diff --git a/app/api/v1/admin_api/config.py b/app/api/v1/admin_api/config.py
index 6148e4a..1b9e0fe 100644
--- a/app/api/v1/admin_api/config.py
+++ b/app/api/v1/admin_api/config.py
@@ -4,6 +4,8 @@ from fastapi import APIRouter, Depends, HTTPException
from app.core.auth import verify_app_key
from app.core.config import config
+from app.services.grok.services.model import ModelService
+from app.services.token import get_token_manager
from app.core.storage import (
get_storage as get_storage_backend,
LocalStorage,
@@ -37,6 +39,27 @@ async def update_config(data: dict):
raise HTTPException(status_code=500, detail=str(e))
+@router.get("/model-routing/meta", dependencies=[Depends(verify_app_key)])
+async def get_model_routing_meta():
+ """获取模型池路由界面所需的模型与池元数据。"""
+ token_mgr = await get_token_manager()
+ pool_names = set(token_mgr.pools.keys())
+ pool_names.update({"ssoBasic", "ssoSuper"})
+
+ models = [
+ {
+ "id": item.model_id,
+ "display_name": item.display_name,
+ }
+ for item in ModelService.list()
+ ]
+
+ return {
+ "models": models,
+ "pools": sorted(pool_names),
+ }
+
+
@router.get("/storage", dependencies=[Depends(verify_app_key)])
async def admin_get_storage():
"""获取当前存储模式"""
diff --git a/app/services/grok/services/model.py b/app/services/grok/services/model.py
index 7ff0fe1..941d9e9 100644
--- a/app/services/grok/services/model.py
+++ b/app/services/grok/services/model.py
@@ -6,6 +6,7 @@ from enum import Enum
from typing import Optional, Tuple, List
from pydantic import BaseModel, Field
+from app.core.config import get_config
from app.core.exceptions import ValidationException
@@ -36,6 +37,7 @@ class ModelInfo(BaseModel):
is_image: bool = False
is_image_edit: bool = False
is_video: bool = False
+ public: bool = True
class ModelService:
@@ -164,16 +166,50 @@ class ModelService:
is_video=False,
),
ModelInfo(
- model_id="grok-4.20-beta",
+ model_id="grok-4.20-auto",
grok_model="grok-420",
- model_mode="MODEL_MODE_GROK_420",
+ model_mode="auto",
tier=Tier.BASIC,
cost=Cost.LOW,
- display_name="GROK-4.20-BETA",
+ display_name="GROK-4.20-AUTO",
is_image=False,
is_image_edit=False,
is_video=False,
),
+ ModelInfo(
+ model_id="grok-4.20-fast",
+ grok_model="grok-420",
+ model_mode="fast",
+ tier=Tier.BASIC,
+ cost=Cost.LOW,
+ display_name="GROK-4.20-FAST",
+ is_image=False,
+ is_image_edit=False,
+ is_video=False,
+ ),
+ ModelInfo(
+ model_id="grok-4.20-expert",
+ grok_model="grok-420",
+ model_mode="expert",
+ tier=Tier.BASIC,
+ cost=Cost.HIGH,
+ display_name="GROK-4.20-EXPERT",
+ is_image=False,
+ is_image_edit=False,
+ is_video=False,
+ ),
+ ModelInfo(
+ model_id="grok-4.20-beta",
+ grok_model="grok-420",
+ model_mode="auto",
+ tier=Tier.BASIC,
+ cost=Cost.LOW,
+ display_name="GROK-4.20-AUTO",
+ is_image=False,
+ is_image_edit=False,
+ is_video=False,
+ public=False,
+ ),
ModelInfo(
model_id="grok-imagine-1.0",
grok_model="grok-3",
@@ -214,6 +250,46 @@ class ModelService:
_map = {m.model_id: m for m in MODELS}
+ @classmethod
+ def _default_pool_candidates(cls, model_id: str) -> List[str]:
+ """返回模型默认池顺序。"""
+ model = cls.get(model_id)
+ if model and model.tier == Tier.SUPER:
+ return ["ssoSuper"]
+ # 基础模型优先使用 basic 池，缺失时可回退到 super 池
+ return ["ssoBasic", "ssoSuper"]
+
+ @classmethod
+ def _routing_lookup_keys(cls, model_id: str) -> List[str]:
+ """返回模型池路由查找顺序，兼容历史别名。"""
+ keys = [model_id]
+ if model_id == "grok-4.20-beta":
+ keys.append("grok-4.20-auto")
+ return keys
+
+ @classmethod
+ def _configured_pool_candidates(cls, model_id: str) -> Optional[List[str]]:
+ """读取管理配置中的模型池路由。"""
+ routing = get_config("model_routing.model_pools", {})
+ if not isinstance(routing, dict):
+ return None
+
+ for key in cls._routing_lookup_keys(model_id):
+ value = routing.get(key)
+ if isinstance(value, str):
+ pool = value.strip()
+ if pool:
+ return [pool]
+ elif isinstance(value, list):
+ pools = [
+ str(item).strip()
+ for item in value
+ if str(item).strip()
+ ]
+ if pools:
+ return pools
+ return None
+
@classmethod
def get(cls, model_id: str) -> Optional[ModelInfo]:
"""获取模型信息"""
@@ -222,7 +298,7 @@ class ModelService:
@classmethod
def list(cls) -> list[ModelInfo]:
"""获取所有模型"""
- return list(cls._map.values())
+ return [model for model in cls._map.values() if model.public]
@classmethod
def valid(cls, model_id: str) -> bool:
@@ -240,19 +316,16 @@ class ModelService:
@classmethod
def pool_for_model(cls, model_id: str) -> str:
"""根据模型选择 Token 池"""
- model = cls.get(model_id)
- if model and model.tier == Tier.SUPER:
- return "ssoSuper"
- return "ssoBasic"
+ pools = cls.pool_candidates_for_model(model_id)
+ return pools[0] if pools else "ssoBasic"
@classmethod
def pool_candidates_for_model(cls, model_id: str) -> List[str]:
"""按优先级返回可用 Token 池列表"""
- model = cls.get(model_id)
- if model and model.tier == Tier.SUPER:
- return ["ssoSuper"]
- # 基础模型优先使用 basic 池，缺失时可回退到 super 池
- return ["ssoBasic", "ssoSuper"]
+ configured = cls._configured_pool_candidates(model_id)
+ if configured:
+ return configured
+ return cls._default_pool_candidates(model_id)
__all__ = ["ModelService"]
diff --git a/app/services/reverse/app_chat.py b/app/services/reverse/app_chat.py
index b255c3a..6215036 100644
--- a/app/services/reverse/app_chat.py
+++ b/app/services/reverse/app_chat.py
@@ -91,7 +91,6 @@ class AppChatReverse:
"isAsyncChat": False,
"isReasoning": False,
"message": message,
- "modelMode": mode,
"modelName": model,
"responseMetadata": {
"requestModelDetails": {"modelId": model},
@@ -108,6 +107,10 @@ class AppChatReverse:
if model == "grok-420":
payload["enable420"] = True
+ if mode:
+ payload["modeId"] = mode
+ elif mode:
+ payload["modelMode"] = mode
custom_personality = AppChatReverse._resolve_custom_personality()
if custom_personality is not None and "Greet the user" not in message[-1000:]:
diff --git a/app/static/admin/css/config.css b/app/static/admin/css/config.css
index ca65dd2..85b61ab 100644
--- a/app/static/admin/css/config.css
+++ b/app/static/admin/css/config.css
@@ -58,6 +58,169 @@
transform: translateY(-50%);
}
+.routing-board {
+ display: grid;
+ grid-template-columns: minmax(0, 1.1fr) minmax(0, 1fr);
+ gap: 16px;
+}
+
+.routing-pane {
+ border: 1px solid var(--border);
+ border-radius: 14px;
+ padding: 14px;
+ background:
+ linear-gradient(180deg, rgba(255, 255, 255, 0.98) 0%, rgba(249, 250, 251, 0.96) 100%);
+}
+
+.routing-pane-header {
+ margin-bottom: 12px;
+}
+
+.routing-pane-title {
+ font-size: 13px;
+ font-weight: 700;
+ color: var(--accents-7);
+}
+
+.routing-pane-subtitle {
+ margin-top: 4px;
+ font-size: 12px;
+ line-height: 1.5;
+ color: var(--accents-4);
+}
+
+.routing-dropzone {
+ min-height: 148px;
+ border: 1px dashed var(--border);
+ border-radius: 12px;
+ padding: 12px;
+ display: flex;
+ flex-wrap: wrap;
+ align-content: flex-start;
+ gap: 10px;
+ background:
+ radial-gradient(circle at top left, rgba(0, 0, 0, 0.03), transparent 42%),
+ linear-gradient(180deg, rgba(255, 255, 255, 0.96), rgba(244, 244, 245, 0.96));
+ transition: border-color 0.2s ease, background-color 0.2s ease, transform 0.2s ease;
+}
+
+.routing-dropzone.is-over {
+ border-color: #111;
+ background:
+ radial-gradient(circle at top left, rgba(0, 0, 0, 0.06), transparent 42%),
+ linear-gradient(180deg, rgba(255, 255, 255, 0.98), rgba(237, 237, 239, 0.98));
+ transform: translateY(-1px);
+}
+
+.routing-dropzone-source {
+ min-height: 320px;
+}
+
+.routing-pool-grid {
+ display: grid;
+ gap: 12px;
+}
+
+.routing-pool-card {
+ border: 1px solid var(--border);
+ border-radius: 12px;
+ background: rgba(255, 255, 255, 0.86);
+ padding: 10px;
+}
+
+.routing-pool-header {
+ display: flex;
+ align-items: baseline;
+ justify-content: space-between;
+ gap: 10px;
+ margin-bottom: 10px;
+}
+
+.routing-pool-title {
+ font-size: 13px;
+ font-weight: 700;
+ color: var(--accents-7);
+}
+
+.routing-pool-hint {
+ font-size: 11px;
+ color: var(--accents-4);
+}
+
+.routing-chip {
+ display: inline-flex;
+ flex-direction: column;
+ align-items: flex-start;
+ gap: 4px;
+ max-width: 100%;
+ padding: 10px 12px;
+ border: 1px solid rgba(17, 17, 17, 0.12);
+ border-radius: 12px;
+ background: linear-gradient(180deg, #111 0%, #262626 100%);
+ color: #fff;
+ cursor: grab;
+ box-shadow: 0 6px 18px rgba(17, 17, 17, 0.1);
+ transition: transform 0.15s ease, box-shadow 0.15s ease, opacity 0.15s ease;
+}
+
+.routing-chip:hover {
+ transform: translateY(-1px);
+ box-shadow: 0 10px 24px rgba(17, 17, 17, 0.14);
+}
+
+.routing-chip.is-dragging {
+ opacity: 0.5;
+ cursor: grabbing;
+}
+
+.routing-chip-name {
+ font-size: 12px;
+ font-weight: 700;
+ line-height: 1.3;
+}
+
+.routing-chip-id {
+ max-width: 100%;
+ font-size: 11px;
+ color: rgba(255, 255, 255, 0.72);
+ white-space: nowrap;
+ overflow: hidden;
+ text-overflow: ellipsis;
+}
+
+.routing-empty {
+ width: 100%;
+ min-height: 72px;
+ border-radius: 10px;
+ display: flex;
+ align-items: center;
+ justify-content: center;
+ text-align: center;
+ padding: 12px;
+ font-size: 12px;
+ color: var(--accents-4);
+ background: rgba(245, 245, 245, 0.9);
+}
+
+html[data-theme='dark'] .routing-pane,
+html[data-theme='dark'] .routing-pool-card {
+ background: linear-gradient(180deg, rgba(29, 29, 31, 0.96), rgba(20, 20, 22, 0.96));
+}
+
+html[data-theme='dark'] .routing-dropzone {
+ background:
+ radial-gradient(circle at top left, rgba(255, 255, 255, 0.04), transparent 42%),
+ linear-gradient(180deg, rgba(24, 24, 27, 0.98), rgba(16, 16, 18, 0.98));
+}
+
+html[data-theme='dark'] .routing-dropzone.is-over {
+ border-color: rgba(255, 255, 255, 0.76);
+}
+
+html[data-theme='dark'] .routing-empty {
+ background: rgba(255, 255, 255, 0.03);
+}
+
@media (max-width: 900px) {
.config-section {
padding: 14px;
@@ -75,4 +238,12 @@
bottom: 0;
transform: none;
}
+
+ .routing-board {
+ grid-template-columns: 1fr;
+ }
+
+ .routing-dropzone-source {
+ min-height: 220px;
+ }
}
diff --git a/app/static/admin/js/config.js b/app/static/admin/js/config.js
index c0ebe0d..c3d69d8 100644
--- a/app/static/admin/js/config.js
+++ b/app/static/admin/js/config.js
@@ -1,5 +1,8 @@
let apiKey = '';
let currentConfig = {};
+let modelRoutingMeta = { models: [], pools: ['ssoBasic', 'ssoSuper'] };
+let modelRoutingAssignments = {};
+let modelRoutingDragId = '';
const byId = (id) => document.getElementById(id);
const NUMERIC_FIELDS = new Set([
'timeout',
@@ -139,6 +142,12 @@ const LOCALE_MAP = {
},
+ "model_routing": {
+ "label": "模型池路由",
+ "model_pools": { title: "模型到 Token 池映射", desc: "为每个模型手动指定使用哪个 Token 池。值支持字符串或字符串数组，例如 {\"grok-4.20-expert\":\"ssoSuper\",\"grok-4.20-auto\":[\"ssoBasic\",\"ssoSuper\"]}。" }
+ },
+
+
"cache": {
"label": "缓存管理",
"enable_auto_clean": { title: "自动清理", desc: "是否启用缓存自动清理，开启后按上限自动回收。" },
@@ -164,7 +173,8 @@ const LOCALE_MAP = {
// 配置部分说明（可选）
const SECTION_DESCRIPTIONS = {
- "proxy": "配置不正确将导致 403 错误。服务首次请求 Grok 时的 IP 必须与获取 CF Clearance 时的 IP 一致，后续服务器请求 IP 变化不会导致 403。"
+ "proxy": "配置不正确将导致 403 错误。服务首次请求 Grok 时的 IP 必须与获取 CF Clearance 时的 IP 一致，后续服务器请求 IP 变化不会导致 403。",
+ "model_routing": "这里可以手动指定每个模型优先走哪个 Token 池。未配置的模型仍会按系统默认路由。"
};
// CF 自动刷新联动禁用字段（全部在 proxy section 内）
@@ -312,9 +322,28 @@ function randomKey(len) {
async function init() {
apiKey = await ensureAdminKey();
if (apiKey === null) return;
+ await loadModelRoutingMeta();
loadData();
}
+async function loadModelRoutingMeta() {
+ try {
+ const res = await fetch('/v1/admin/model-routing/meta', {
+ headers: buildAuthHeaders(apiKey)
+ });
+ if (!res.ok) return;
+ const data = await res.json();
+ if (Array.isArray(data.models)) {
+ modelRoutingMeta.models = data.models;
+ }
+ if (Array.isArray(data.pools) && data.pools.length) {
+ modelRoutingMeta.pools = data.pools;
+ }
+ } catch (e) {
+ console.warn('模型池路由元数据加载失败', e);
+ }
+}
+
async function loadData() {
try {
const res = await fetch('/v1/admin/config', {
@@ -386,6 +415,182 @@ function renderConfig(data) {
applyCfRefreshState(cfEnabled);
}
+function normalizeModelRoutingAssignments(raw) {
+ const next = {};
+ const source = raw && typeof raw === 'object' ? raw : {};
+ Object.entries(source).forEach(([modelId, value]) => {
+ if (typeof value === 'string' && value.trim()) {
+ next[modelId] = value.trim();
+ return;
+ }
+ if (Array.isArray(value)) {
+ const first = value.map(item => String(item).trim()).find(Boolean);
+ if (first) next[modelId] = first;
+ }
+ });
+ return next;
+}
+
+function buildModelRoutingInput(section, key, val) {
+ modelRoutingAssignments = normalizeModelRoutingAssignments(val);
+
+ const wrapper = document.createElement('div');
+ wrapper.className = 'routing-board';
+ wrapper.dataset.type = 'model-routing';
+ wrapper.dataset.section = section;
+ wrapper.dataset.key = key;
+
+ const leftPane = document.createElement('div');
+ leftPane.className = 'routing-pane routing-pane-models';
+ leftPane.innerHTML = `
+
+ `;
+
+ const unassignedZone = document.createElement('div');
+ unassignedZone.className = 'routing-dropzone routing-dropzone-source';
+ unassignedZone.dataset.pool = '';
+ registerRoutingDropzone(unassignedZone, '');
+
+ const rightPane = document.createElement('div');
+ rightPane.className = 'routing-pane routing-pane-pools';
+ rightPane.innerHTML = `
+
+ `;
+
+ const poolGrid = document.createElement('div');
+ poolGrid.className = 'routing-pool-grid';
+ const pools = Array.isArray(modelRoutingMeta.pools) && modelRoutingMeta.pools.length
+ ? modelRoutingMeta.pools
+ : ['ssoBasic', 'ssoSuper'];
+
+ pools.forEach(poolName => {
+ const card = document.createElement('div');
+ card.className = 'routing-pool-card';
+
+ const header = document.createElement('div');
+ header.className = 'routing-pool-header';
+ header.innerHTML = `
+ ${poolName}
+ 拖拽模型到这里
+ `;
+
+ const zone = document.createElement('div');
+ zone.className = 'routing-dropzone routing-dropzone-pool';
+ zone.dataset.pool = poolName;
+ registerRoutingDropzone(zone, poolName);
+
+ card.appendChild(header);
+ card.appendChild(zone);
+ poolGrid.appendChild(card);
+ });
+
+ leftPane.appendChild(unassignedZone);
+ rightPane.appendChild(poolGrid);
+ wrapper.appendChild(leftPane);
+ wrapper.appendChild(rightPane);
+
+ renderModelRoutingBoard(wrapper);
+ return { input: wrapper, node: wrapper };
+}
+
+function registerRoutingDropzone(zone, poolName) {
+ zone.addEventListener('dragover', (event) => {
+ event.preventDefault();
+ zone.classList.add('is-over');
+ });
+ zone.addEventListener('dragleave', () => {
+ zone.classList.remove('is-over');
+ });
+ zone.addEventListener('drop', (event) => {
+ event.preventDefault();
+ zone.classList.remove('is-over');
+ const modelId = event.dataTransfer.getData('text/plain') || modelRoutingDragId;
+ if (!modelId) return;
+ if (poolName) {
+ modelRoutingAssignments[modelId] = poolName;
+ } else {
+ delete modelRoutingAssignments[modelId];
+ }
+ const board = zone.closest('.routing-board');
+ if (board) renderModelRoutingBoard(board);
+ modelRoutingDragId = '';
+ });
+}
+
+function createModelChip(model) {
+ const chip = document.createElement('button');
+ chip.type = 'button';
+ chip.className = 'routing-chip';
+ chip.draggable = true;
+ chip.dataset.modelId = model.id;
+ chip.innerHTML = `
+ ${model.display_name || model.id}
+ ${model.id}
+ `;
+
+ chip.addEventListener('dragstart', (event) => {
+ modelRoutingDragId = model.id;
+ event.dataTransfer.setData('text/plain', model.id);
+ event.dataTransfer.effectAllowed = 'move';
+ chip.classList.add('is-dragging');
+ });
+ chip.addEventListener('dragend', () => {
+ chip.classList.remove('is-dragging');
+ modelRoutingDragId = '';
+ document.querySelectorAll('.routing-dropzone.is-over').forEach((node) => {
+ node.classList.remove('is-over');
+ });
+ });
+
+ return chip;
+}
+
+function renderModelRoutingBoard(board) {
+ const models = Array.isArray(modelRoutingMeta.models) ? [...modelRoutingMeta.models] : [];
+ models.sort((a, b) => String(a.id).localeCompare(String(b.id)));
+
+ const sourceZone = board.querySelector('.routing-dropzone-source');
+ const poolZones = Array.from(board.querySelectorAll('.routing-dropzone-pool'));
+ if (!sourceZone) return;
+
+ sourceZone.replaceChildren();
+ poolZones.forEach(zone => zone.replaceChildren());
+
+ const assignedModelIds = new Set();
+ poolZones.forEach(zone => {
+ const poolName = zone.dataset.pool || '';
+ const assigned = models.filter(model => modelRoutingAssignments[model.id] === poolName);
+ assigned.forEach(model => {
+ assignedModelIds.add(model.id);
+ zone.appendChild(createModelChip(model));
+ });
+ if (!assigned.length) {
+ zone.appendChild(createRoutingEmptyState('拖拽模型到这里'));
+ }
+ });
+
+ const unassigned = models.filter(model => !assignedModelIds.has(model.id));
+ unassigned.forEach(model => {
+ sourceZone.appendChild(createModelChip(model));
+ });
+ if (!unassigned.length) {
+ sourceZone.appendChild(createRoutingEmptyState('所有模型都已分配到池'));
+ }
+}
+
+function createRoutingEmptyState(text) {
+ const empty = document.createElement('div');
+ empty.className = 'routing-empty';
+ empty.textContent = text;
+ return empty;
+}
+
function applyCfRefreshState(enabled) {
// 设置字段禁用状态的辅助函数
function setFieldDisabled(section, key, disabled) {
@@ -468,7 +673,11 @@ function buildFieldCard(section, key, val) {
]);
}
else if (Array.isArray(val) || typeof val === 'object') {
- built = buildJsonInput(section, key, val);
+ if (section === 'model_routing' && key === 'model_pools') {
+ built = buildModelRoutingInput(section, key, val);
+ } else {
+ built = buildJsonInput(section, key, val);
+ }
}
else {
if (key === 'api_key' || key === 'app_key' || key === 'public_key') {
@@ -522,11 +731,16 @@ async function saveConfig() {
const newConfig = typeof structuredClone === 'function'
? structuredClone(currentConfig)
: JSON.parse(JSON.stringify(currentConfig));
- const inputs = document.querySelectorAll('input[data-section], textarea[data-section], select[data-section]');
+ const inputs = document.querySelectorAll('input[data-section], textarea[data-section], select[data-section], [data-type="model-routing"][data-section]');
inputs.forEach(input => {
const s = input.dataset.section;
const k = input.dataset.key;
+ if (input.dataset.type === 'model-routing') {
+ if (!newConfig[s]) newConfig[s] = {};
+ newConfig[s][k] = { ...modelRoutingAssignments };
+ return;
+ }
let val = input.value;
if (input.type === 'checkbox') {
diff --git a/app/static/public/js/chat.js b/app/static/public/js/chat.js
index 1c75ee8..c3fdd22 100644
--- a/app/static/public/js/chat.js
+++ b/app/static/public/js/chat.js
@@ -1405,8 +1405,8 @@
async function loadModels() {
if (!modelSelect) return;
- const fallback = ['grok-4.1-fast', 'grok-4', 'grok-3', 'grok-3-mini', 'grok-3-thinking', 'grok-4.20-beta'];
- const preferred = 'grok-4.20-beta';
+ const fallback = ['grok-4.1-fast', 'grok-4', 'grok-3', 'grok-3-mini', 'grok-3-thinking', 'grok-4.20-fast', 'grok-4.20-expert', 'grok-4.20-auto'];
+ const preferred = 'grok-4.20-auto';
let list = fallback;
try {
diff --git a/config.defaults.toml b/config.defaults.toml
index 1426e44..dbe9d38 100644
--- a/config.defaults.toml
+++ b/config.defaults.toml
@@ -81,6 +81,13 @@ save_delay_ms = 500
# 多 worker 状态同步间隔（秒）
reload_interval_sec = 30
+# ==================== 模型路由配置 ====================
+[model_routing]
+# 模型到 Token 池的手动路由配置。
+# 值支持字符串（单池）或字符串数组（按顺序回退）。
+# 例如：{ "grok-4.20-expert" = "ssoSuper", "grok-4.20-auto" = ["ssoBasic", "ssoSuper"] }
+model_pools = {}
+
# ==================== 缓存管理 ====================
[cache]
# 是否启用自动清理