From ff3da6a5ee1d6f34215a345dfdc335cd2028afc4 Mon Sep 17 00:00:00 2001
From: Mord0reK <135718526+Mord0reK@users.noreply.github.com>
Date: Mon, 27 Jul 2026 17:29:22 +0200
Subject: [PATCH] feat: add configurable subagent model
---
backend/open_webui/config.py | 2 +
backend/open_webui/routers/configs.py | 2 +
backend/open_webui/utils/subagents.py | 43 ++++++++++---
.../admin/Settings/Subagents.svelte | 61 ++++++++++++++++++-
4 files changed, 98 insertions(+), 10 deletions(-)
diff --git a/backend/open_webui/config.py b/backend/open_webui/config.py
index 262e7c0818..eacfe43e16 100644
--- a/backend/open_webui/config.py
+++ b/backend/open_webui/config.py
@@ -2028,6 +2028,7 @@ ENABLE_CALENDAR = os.getenv('ENABLE_CALENDAR', 'True').lower() == 'true'
ENABLE_AUTOMATIONS = os.getenv('ENABLE_AUTOMATIONS', 'True').lower() == 'true'
ENABLE_SUBAGENTS = os.getenv('ENABLE_SUBAGENTS', 'False').lower() == 'true'
+SUBAGENTS_MODEL_ID = os.getenv('SUBAGENTS_MODEL_ID', '')
SUBAGENTS_BACKGROUND_ENABLED = os.getenv('SUBAGENTS_BACKGROUND_ENABLED', 'False').lower() == 'true'
SUBAGENTS_MAX_CONCURRENT = int(os.getenv('SUBAGENTS_MAX_CONCURRENT', '20'))
SUBAGENTS_MAX_ASYNC = int(os.getenv('SUBAGENTS_MAX_ASYNC', '20'))
@@ -3092,6 +3093,7 @@ DEFAULT_CONFIG = {
'calendar.enable': ENABLE_CALENDAR,
'automations.enable': ENABLE_AUTOMATIONS,
'subagents.enable': ENABLE_SUBAGENTS,
+ 'subagents.model_id': SUBAGENTS_MODEL_ID,
'subagents.background_enabled': SUBAGENTS_BACKGROUND_ENABLED,
'subagents.max_concurrent': SUBAGENTS_MAX_CONCURRENT,
'subagents.max_async': SUBAGENTS_MAX_ASYNC,
diff --git a/backend/open_webui/routers/configs.py b/backend/open_webui/routers/configs.py
index f591908a6d..4028dcc5b0 100644
--- a/backend/open_webui/routers/configs.py
+++ b/backend/open_webui/routers/configs.py
@@ -69,6 +69,7 @@ MODELS_CONFIG_KEYS = {
}
SUBAGENTS_CONFIG_KEYS = {
'ENABLE_SUBAGENTS': 'subagents.enable',
+ 'SUBAGENTS_MODEL_ID': 'subagents.model_id',
'SUBAGENTS_BACKGROUND_ENABLED': 'subagents.background_enabled',
'SUBAGENTS_MAX_CONCURRENT': 'subagents.max_concurrent',
'SUBAGENTS_MAX_ASYNC': 'subagents.max_async',
@@ -766,6 +767,7 @@ async def set_models_config(request: Request, form_data: ModelsConfigForm, user=
class SubagentsConfigForm(BaseModel):
ENABLE_SUBAGENTS: bool
+ SUBAGENTS_MODEL_ID: str = ''
SUBAGENTS_BACKGROUND_ENABLED: bool
SUBAGENTS_MAX_CONCURRENT: int
SUBAGENTS_MAX_ASYNC: int
diff --git a/backend/open_webui/utils/subagents.py b/backend/open_webui/utils/subagents.py
index 7092df44c2..3a53100188 100644
--- a/backend/open_webui/utils/subagents.py
+++ b/backend/open_webui/utils/subagents.py
@@ -44,6 +44,13 @@ _foreground_semaphore: asyncio.Semaphore | None = None
_parent_locks: dict[str, asyncio.Lock] = {}
+def get_subagent_model_ids(
+ parent_model_id: str | None, configured_model_id: str | None
+) -> tuple[str | None, str | None]:
+ subagent_model_id = configured_model_id or parent_model_id
+ return parent_model_id, subagent_model_id
+
+
def _build_request(source: Request, user_id: str, *, internal: bool) -> Request:
scope = {
'type': 'http',
@@ -123,7 +130,17 @@ async def process_pending_internal_messages(
if kind == 'timer' and first_meta.get('timer_id'):
timer = await Chats.get_chat_by_id(first_meta['timer_id'])
run = {**run, **(((timer.meta or {}).get('run') if timer else None) or {})}
- model_id = first.get('model') or run['model_id']
+ parent_model_id = first_meta.get('parent_model_id') or run.get('parent_model_id')
+ subagent_model_id = (
+ first.get('model')
+ or first_meta.get('subagent_model_id')
+ or run.get('subagent_model_id')
+ or run.get('model_id')
+ )
+ if kind == 'subagent':
+ model_id = parent_model_id or subagent_model_id
+ else:
+ model_id = subagent_model_id
if kind == 'timer':
batch = [first]
else:
@@ -132,7 +149,7 @@ async def process_pending_internal_messages(
for message in pending
for meta in [message.get('meta') or {}]
if message.get('parentId') == parent_id
- and (message.get('model') or model_id) == model_id
+ and (message.get('model') or subagent_model_id) == subagent_model_id
and (
meta.get('internal') is True
and meta.get('type') == 'subagent'
@@ -294,6 +311,7 @@ async def delegate(
'subagents.max_iterations',
'subagents.max_output',
'subagents.system_prompt',
+ 'subagents.model_id',
)
max_concurrent = int(config.get('subagents.max_concurrent') or 20)
max_async = int(config.get('subagents.max_async') or 20)
@@ -314,8 +332,13 @@ async def delegate(
and await Config.get('code_interpreter.engine', 'pyodide') != 'jupyter'
):
features.pop('code_interpreter')
+ parent_model_id, subagent_model_id = get_subagent_model_ids(
+ metadata.get('model_id') or (metadata.get('model') or {}).get('id'),
+ config.get('subagents.model_id'),
+ )
run = {
- 'model_id': metadata.get('model_id') or (metadata.get('model') or {}).get('id'),
+ 'parent_model_id': parent_model_id,
+ 'subagent_model_id': subagent_model_id,
'session_id': metadata.get('session_id'),
'tool_ids': copy.deepcopy(metadata.get('tool_ids') or []),
'skill_ids': copy.deepcopy(metadata.get('skill_ids') or []),
@@ -328,7 +351,7 @@ async def delegate(
'variables': copy.deepcopy(metadata.get('variables') or {}),
'direct': bool(metadata.get('direct')),
}
- if not run.get('model_id'):
+ if not run.get('subagent_model_id'):
return 'Error: model context is required.'
if run.get('direct'):
return 'Error: sub-agents are unavailable for direct connections.'
@@ -391,7 +414,7 @@ async def delegate(
'role': 'user',
'content': prompt,
'timestamp': int(time.time()),
- 'models': [run['model_id']],
+ 'models': [run['subagent_model_id']],
**({'files': prompt_files} if prompt_files else {}),
}
chat = await Chats.insert_new_chat(
@@ -401,7 +424,7 @@ async def delegate(
chat={
'id': chat_id,
'title': f'Sub-agent: {task[:60]}',
- 'models': [run['model_id']],
+ 'models': [run['subagent_model_id']],
'history': {
'currentId': assistant_message_id,
'messages': {
@@ -413,7 +436,7 @@ async def delegate(
'role': 'assistant',
'content': '',
'done': False,
- 'model': run['model_id'],
+ 'model': run['subagent_model_id'],
'timestamp': int(time.time()),
},
},
@@ -457,7 +480,7 @@ async def delegate(
str(config.get('subagents.system_prompt') or '').strip() or DEFAULT_SUBAGENT_SYSTEM_PROMPT
)
form_data = {
- 'model': run['model_id'],
+ 'model': run['subagent_model_id'],
'messages': [
{
'role': 'system',
@@ -589,6 +612,8 @@ async def delegate(
'type': 'subagent',
'delegation_id': delegation_id,
'subagent_chat_id': chat_id,
+ 'parent_model_id': run['parent_model_id'],
+ 'subagent_model_id': run['subagent_model_id'],
}
pending_message = {
'id': pending_message_id,
@@ -596,7 +621,7 @@ async def delegate(
'childrenIds': [],
'role': 'user',
'content': '\n'.join(lines),
- 'model': run['model_id'],
+ 'model': run['subagent_model_id'],
'meta': pending_meta,
'timestamp': int(time.time()),
}
diff --git a/src/lib/components/admin/Settings/Subagents.svelte b/src/lib/components/admin/Settings/Subagents.svelte
index 47b72233f0..1f132edddc 100644
--- a/src/lib/components/admin/Settings/Subagents.svelte
+++ b/src/lib/components/admin/Settings/Subagents.svelte
@@ -2,7 +2,10 @@
import { getContext, onMount } from 'svelte';
import { toast } from 'svelte-sonner';
+ import { getModels } from '$lib/apis';
import { getSubagentsConfig, setSubagentsConfig } from '$lib/apis/configs';
+ import { getBaseModels } from '$lib/apis/models';
+ import SettingsSelect from '$lib/components/common/SettingsSelect.svelte';
import Spinner from '$lib/components/common/Spinner.svelte';
import Switch from '$lib/components/common/Switch.svelte';
@@ -11,23 +14,50 @@
let loading = true;
let saving = false;
let enabled = false;
+ let subagentModelId = '';
let backgroundEnabled = false;
let maxConcurrent = 20;
let maxAsync = 20;
let maxIterations = 30;
let maxOutput = 30000;
let systemPrompt = '';
+ let models: any[] = [];
+
+ const normalizeModelSelection = (modelId: string | null | undefined) => {
+ if (!modelId) {
+ return '';
+ }
+
+ const model = models.find((item: any) => item.id === modelId);
+ if (!model) {
+ return modelId;
+ }
+
+ return model.id;
+ };
onMount(async () => {
try {
- const config = await getSubagentsConfig(localStorage.token);
+ const [config, workspaceModels, baseModels] = await Promise.all([
+ getSubagentsConfig(localStorage.token),
+ getBaseModels(localStorage.token),
+ getModels(localStorage.token, null, false)
+ ]);
enabled = config?.ENABLE_SUBAGENTS ?? false;
+ subagentModelId = config?.SUBAGENTS_MODEL_ID ?? '';
backgroundEnabled = config?.SUBAGENTS_BACKGROUND_ENABLED ?? false;
maxConcurrent = Number(config?.SUBAGENTS_MAX_CONCURRENT) || 20;
maxAsync = Number(config?.SUBAGENTS_MAX_ASYNC) || 20;
maxIterations = Number(config?.SUBAGENTS_MAX_ITERATIONS) || 30;
maxOutput = Number(config?.SUBAGENTS_MAX_OUTPUT) || 30000;
systemPrompt = config?.SUBAGENTS_SYSTEM_PROMPT ?? '';
+ models = baseModels.map((model: any) => {
+ const workspaceModel = workspaceModels.find((item: any) => item.id === model.id);
+
+ return workspaceModel
+ ? { ...model, ...workspaceModel }
+ : { ...model, id: model.id, name: model.name, is_active: true };
+ });
} catch (error) {
toast.error(`${error}`);
} finally {
@@ -40,6 +70,7 @@
try {
await setSubagentsConfig(localStorage.token, {
ENABLE_SUBAGENTS: enabled,
+ SUBAGENTS_MODEL_ID: subagentModelId,
SUBAGENTS_BACKGROUND_ENABLED: backgroundEnabled,
SUBAGENTS_MAX_CONCURRENT: maxConcurrent,
SUBAGENTS_MAX_ASYNC: maxAsync,
@@ -77,6 +108,34 @@
{#if enabled}
+
+
+
{
+ subagentModelId = normalizeModelSelection(subagentModelId);
+ }}
+ >
+
+ {#each models as model}
+
+ {/each}
+
+
+ {$i18n.t(
+ 'Choose a dedicated model for sub-agents. Current Model follows the active chat model.'
+ )}
+
+