ CREATE TABLE IF NOT EXISTS bot_history_messages (
   role TEXT NOT NULL CHECK (role IN ('user', 'assistant', 'system', 'tool')),
   content JSONB NOT NULL,
   metadata JSONB NOT NULL DEFAULT '{}'::jsonb,
+  usage JSONB,
   created_at TIMESTAMPTZ NOT NULL DEFAULT now()
 );
 

db/migrations/0008_max_context_tokens.down.sql

@@ -0,0 +1,4 @@
+-- 0008_max_context_tokens
+-- Remove max_context_tokens column from bots table
+
+ALTER TABLE bots DROP COLUMN IF EXISTS max_context_tokens;

db/migrations/0008_max_context_tokens.up.sql

@@ -0,0 +1,4 @@
+-- 0008_max_context_tokens
+-- Add max_context_tokens column to bots table for token-based context trimming
+
+ALTER TABLE bots ADD COLUMN IF NOT EXISTS max_context_tokens INTEGER NOT NULL DEFAULT 0;

db/migrations/0009_message_usage.down.sql

@@ -0,0 +1,4 @@
+-- 0009_message_usage
+-- Remove usage JSONB column from bot_history_messages
+
+ALTER TABLE bot_history_messages DROP COLUMN IF EXISTS usage;

db/migrations/0009_message_usage.up.sql

@@ -0,0 +1,4 @@
+-- 0009_message_usage
+-- Add usage JSONB column to bot_history_messages for storing LLM token usage
+
+ALTER TABLE bot_history_messages ADD COLUMN IF NOT EXISTS usage JSONB;

db/queries/bots.sql

@@ -1,21 +1,21 @@
 -- name: CreateBot :one
 INSERT INTO bots (owner_user_id, type, display_name, avatar_url, is_active, metadata, status)
 VALUES ($1, $2, $3, $4, $5, $6, $7)
-RETURNING id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at;
+RETURNING id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, max_context_tokens, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at;
 
 -- name: GetBotByID :one
-SELECT id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at
+SELECT id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, max_context_tokens, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at
 FROM bots
 WHERE id = $1;
 
 -- name: ListBotsByOwner :many
-SELECT id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at
+SELECT id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, max_context_tokens, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at
 FROM bots
 WHERE owner_user_id = $1
 ORDER BY created_at DESC;
 
 -- name: ListBotsByMember :many
-SELECT b.id, b.owner_user_id, b.type, b.display_name, b.avatar_url, b.is_active, b.status, b.max_context_load_time, b.language, b.allow_guest, b.chat_model_id, b.memory_model_id, b.embedding_model_id, b.search_provider_id, b.metadata, b.created_at, b.updated_at
+SELECT b.id, b.owner_user_id, b.type, b.display_name, b.avatar_url, b.is_active, b.status, b.max_context_load_time, b.max_context_tokens, b.language, b.allow_guest, b.chat_model_id, b.memory_model_id, b.embedding_model_id, b.search_provider_id, b.metadata, b.created_at, b.updated_at
 FROM bots b
 JOIN bot_members m ON m.bot_id = b.id
 WHERE m.user_id = $1
@@ -29,14 +29,14 @@ SET display_name = $2,
     metadata = $5,
     updated_at = now()
 WHERE id = $1
-RETURNING id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at;
+RETURNING id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, max_context_tokens, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at;
 
 -- name: UpdateBotOwner :one
 UPDATE bots
 SET owner_user_id = $2,
     updated_at = now()
 WHERE id = $1
-RETURNING id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at;
+RETURNING id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, max_context_tokens, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at;
 
 -- name: UpdateBotStatus :exec
 UPDATE bots

db/queries/messages.sql

@@ -9,7 +9,8 @@ INSERT INTO bot_history_messages (
   source_reply_to_message_id,
   role,
   content,
-  metadata
+  metadata,
+  usage
 )
 VALUES (
   sqlc.arg(bot_id),
@@ -21,7 +22,8 @@ VALUES (
   sqlc.narg(source_reply_to_message_id)::text,
   sqlc.arg(role),
   sqlc.arg(content),
-  sqlc.arg(metadata)
+  sqlc.arg(metadata),
+  sqlc.arg(usage)
 )
 RETURNING
   id,
@@ -35,6 +37,7 @@ RETURNING
   role,
   content,
   metadata,
+  usage,
   created_at;
 
 -- name: ListMessages :many
@@ -50,6 +53,7 @@ SELECT
   m.role,
   m.content,
   m.metadata,
+  m.usage,
   m.created_at,
   ci.display_name AS sender_display_name,
   ci.avatar_url AS sender_avatar_url
@@ -72,6 +76,7 @@ SELECT
   m.role,
   m.content,
   m.metadata,
+  m.usage,
   m.created_at,
   ci.display_name AS sender_display_name,
   ci.avatar_url AS sender_avatar_url
@@ -94,6 +99,7 @@ SELECT
   m.role,
   m.content,
   m.metadata,
+  m.usage,
   m.created_at,
   ci.display_name AS sender_display_name,
   ci.avatar_url AS sender_avatar_url
@@ -117,6 +123,7 @@ SELECT
   m.role,
   m.content,
   m.metadata,
+  m.usage,
   m.created_at,
   ci.display_name AS sender_display_name,
   ci.avatar_url AS sender_avatar_url

db/queries/settings.sql

@@ -2,6 +2,7 @@
 SELECT
   bots.id AS bot_id,
   bots.max_context_load_time,
+  bots.max_context_tokens,
   bots.language,
   bots.allow_guest,
   chat_models.model_id AS chat_model_id,
@@ -19,6 +20,7 @@ WHERE bots.id = $1;
 WITH updated AS (
   UPDATE bots
   SET max_context_load_time = sqlc.arg(max_context_load_time),
+      max_context_tokens = sqlc.arg(max_context_tokens),
       language = sqlc.arg(language),
       allow_guest = sqlc.arg(allow_guest),
       chat_model_id = COALESCE(sqlc.narg(chat_model_id)::uuid, bots.chat_model_id),
@@ -27,11 +29,12 @@ WITH updated AS (
       search_provider_id = COALESCE(sqlc.narg(search_provider_id)::uuid, bots.search_provider_id),
       updated_at = now()
   WHERE bots.id = sqlc.arg(id)
-  RETURNING bots.id, bots.max_context_load_time, bots.language, bots.allow_guest, bots.chat_model_id, bots.memory_model_id, bots.embedding_model_id, bots.search_provider_id
+  RETURNING bots.id, bots.max_context_load_time, bots.max_context_tokens, bots.language, bots.allow_guest, bots.chat_model_id, bots.memory_model_id, bots.embedding_model_id, bots.search_provider_id
 )
 SELECT
   updated.id AS bot_id,
   updated.max_context_load_time,
+  updated.max_context_tokens,
   updated.language,
   updated.allow_guest,
   chat_models.model_id AS chat_model_id,
@@ -47,6 +50,7 @@ LEFT JOIN search_providers ON search_providers.id = updated.search_provider_id;
 -- name: DeleteSettingsByBotID :exec
 UPDATE bots
 SET max_context_load_time = 1440,
+    max_context_tokens = 0,
     language = 'auto',
     allow_guest = false,
     chat_model_id = NULL,

internal/conversation/flow/resolver.go

@@ -150,6 +150,12 @@ type gatewayRequest struct {
 type gatewayResponse struct {
 	Messages []conversation.ModelMessage `json:"messages"`
 	Skills   []string                    `json:"skills"`
+	Usage    json.RawMessage             `json:"usage,omitempty"`
+}
+
+type gatewayUsage struct {
+	InputTokens  *int `json:"inputTokens"`
+	OutputTokens *int `json:"outputTokens"`
 }
 
 // gatewaySchedule matches the agent gateway ScheduleModel for /chat/trigger-schedule.
@@ -212,13 +218,15 @@ func (r *Resolver) resolve(ctx context.Context, req conversation.ChatRequest) (r
 		return resolvedContext{}, err
 	}
 	maxCtx := coalescePositiveInt(req.MaxContextLoadTime, botSettings.MaxContextLoadTime, defaultMaxContextMinutes)
+	maxTokens := botSettings.MaxContextTokens
 
 	var messages []conversation.ModelMessage
 	if !skipHistory && r.conversationSvc != nil {
-		messages, err = r.loadMessages(ctx, req.ChatID, maxCtx)
+		loaded, loadErr := r.loadMessages(ctx, req.ChatID, maxCtx)
-		if err != nil {
+		if loadErr != nil {
-			return resolvedContext{}, err
+			return resolvedContext{}, loadErr
 		}
+		messages = trimMessagesByTokens(loaded, maxTokens)
 	}
 	if memoryMsg := r.loadMemoryContextMessage(ctx, req); memoryMsg != nil {
 		messages = append(messages, *memoryMsg)
@@ -291,7 +299,7 @@ func (r *Resolver) Chat(ctx context.Context, req conversation.ChatRequest) (conv
 	if err != nil {
 		return conversation.ChatResponse{}, err
 	}
-	if err := r.storeRound(ctx, req, resp.Messages); err != nil {
+	if err := r.storeRound(ctx, req, resp.Messages, resp.Usage); err != nil {
 		return conversation.ChatResponse{}, err
 	}
 	return conversation.ChatResponse{
@@ -345,7 +353,7 @@ func (r *Resolver) TriggerSchedule(ctx context.Context, botID string, payload sc
 	if err != nil {
 		return err
 	}
-	return r.storeRound(ctx, req, resp.Messages)
+	return r.storeRound(ctx, req, resp.Messages, resp.Usage)
 }
 
 // --- StreamChat ---
@@ -552,7 +560,7 @@ func (r *Resolver) tryStoreStream(ctx context.Context, req conversation.ChatRequ
 	if eventType == "done" {
 		var resp gatewayResponse
 		if err := json.Unmarshal([]byte(data), &resp); err == nil && len(resp.Messages) > 0 {
-			return true, r.storeRound(ctx, req, resp.Messages)
+			return true, r.storeRound(ctx, req, resp.Messages, resp.Usage)
 		}
 	}
 
@@ -562,15 +570,16 @@ func (r *Resolver) tryStoreStream(ctx context.Context, req conversation.ChatRequ
 		Data     json.RawMessage             `json:"data"`
 		Messages []conversation.ModelMessage `json:"messages"`
 		Skills   []string                    `json:"skills"`
+		Usage    json.RawMessage             `json:"usage,omitempty"`
 	}
 	if err := json.Unmarshal([]byte(data), &envelope); err == nil {
 		if (envelope.Type == "agent_end" || envelope.Type == "done") && len(envelope.Messages) > 0 {
-			return true, r.storeRound(ctx, req, envelope.Messages)
+			return true, r.storeRound(ctx, req, envelope.Messages, envelope.Usage)
 		}
 		if envelope.Type == "done" && len(envelope.Data) > 0 {
 			var resp gatewayResponse
 			if err := json.Unmarshal(envelope.Data, &resp); err == nil && len(resp.Messages) > 0 {
-				return true, r.storeRound(ctx, req, resp.Messages)
+				return true, r.storeRound(ctx, req, resp.Messages, resp.Usage)
 			}
 		}
 	}
@@ -578,7 +587,7 @@ func (r *Resolver) tryStoreStream(ctx context.Context, req conversation.ChatRequ
 	// fallback: data: {messages: [...]}
 	var resp gatewayResponse
 	if err := json.Unmarshal([]byte(data), &resp); err == nil && len(resp.Messages) > 0 {
-		return true, r.storeRound(ctx, req, resp.Messages)
+		return true, r.storeRound(ctx, req, resp.Messages, resp.Usage)
 	}
 	return false, nil
 }
@@ -648,7 +657,12 @@ func (r *Resolver) resolveContainerID(ctx context.Context, botID, explicit strin
 
 // --- message loading ---
 
-func (r *Resolver) loadMessages(ctx context.Context, chatID string, maxContextMinutes int) ([]conversation.ModelMessage, error) {
+type messageWithUsage struct {
+	Message          conversation.ModelMessage
+	UsageInputTokens *int
+}
+
+func (r *Resolver) loadMessages(ctx context.Context, chatID string, maxContextMinutes int) ([]messageWithUsage, error) {
 	if r.messageService == nil {
 		return nil, nil
 	}
@@ -657,7 +671,7 @@ func (r *Resolver) loadMessages(ctx context.Context, chatID string, maxContextMi
 	if err != nil {
 		return nil, err
 	}
-	var result []conversation.ModelMessage
+	var result []messageWithUsage
 	for _, m := range msgs {
 		var mm conversation.ModelMessage
 		if err := json.Unmarshal(m.Content, &mm); err != nil {
@@ -667,11 +681,84 @@ func (r *Resolver) loadMessages(ctx context.Context, chatID string, maxContextMi
 		} else {
 			mm.Role = m.Role
 		}
-		result = append(result, mm)
+		var inputTokens *int
+		if len(m.Usage) > 0 {
+			var u gatewayUsage
+			if json.Unmarshal(m.Usage, &u) == nil {
+				inputTokens = u.InputTokens
+			}
+		}
+		result = append(result, messageWithUsage{Message: mm, UsageInputTokens: inputTokens})
 	}
 	return result, nil
 }
 
+func estimateMessageTokens(msg conversation.ModelMessage) int {
+	text := msg.TextContent()
+	if len(text) == 0 {
+		data, _ := json.Marshal(msg.Content)
+		return len(data) / 4
+	}
+	return len(text) / 4
+}
+
+func trimMessagesByTokens(messages []messageWithUsage, maxTokens int) []conversation.ModelMessage {
+	if maxTokens <= 0 || len(messages) == 0 {
+		result := make([]conversation.ModelMessage, len(messages))
+		for i, m := range messages {
+			result[i] = m.Message
+		}
+		return result
+	}
+
+	// Scan backwards. When a message with UsageInputTokens is found, that value
+	// represents the cumulative input tokens for all messages up to and including
+	// that message. Messages after it are estimated with chars/4.
+	totalTokens := 0
+	anchorFound := false
+	cutoff := 0
+
+	tailEstimate := 0
+	for i := len(messages) - 1; i >= 0; i-- {
+		if !anchorFound && messages[i].UsageInputTokens != nil {
+			anchorFound = true
+			totalTokens = *messages[i].UsageInputTokens + tailEstimate
+			if totalTokens > maxTokens {
+				cutoff = i + 1
+				break
+			}
+			continue
+		}
+		est := estimateMessageTokens(messages[i].Message)
+		if anchorFound {
+			totalTokens += est
+			if totalTokens > maxTokens {
+				cutoff = i + 1
+				break
+			}
+		} else {
+			tailEstimate += est
+		}
+	}
+
+	if !anchorFound {
+		totalTokens = 0
+		for i := len(messages) - 1; i >= 0; i-- {
+			totalTokens += estimateMessageTokens(messages[i].Message)
+			if totalTokens > maxTokens {
+				cutoff = i + 1
+				break
+			}
+		}
+	}
+
+	result := make([]conversation.ModelMessage, 0, len(messages)-cutoff)
+	for _, m := range messages[cutoff:] {
+		result = append(result, m.Message)
+	}
+	return result
+}
+
 type memoryContextItem struct {
 	Namespace string
 	Item      memory.MemoryItem
@@ -792,9 +879,7 @@ func (r *Resolver) persistUserMessage(ctx context.Context, req conversation.Chat
 	return err
 }
 
-func (r *Resolver) storeRound(ctx context.Context, req conversation.ChatRequest, messages []conversation.ModelMessage) error {
+func (r *Resolver) storeRound(ctx context.Context, req conversation.ChatRequest, messages []conversation.ModelMessage, usage json.RawMessage) error {
-	// Add user query as the first message if not already present in the round.
-	// This ensures the user's prompt is persisted alongside the assistant's response.
 	fullRound := make([]conversation.ModelMessage, 0, len(messages)+1)
 	hasUserQuery := false
 	for _, m := range messages {
@@ -813,7 +898,6 @@ func (r *Resolver) storeRound(ctx context.Context, req conversation.ChatRequest,
 	}
 	for _, m := range messages {
 		if req.UserMessagePersisted && m.Role == "user" && strings.TrimSpace(m.TextContent()) == strings.TrimSpace(req.Query) {
-			// User message was already persisted before streaming; skip duplicate copy in round payload.
 			continue
 		}
 		fullRound = append(fullRound, m)
@@ -822,14 +906,12 @@ func (r *Resolver) storeRound(ctx context.Context, req conversation.ChatRequest,
 		return nil
 	}
 
-	r.storeMessages(ctx, req, fullRound)
+	r.storeMessages(ctx, req, fullRound, usage)
-	// Run memory extraction in the background so that the SSE stream can
-	// finish immediately after messages are persisted.
 	go r.storeMemory(context.WithoutCancel(ctx), req.BotID, fullRound)
 	return nil
 }
 
-func (r *Resolver) storeMessages(ctx context.Context, req conversation.ChatRequest, messages []conversation.ModelMessage) {
+func (r *Resolver) storeMessages(ctx context.Context, req conversation.ChatRequest, messages []conversation.ModelMessage, usage json.RawMessage) {
 	if r.messageService == nil {
 		return
 	}
@@ -838,7 +920,7 @@ func (r *Resolver) storeMessages(ctx context.Context, req conversation.ChatReque
 	}
 	meta := buildRouteMetadata(req)
 	senderChannelIdentityID, senderUserID := r.resolvePersistSenderIDs(ctx, req)
-	for _, msg := range messages {
+	for i, msg := range messages {
 		content, err := json.Marshal(msg)
 		if err != nil {
 			r.logger.Warn("storeMessages: marshal failed", slog.Any("error", err))
@@ -857,9 +939,12 @@ func (r *Resolver) storeMessages(ctx context.Context, req conversation.ChatReque
 				assets = chatAttachmentsToAssetRefs(req.Attachments)
 			}
 		} else if strings.TrimSpace(req.ExternalMessageID) != "" {
-			// Assistant/tool/system outputs are linked to the inbound source message for cross-channel reply threading.
 			sourceReplyToMessageID = req.ExternalMessageID
 		}
+		var msgUsage json.RawMessage
+		if i == len(messages)-1 && len(usage) > 0 {
+			msgUsage = usage
+		}
 		if _, err := r.messageService.Persist(ctx, messagepkg.PersistInput{
 			BotID:                   req.BotID,
 			RouteID:                 req.RouteID,
@@ -871,6 +956,7 @@ func (r *Resolver) storeMessages(ctx context.Context, req conversation.ChatReque
 			Role:                    msg.Role,
 			Content:                 content,
 			Metadata:                meta,
+			Usage:                   msgUsage,
 			Assets:                  assets,
 		}); err != nil {
 			r.logger.Warn("persist message failed", slog.Any("error", err))

internal/db/sqlc/bots.sql.go

@@ -14,7 +14,7 @@ import (
 const createBot = `-- name: CreateBot :one
 INSERT INTO bots (owner_user_id, type, display_name, avatar_url, is_active, metadata, status)
 VALUES ($1, $2, $3, $4, $5, $6, $7)
-RETURNING id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at
+RETURNING id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, max_context_tokens, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at
 `
 
 type CreateBotParams struct {
@@ -47,6 +47,7 @@ func (q *Queries) CreateBot(ctx context.Context, arg CreateBotParams) (Bot, erro
 		&i.IsActive,
 		&i.Status,
 		&i.MaxContextLoadTime,
+		&i.MaxContextTokens,
 		&i.Language,
 		&i.AllowGuest,
 		&i.ChatModelID,
@@ -84,7 +85,7 @@ func (q *Queries) DeleteBotMember(ctx context.Context, arg DeleteBotMemberParams
 }
 
 const getBotByID = `-- name: GetBotByID :one
-SELECT id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at
+SELECT id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, max_context_tokens, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at
 FROM bots
 WHERE id = $1
 `
@@ -101,6 +102,7 @@ func (q *Queries) GetBotByID(ctx context.Context, id pgtype.UUID) (Bot, error) {
 		&i.IsActive,
 		&i.Status,
 		&i.MaxContextLoadTime,
+		&i.MaxContextTokens,
 		&i.Language,
 		&i.AllowGuest,
 		&i.ChatModelID,
@@ -171,7 +173,7 @@ func (q *Queries) ListBotMembers(ctx context.Context, botID pgtype.UUID) ([]BotM
 }
 
 const listBotsByMember = `-- name: ListBotsByMember :many
-SELECT b.id, b.owner_user_id, b.type, b.display_name, b.avatar_url, b.is_active, b.status, b.max_context_load_time, b.language, b.allow_guest, b.chat_model_id, b.memory_model_id, b.embedding_model_id, b.search_provider_id, b.metadata, b.created_at, b.updated_at
+SELECT b.id, b.owner_user_id, b.type, b.display_name, b.avatar_url, b.is_active, b.status, b.max_context_load_time, b.max_context_tokens, b.language, b.allow_guest, b.chat_model_id, b.memory_model_id, b.embedding_model_id, b.search_provider_id, b.metadata, b.created_at, b.updated_at
 FROM bots b
 JOIN bot_members m ON m.bot_id = b.id
 WHERE m.user_id = $1
@@ -196,6 +198,7 @@ func (q *Queries) ListBotsByMember(ctx context.Context, userID pgtype.UUID) ([]B
 			&i.IsActive,
 			&i.Status,
 			&i.MaxContextLoadTime,
+			&i.MaxContextTokens,
 			&i.Language,
 			&i.AllowGuest,
 			&i.ChatModelID,
@@ -217,7 +220,7 @@ func (q *Queries) ListBotsByMember(ctx context.Context, userID pgtype.UUID) ([]B
 }
 
 const listBotsByOwner = `-- name: ListBotsByOwner :many
-SELECT id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at
+SELECT id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, max_context_tokens, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at
 FROM bots
 WHERE owner_user_id = $1
 ORDER BY created_at DESC
@@ -241,6 +244,7 @@ func (q *Queries) ListBotsByOwner(ctx context.Context, ownerUserID pgtype.UUID)
 			&i.IsActive,
 			&i.Status,
 			&i.MaxContextLoadTime,
+			&i.MaxContextTokens,
 			&i.Language,
 			&i.AllowGuest,
 			&i.ChatModelID,
@@ -266,7 +270,7 @@ UPDATE bots
 SET owner_user_id = $2,
     updated_at = now()
 WHERE id = $1
-RETURNING id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at
+RETURNING id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, max_context_tokens, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at
 `
 
 type UpdateBotOwnerParams struct {
@@ -286,6 +290,7 @@ func (q *Queries) UpdateBotOwner(ctx context.Context, arg UpdateBotOwnerParams)
 		&i.IsActive,
 		&i.Status,
 		&i.MaxContextLoadTime,
+		&i.MaxContextTokens,
 		&i.Language,
 		&i.AllowGuest,
 		&i.ChatModelID,
@@ -307,7 +312,7 @@ SET display_name = $2,
     metadata = $5,
     updated_at = now()
 WHERE id = $1
-RETURNING id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at
+RETURNING id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, max_context_tokens, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at
 `
 
 type UpdateBotProfileParams struct {
@@ -336,6 +341,7 @@ func (q *Queries) UpdateBotProfile(ctx context.Context, arg UpdateBotProfilePara
 		&i.IsActive,
 		&i.Status,
 		&i.MaxContextLoadTime,
+		&i.MaxContextTokens,
 		&i.Language,
 		&i.AllowGuest,
 		&i.ChatModelID,

internal/db/sqlc/conversations.sql.go

@@ -590,7 +590,7 @@ WITH updated AS (
   SET display_name = $1,
       updated_at = now()
   WHERE bots.id = $2
-  RETURNING id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at
+  RETURNING id, owner_user_id, type, display_name, avatar_url, is_active, status, max_context_load_time, max_context_tokens, language, allow_guest, chat_model_id, memory_model_id, embedding_model_id, search_provider_id, metadata, created_at, updated_at
 )
 SELECT
   updated.id AS id,

internal/db/sqlc/messages.sql.go

@@ -22,7 +22,8 @@ INSERT INTO bot_history_messages (
   source_reply_to_message_id,
   role,
   content,
-  metadata
+  metadata,
+  usage
 )
 VALUES (
   $1,
@@ -34,7 +35,8 @@ VALUES (
   $7::text,
   $8,
   $9,
-  $10
+  $10,
+  $11
 )
 RETURNING
   id,
@@ -48,6 +50,7 @@ RETURNING
   role,
   content,
   metadata,
+  usage,
   created_at
 `
 
@@ -62,6 +65,7 @@ type CreateMessageParams struct {
 	Role                    string      `json:"role"`
 	Content                 []byte      `json:"content"`
 	Metadata                []byte      `json:"metadata"`
+	Usage                   []byte      `json:"usage"`
 }
 
 type CreateMessageRow struct {
@@ -76,6 +80,7 @@ type CreateMessageRow struct {
 	Role                    string             `json:"role"`
 	Content                 []byte             `json:"content"`
 	Metadata                []byte             `json:"metadata"`
+	Usage                   []byte             `json:"usage"`
 	CreatedAt               pgtype.Timestamptz `json:"created_at"`
 }
 
@@ -91,6 +96,7 @@ func (q *Queries) CreateMessage(ctx context.Context, arg CreateMessageParams) (C
 		arg.Role,
 		arg.Content,
 		arg.Metadata,
+		arg.Usage,
 	)
 	var i CreateMessageRow
 	err := row.Scan(
@@ -105,6 +111,7 @@ func (q *Queries) CreateMessage(ctx context.Context, arg CreateMessageParams) (C
 		&i.Role,
 		&i.Content,
 		&i.Metadata,
+		&i.Usage,
 		&i.CreatedAt,
 	)
 	return i, err
@@ -133,6 +140,7 @@ SELECT
   m.role,
   m.content,
   m.metadata,
+  m.usage,
   m.created_at,
   ci.display_name AS sender_display_name,
   ci.avatar_url AS sender_avatar_url
@@ -155,6 +163,7 @@ type ListMessagesRow struct {
 	Role                    string             `json:"role"`
 	Content                 []byte             `json:"content"`
 	Metadata                []byte             `json:"metadata"`
+	Usage                   []byte             `json:"usage"`
 	CreatedAt               pgtype.Timestamptz `json:"created_at"`
 	SenderDisplayName       pgtype.Text        `json:"sender_display_name"`
 	SenderAvatarUrl         pgtype.Text        `json:"sender_avatar_url"`
@@ -181,6 +190,7 @@ func (q *Queries) ListMessages(ctx context.Context, botID pgtype.UUID) ([]ListMe
 			&i.Role,
 			&i.Content,
 			&i.Metadata,
+			&i.Usage,
 			&i.CreatedAt,
 			&i.SenderDisplayName,
 			&i.SenderAvatarUrl,
@@ -208,6 +218,7 @@ SELECT
   m.role,
   m.content,
   m.metadata,
+  m.usage,
   m.created_at,
   ci.display_name AS sender_display_name,
   ci.avatar_url AS sender_avatar_url
@@ -237,6 +248,7 @@ type ListMessagesBeforeRow struct {
 	Role                    string             `json:"role"`
 	Content                 []byte             `json:"content"`
 	Metadata                []byte             `json:"metadata"`
+	Usage                   []byte             `json:"usage"`
 	CreatedAt               pgtype.Timestamptz `json:"created_at"`
 	SenderDisplayName       pgtype.Text        `json:"sender_display_name"`
 	SenderAvatarUrl         pgtype.Text        `json:"sender_avatar_url"`
@@ -263,6 +275,7 @@ func (q *Queries) ListMessagesBefore(ctx context.Context, arg ListMessagesBefore
 			&i.Role,
 			&i.Content,
 			&i.Metadata,
+			&i.Usage,
 			&i.CreatedAt,
 			&i.SenderDisplayName,
 			&i.SenderAvatarUrl,
@@ -290,6 +303,7 @@ SELECT
   m.role,
   m.content,
   m.metadata,
+  m.usage,
   m.created_at,
   ci.display_name AS sender_display_name,
   ci.avatar_url AS sender_avatar_url
@@ -317,6 +331,7 @@ type ListMessagesLatestRow struct {
 	Role                    string             `json:"role"`
 	Content                 []byte             `json:"content"`
 	Metadata                []byte             `json:"metadata"`
+	Usage                   []byte             `json:"usage"`
 	CreatedAt               pgtype.Timestamptz `json:"created_at"`
 	SenderDisplayName       pgtype.Text        `json:"sender_display_name"`
 	SenderAvatarUrl         pgtype.Text        `json:"sender_avatar_url"`
@@ -343,6 +358,7 @@ func (q *Queries) ListMessagesLatest(ctx context.Context, arg ListMessagesLatest
 			&i.Role,
 			&i.Content,
 			&i.Metadata,
+			&i.Usage,
 			&i.CreatedAt,
 			&i.SenderDisplayName,
 			&i.SenderAvatarUrl,
@@ -370,6 +386,7 @@ SELECT
   m.role,
   m.content,
   m.metadata,
+  m.usage,
   m.created_at,
   ci.display_name AS sender_display_name,
   ci.avatar_url AS sender_avatar_url
@@ -397,6 +414,7 @@ type ListMessagesSinceRow struct {
 	Role                    string             `json:"role"`
 	Content                 []byte             `json:"content"`
 	Metadata                []byte             `json:"metadata"`
+	Usage                   []byte             `json:"usage"`
 	CreatedAt               pgtype.Timestamptz `json:"created_at"`
 	SenderDisplayName       pgtype.Text        `json:"sender_display_name"`
 	SenderAvatarUrl         pgtype.Text        `json:"sender_avatar_url"`
@@ -423,6 +441,7 @@ func (q *Queries) ListMessagesSince(ctx context.Context, arg ListMessagesSincePa
 			&i.Role,
 			&i.Content,
 			&i.Metadata,
+			&i.Usage,
 			&i.CreatedAt,
 			&i.SenderDisplayName,
 			&i.SenderAvatarUrl,

internal/db/sqlc/models.go

@@ -17,6 +17,7 @@ type Bot struct {
 	IsActive           bool               `json:"is_active"`
 	Status             string             `json:"status"`
 	MaxContextLoadTime int32              `json:"max_context_load_time"`
+	MaxContextTokens   int32              `json:"max_context_tokens"`
 	Language           string             `json:"language"`
 	AllowGuest         bool               `json:"allow_guest"`
 	ChatModelID        pgtype.UUID        `json:"chat_model_id"`
@@ -69,6 +70,7 @@ type BotHistoryMessage struct {
 	Role                    string             `json:"role"`
 	Content                 []byte             `json:"content"`
 	Metadata                []byte             `json:"metadata"`
+	Usage                   []byte             `json:"usage"`
 	CreatedAt               pgtype.Timestamptz `json:"created_at"`
 }
 

internal/db/sqlc/settings.sql.go

@@ -14,6 +14,7 @@ import (
 const deleteSettingsByBotID = `-- name: DeleteSettingsByBotID :exec
 UPDATE bots
 SET max_context_load_time = 1440,
+    max_context_tokens = 0,
     language = 'auto',
     allow_guest = false,
     chat_model_id = NULL,
@@ -33,6 +34,7 @@ const getSettingsByBotID = `-- name: GetSettingsByBotID :one
 SELECT
   bots.id AS bot_id,
   bots.max_context_load_time,
+  bots.max_context_tokens,
   bots.language,
   bots.allow_guest,
   chat_models.model_id AS chat_model_id,
@@ -50,6 +52,7 @@ WHERE bots.id = $1
 type GetSettingsByBotIDRow struct {
 	BotID              pgtype.UUID `json:"bot_id"`
 	MaxContextLoadTime int32       `json:"max_context_load_time"`
+	MaxContextTokens   int32       `json:"max_context_tokens"`
 	Language           string      `json:"language"`
 	AllowGuest         bool        `json:"allow_guest"`
 	ChatModelID        pgtype.Text `json:"chat_model_id"`
@@ -64,6 +67,7 @@ func (q *Queries) GetSettingsByBotID(ctx context.Context, id pgtype.UUID) (GetSe
 	err := row.Scan(
 		&i.BotID,
 		&i.MaxContextLoadTime,
+		&i.MaxContextTokens,
 		&i.Language,
 		&i.AllowGuest,
 		&i.ChatModelID,
@@ -78,19 +82,21 @@ const upsertBotSettings = `-- name: UpsertBotSettings :one
 WITH updated AS (
   UPDATE bots
   SET max_context_load_time = $1,
-      language = $2,
+      max_context_tokens = $2,
-      allow_guest = $3,
+      language = $3,
-      chat_model_id = COALESCE($4::uuid, bots.chat_model_id),
+      allow_guest = $4,
-      memory_model_id = COALESCE($5::uuid, bots.memory_model_id),
+      chat_model_id = COALESCE($5::uuid, bots.chat_model_id),
-      embedding_model_id = COALESCE($6::uuid, bots.embedding_model_id),
+      memory_model_id = COALESCE($6::uuid, bots.memory_model_id),
-      search_provider_id = COALESCE($7::uuid, bots.search_provider_id),
+      embedding_model_id = COALESCE($7::uuid, bots.embedding_model_id),
+      search_provider_id = COALESCE($8::uuid, bots.search_provider_id),
       updated_at = now()
-  WHERE bots.id = $8
+  WHERE bots.id = $9
-  RETURNING bots.id, bots.max_context_load_time, bots.language, bots.allow_guest, bots.chat_model_id, bots.memory_model_id, bots.embedding_model_id, bots.search_provider_id
+  RETURNING bots.id, bots.max_context_load_time, bots.max_context_tokens, bots.language, bots.allow_guest, bots.chat_model_id, bots.memory_model_id, bots.embedding_model_id, bots.search_provider_id
 )
 SELECT
   updated.id AS bot_id,
   updated.max_context_load_time,
+  updated.max_context_tokens,
   updated.language,
   updated.allow_guest,
   chat_models.model_id AS chat_model_id,
@@ -106,6 +112,7 @@ LEFT JOIN search_providers ON search_providers.id = updated.search_provider_id
 
 type UpsertBotSettingsParams struct {
 	MaxContextLoadTime int32       `json:"max_context_load_time"`
+	MaxContextTokens   int32       `json:"max_context_tokens"`
 	Language           string      `json:"language"`
 	AllowGuest         bool        `json:"allow_guest"`
 	ChatModelID        pgtype.UUID `json:"chat_model_id"`
@@ -118,6 +125,7 @@ type UpsertBotSettingsParams struct {
 type UpsertBotSettingsRow struct {
 	BotID              pgtype.UUID `json:"bot_id"`
 	MaxContextLoadTime int32       `json:"max_context_load_time"`
+	MaxContextTokens   int32       `json:"max_context_tokens"`
 	Language           string      `json:"language"`
 	AllowGuest         bool        `json:"allow_guest"`
 	ChatModelID        pgtype.Text `json:"chat_model_id"`
@@ -129,6 +137,7 @@ type UpsertBotSettingsRow struct {
 func (q *Queries) UpsertBotSettings(ctx context.Context, arg UpsertBotSettingsParams) (UpsertBotSettingsRow, error) {
 	row := q.db.QueryRow(ctx, upsertBotSettings,
 		arg.MaxContextLoadTime,
+		arg.MaxContextTokens,
 		arg.Language,
 		arg.AllowGuest,
 		arg.ChatModelID,
@@ -141,6 +150,7 @@ func (q *Queries) UpsertBotSettings(ctx context.Context, arg UpsertBotSettingsPa
 	err := row.Scan(
 		&i.BotID,
 		&i.MaxContextLoadTime,
+		&i.MaxContextTokens,
 		&i.Language,
 		&i.AllowGuest,
 		&i.ChatModelID,

internal/message/service.go

@@ -79,6 +79,7 @@ func (s *DBService) Persist(ctx context.Context, input PersistInput) (Message, e
 		Role:                    input.Role,
 		Content:                 content,
 		Metadata:                metaBytes,
+		Usage:                   input.Usage,
 	})
 	if err != nil {
 		return Message{}, err
@@ -213,6 +214,7 @@ func toMessageFromCreate(row sqlc.CreateMessageRow) Message {
 		row.Role,
 		row.Content,
 		row.Metadata,
+		row.Usage,
 		row.CreatedAt,
 	)
 }
@@ -232,6 +234,7 @@ func toMessageFromListRow(row sqlc.ListMessagesRow) Message {
 		row.Role,
 		row.Content,
 		row.Metadata,
+		row.Usage,
 		row.CreatedAt,
 	)
 }
@@ -251,6 +254,7 @@ func toMessageFromSinceRow(row sqlc.ListMessagesSinceRow) Message {
 		row.Role,
 		row.Content,
 		row.Metadata,
+		row.Usage,
 		row.CreatedAt,
 	)
 }
@@ -270,6 +274,7 @@ func toMessageFromLatestRow(row sqlc.ListMessagesLatestRow) Message {
 		row.Role,
 		row.Content,
 		row.Metadata,
+		row.Usage,
 		row.CreatedAt,
 	)
 }
@@ -288,6 +293,7 @@ func toMessageFields(
 	role string,
 	content []byte,
 	metadata []byte,
+	usage []byte,
 	createdAt pgtype.Timestamptz,
 ) Message {
 	return Message{
@@ -304,6 +310,7 @@ func toMessageFields(
 		Role:                    role,
 		Content:                 json.RawMessage(content),
 		Metadata:                parseJSONMap(metadata),
+		Usage:                   json.RawMessage(usage),
 		CreatedAt:               createdAt.Time,
 	}
 }
@@ -347,6 +354,7 @@ func toMessageFromBeforeRow(row sqlc.ListMessagesBeforeRow) Message {
 		row.Role,
 		row.Content,
 		row.Metadata,
+		row.Usage,
 		row.CreatedAt,
 	)
 }

internal/message/types.go

@@ -36,6 +36,7 @@ type Message struct {
 	Role                    string          `json:"role"`
 	Content                 json.RawMessage `json:"content"`
 	Metadata                map[string]any  `json:"metadata,omitempty"`
+	Usage                   json.RawMessage `json:"usage,omitempty"`
 	Assets                  []MessageAsset  `json:"assets,omitempty"`
 	CreatedAt               time.Time       `json:"created_at"`
 }
@@ -59,6 +60,7 @@ type PersistInput struct {
 	Role                    string
 	Content                 json.RawMessage
 	Metadata                map[string]any
+	Usage                   json.RawMessage
 	Assets                  []AssetRef
 }
 

internal/settings/service.go

@@ -54,10 +54,13 @@ func (s *Service) UpsertBot(ctx context.Context, botID string, req UpsertRequest
 	}
 	isPersonalBot := strings.EqualFold(strings.TrimSpace(botRow.Type), "personal")
 
-	current := normalizeBotSetting(botRow.MaxContextLoadTime, botRow.Language, botRow.AllowGuest)
+	current := normalizeBotSetting(botRow.MaxContextLoadTime, botRow.MaxContextTokens, botRow.Language, botRow.AllowGuest)
 	if req.MaxContextLoadTime != nil && *req.MaxContextLoadTime > 0 {
 		current.MaxContextLoadTime = *req.MaxContextLoadTime
 	}
+	if req.MaxContextTokens != nil && *req.MaxContextTokens >= 0 {
+		current.MaxContextTokens = *req.MaxContextTokens
+	}
 	if strings.TrimSpace(req.Language) != "" {
 		current.Language = strings.TrimSpace(req.Language)
 	}
@@ -106,6 +109,7 @@ func (s *Service) UpsertBot(ctx context.Context, botID string, req UpsertRequest
 	updated, err := s.queries.UpsertBotSettings(ctx, sqlc.UpsertBotSettingsParams{
 		ID:                 pgID,
 		MaxContextLoadTime: int32(current.MaxContextLoadTime),
+		MaxContextTokens:   int32(current.MaxContextTokens),
 		Language:           current.Language,
 		AllowGuest:         current.AllowGuest,
 		ChatModelID:        chatModelUUID,
@@ -130,15 +134,19 @@ func (s *Service) Delete(ctx context.Context, botID string) error {
 	return s.queries.DeleteSettingsByBotID(ctx, pgID)
 }
 
-func normalizeBotSetting(maxContextLoadTime int32, language string, allowGuest bool) Settings {
+func normalizeBotSetting(maxContextLoadTime int32, maxContextTokens int32, language string, allowGuest bool) Settings {
 	settings := Settings{
 		MaxContextLoadTime: int(maxContextLoadTime),
+		MaxContextTokens:   int(maxContextTokens),
 		Language:           strings.TrimSpace(language),
 		AllowGuest:         allowGuest,
 	}
 	if settings.MaxContextLoadTime <= 0 {
 		settings.MaxContextLoadTime = DefaultMaxContextLoadTime
 	}
+	if settings.MaxContextTokens < 0 {
+		settings.MaxContextTokens = 0
+	}
 	if settings.Language == "" {
 		settings.Language = DefaultLanguage
 	}
@@ -148,6 +156,7 @@ func normalizeBotSetting(maxContextLoadTime int32, language string, allowGuest b
 func normalizeBotSettingsReadRow(row sqlc.GetSettingsByBotIDRow) Settings {
 	return normalizeBotSettingsFields(
 		row.MaxContextLoadTime,
+		row.MaxContextTokens,
 		row.Language,
 		row.AllowGuest,
 		row.ChatModelID,
@@ -160,6 +169,7 @@ func normalizeBotSettingsReadRow(row sqlc.GetSettingsByBotIDRow) Settings {
 func normalizeBotSettingsWriteRow(row sqlc.UpsertBotSettingsRow) Settings {
 	return normalizeBotSettingsFields(
 		row.MaxContextLoadTime,
+		row.MaxContextTokens,
 		row.Language,
 		row.AllowGuest,
 		row.ChatModelID,
@@ -171,6 +181,7 @@ func normalizeBotSettingsWriteRow(row sqlc.UpsertBotSettingsRow) Settings {
 
 func normalizeBotSettingsFields(
 	maxContextLoadTime int32,
+	maxContextTokens int32,
 	language string,
 	allowGuest bool,
 	chatModelID pgtype.Text,
@@ -178,7 +189,7 @@ func normalizeBotSettingsFields(
 	embeddingModelID pgtype.Text,
 	searchProviderID pgtype.UUID,
 ) Settings {
-	settings := normalizeBotSetting(maxContextLoadTime, language, allowGuest)
+	settings := normalizeBotSetting(maxContextLoadTime, maxContextTokens, language, allowGuest)
 	settings.ChatModelID = strings.TrimSpace(chatModelID.String)
 	settings.MemoryModelID = strings.TrimSpace(memoryModelID.String)
 	settings.EmbeddingModelID = strings.TrimSpace(embeddingModelID.String)

internal/settings/types.go

@@ -11,6 +11,7 @@ type Settings struct {
 	EmbeddingModelID   string `json:"embedding_model_id"`
 	SearchProviderID   string `json:"search_provider_id"`
 	MaxContextLoadTime int    `json:"max_context_load_time"`
+	MaxContextTokens   int    `json:"max_context_tokens"`
 	Language           string `json:"language"`
 	AllowGuest         bool   `json:"allow_guest"`
 }
@@ -21,6 +22,7 @@ type UpsertRequest struct {
 	EmbeddingModelID   string `json:"embedding_model_id,omitempty"`
 	SearchProviderID   string `json:"search_provider_id,omitempty"`
 	MaxContextLoadTime *int   `json:"max_context_load_time,omitempty"`
+	MaxContextTokens   *int   `json:"max_context_tokens,omitempty"`
 	Language           string `json:"language,omitempty"`
 	AllowGuest         *bool  `json:"allow_guest,omitempty"`
 }

packages/sdk/src/types.gen.ts

@@ -673,6 +673,7 @@ export type MessageMessage = {
     sender_display_name?: string;
     sender_user_id?: string;
     source_reply_to_message_id?: string;
+    usage?: Array<number>;
 };
 
 export type MessageMessageAsset = {
@@ -866,6 +867,7 @@ export type SettingsSettings = {
     embedding_model_id?: string;
     language?: string;
     max_context_load_time?: number;
+    max_context_tokens?: number;
     memory_model_id?: string;
     search_provider_id?: string;
 };
@@ -876,6 +878,7 @@ export type SettingsUpsertRequest = {
     embedding_model_id?: string;
     language?: string;
     max_context_load_time?: number;
+    max_context_tokens?: number;
     memory_model_id?: string;
     search_provider_id?: string;
 };

packages/web/src/i18n/locales/en.json

@@ -354,6 +354,7 @@
       "searchProvider": "Search Provider",
       "searchProviderPlaceholder": "Select search provider",
       "maxContextLoadTime": "Max Context Load Time",
+      "maxContextTokens": "Max Context Tokens",
       "language": "Language",
       "allowGuest": "Allow Guest Access",
       "allowGuestPersonalHint": "Personal bots do not support guest access. Use a public bot instead.",

packages/web/src/i18n/locales/zh.json

@@ -350,6 +350,7 @@
       "searchProvider": "搜索提供方",
       "searchProviderPlaceholder": "选择搜索提供方",
       "maxContextLoadTime": "最大上下文加载时间",
+      "maxContextTokens": "最大上下文Token数",
       "language": "语言",
       "allowGuest": "允许游客访问",
       "allowGuestPersonalHint": "个人 Bot 不支持游客访问，请使用公开 Bot。",

packages/web/src/pages/bots/components/bot-settings.vue

@@ -58,6 +58,17 @@
       />
     </div>
 
+    <!-- Max Context Tokens -->
+    <div class="space-y-2">
+      <Label>{{ $t('bots.settings.maxContextTokens') }}</Label>
+      <Input
+        v-model.number="form.max_context_tokens"
+        type="number"
+        :min="0"
+        placeholder="0"
+      />
+    </div>
+
     <!-- Language -->
     <div class="space-y-2">
       <Label>{{ $t('bots.settings.language') }}</Label>
@@ -226,6 +237,7 @@ const form = reactive<SettingsSettings>({
   embedding_model_id: '',
   search_provider_id: '',
   max_context_load_time: 0,
+  max_context_tokens: 0,
   language: '',
   allow_guest: false,
 })
@@ -238,6 +250,7 @@ watch(settings, (val) => {
     form.embedding_model_id = val.embedding_model_id ?? ''
     form.search_provider_id = val.search_provider_id ?? ''
     form.max_context_load_time = val.max_context_load_time ?? 0
+    form.max_context_tokens = val.max_context_tokens ?? 0
     form.language = val.language ?? ''
     form.allow_guest = val.allow_guest ?? false
   }
@@ -252,6 +265,7 @@ const hasChanges = computed(() => {
     || form.embedding_model_id !== (s.embedding_model_id ?? '')
     || form.search_provider_id !== (s.search_provider_id ?? '')
     || form.max_context_load_time !== (s.max_context_load_time ?? 0)
+    || form.max_context_tokens !== (s.max_context_tokens ?? 0)
     || form.language !== (s.language ?? '')
   if (isPublicBot.value) {
     changed = changed || form.allow_guest !== (s.allow_guest ?? false)

spec/docs.go

@@ -6451,6 +6451,12 @@ const docTemplate = `{
                 },
                 "source_reply_to_message_id": {
                     "type": "string"
+                },
+                "usage": {
+                    "type": "array",
+                    "items": {
+                        "type": "integer"
+                    }
                 }
             }
         },
@@ -6949,6 +6955,9 @@ const docTemplate = `{
                 "max_context_load_time": {
                     "type": "integer"
                 },
+                "max_context_tokens": {
+                    "type": "integer"
+                },
                 "memory_model_id": {
                     "type": "string"
                 },
@@ -6975,6 +6984,9 @@ const docTemplate = `{
                 "max_context_load_time": {
                     "type": "integer"
                 },
+                "max_context_tokens": {
+                    "type": "integer"
+                },
                 "memory_model_id": {
                     "type": "string"
                 },

spec/swagger.json

@@ -6442,6 +6442,12 @@
                 },
                 "source_reply_to_message_id": {
                     "type": "string"
+                },
+                "usage": {
+                    "type": "array",
+                    "items": {
+                        "type": "integer"
+                    }
                 }
             }
         },
@@ -6940,6 +6946,9 @@
                 "max_context_load_time": {
                     "type": "integer"
                 },
+                "max_context_tokens": {
+                    "type": "integer"
+                },
                 "memory_model_id": {
                     "type": "string"
                 },
@@ -6966,6 +6975,9 @@
                 "max_context_load_time": {
                     "type": "integer"
                 },
+                "max_context_tokens": {
+                    "type": "integer"
+                },
                 "memory_model_id": {
                     "type": "string"
                 },

spec/swagger.yaml

@@ -1101,6 +1101,10 @@ definitions:
         type: string
       source_reply_to_message_id:
         type: string
+      usage:
+        items:
+          type: integer
+        type: array
     type: object
   message.MessageAsset:
     properties:
@@ -1435,6 +1439,8 @@ definitions:
         type: string
       max_context_load_time:
         type: integer
+      max_context_tokens:
+        type: integer
       memory_model_id:
         type: string
       search_provider_id:
@@ -1452,6 +1458,8 @@ definitions:
         type: string
       max_context_load_time:
         type: integer
+      max_context_tokens:
+        type: integer
       memory_model_id:
         type: string
       search_provider_id: