Agent Builder list All models when accessed from Marketplace

[romy651]

What happened?

When creating or editing an agent from the Agent Marketplace, all models for each configured provider are listed. The model provider dropdown shows ALL available models (for each provider) instead of respecting the configuration from librechat.yaml. However, when accessing the agent builder from the main chat interface, everything works correctly.
Expected Behavior:
The agent builder should only show the models configured from the YAML configuration file, regardless of whether it's opened from the marketplace or main chat.
Actual Behavior:
✅ Main chat → Agent builder: Shows the configured models in each provider (works correctly)
❌ Marketplace → Agent builder: Shows every model in each provider

Screen.Recording.2026-01-07.at.19.02.55.mov

Version Information

0.8.2-rc2

Steps to Reproduce

1. Configure allowedProviders in librechat.yaml under the agents endpoint (e.g., allowedProviders: ["Anthropic"]) and allow only one model, e.g., Haiku 4.5.

2. Open Agent Marketplace and refresh → Open the agent builder (by opening the sidebar),

3. Click any agent card to edit OR create a new agent → Open the model selection panel

4. Observe: for each provider, every model is shown (even the ones that are not configured) except the ones that are actually configured (e.g, Haiku 4.5)

What browsers are you seeing the problem on?

No response

Relevant log output

n/a

Screenshots

No response

Code of Conduct

• I agree to follow this project's Code of Conduct

[danny-avila]

Steps to Reproduce

I followed these steps and was not able to reproduce, can you record a screen capture?

Open Agent Marketplace and refresh → Click any agent card to edit OR create a new agent → Open the model selection panel

Creating/editing a new agent from the agent marketplace UI is not possible (center of screen). Doing so from the builder (right-hand side of screen) did not have the observed effect:

Observe: All providers are shown instead of only the configured ones (compare with agent builder opened from main chat, which correctly filters providers)

Any combination of navigating to or refreshing from the marketplace page did not reproduce the behavior.

Also, try updating to 0.8.2-rc2 as it may be due to a separate issue already resolved.

[romy651]

Hi @danny-avila,
Thanks for the quick response. I just upgraded to 0.8.2-rc2, and the issue persists. I recorded and uploaded a video regarding this.
When opening the agent builder from the marketplace, the providers are listed correctly, but the models for the providers are not.

[danny-avila]

can you share your configuration? note, it's case sensitive and should be anthropic for the official Anthropic endpoint.

[romy651]

Here is the config file I am working on:

interface:
  customWelcome: 'Welcome to Librechat demo! Enjoy your experience.'
  fileSearch: false
  privacyPolicy:
    externalUrl: 'https://librechat.ai/privacy-policy'
    openNewTab: true

  termsOfService:
    externalUrl: 'https://librechat.ai/tos'
    openNewTab: true
    modalAcceptance: true
    modalTitle: 'Terms of Service for LibreChat'
    modalContent: ...

  endpointsMenu: true
  modelSelect: true
  parameters: true
  sidePanel: true
  presets: true
  prompts: true
  bookmarks: true
  multiConvo: true
  agents: true
  peoplePicker:
    users: true
    groups: true
    roles: true
  marketplace:
      use: true
  fileCitations: true
  # temporaryChatRetention: 1
filteredTools:
  - "execute_code"

registration:
  socialLogins: ['github', 'google', 'discord', 'openid', 'facebook', 'apple', 'saml']

actions:
  allowedDomains:
    - 'swapi.dev'
    - 'librechat.ai'
    - 'google.com'


 
endpoints:
  openAI:
    plugins: false

  assistants: 
    disableBuilder: true # Disable Assistants Builder Interface by setting to `true`

  agents:
    recursionLimit: 50
    maxRecursionLimit: 100
    disableBuilder: false
    maxCitations: 30
    maxCitationsPerFile: 7
    minRelevanceScore: 0.45
    capabilities: 
      - "execute_code"
      - "file_search"
      - "actions"
      - "tools"
      - "artifacts"
      - "context"
      - "chain"
      - "web_search"
    
    allowedProviders: ["bedrock", "anthropic"]

  custom:
    - name: 'groq'
      apiKey: '${GROQ_API_KEY}'
      baseURL: 'https://api.groq.com/openai/v1/'
      models:
        default:
          - 'llama3-70b-8192'
          - 'llama3-8b-8192'
          - 'llama2-70b-4096'
          - 'mixtral-8x7b-32768'
          - 'gemma-7b-it'
        fetch: false
      titleConvo: true
      titleModel: 'mixtral-8x7b-32768'
      modelDisplayLabel: 'groq'

      # Models configuration
      models:
        default: ['mistral-tiny', 'mistral-small', 'mistral-medium']
        fetch: true # Defaults to false.

      titleConvo: true # Set to true to enable title conversation
      titleModel: 'mistral-tiny' # Defaults to "gpt-3.5-turbo" if omitted.
      modelDisplayLabel: 'Mistral' # Default is "AI" when not set.
      dropParams: ['stop', 'user', 'frequency_penalty', 'presence_penalty']

    - name: 'OpenRouter'
      # For `apiKey` and `baseURL`, you can use environment variables that you define.
      # recommended environment variables:
      apiKey: '${OPENROUTER_KEY}'
      baseURL: 'https://openrouter.ai/api/v1'
      headers:
          x-librechat-body-parentmessageid: '{{LIBRECHAT_BODY_PARENTMESSAGEID}}'
      models:
        default: ['meta-llama/llama-3-70b-instruct']
        fetch: true
      titleConvo: true
      titleModel: 'meta-llama/llama-3-70b-instruct'
      dropParams: ['stop']
      modelDisplayLabel: 'OpenRouter'

    # Portkey AI Example
    - name: 'Portkey'
      apiKey: 'dummy'
      baseURL: 'https://api.portkey.ai/v1'
      headers:
        x-portkey-api-key: '${PORTKEY_API_KEY}'
        x-portkey-virtual-key: '${PORTKEY_OPENAI_VIRTUAL_KEY}'
      models:
        default: ['gpt-4o-mini', 'gpt-4o', 'chatgpt-4o-latest']
        fetch: true
      titleConvo: true
      titleModel: 'current_model'
      summarize: false
      summaryModel: 'current_model'
      forcePrompt: false
      modelDisplayLabel: 'Portkey'
      iconURL: https://images.crunchbase.com/image/upload/c_pad,f_auto,q_auto:eco,dpr_1/rjqy7ghvjoiu4cd1xjbf

    - name: "AI Studio"
      label: "AI Studio"
      description: "AI Studio assistant"
      apiKey: "0p3n-w3bu!"
      baseURL: "http://0.0.0.0:9099"
      models:
        default: ["open_ai_pipeline"]
        fetch: true
      # No 'group' field - appears as standalone item at top level (not nested)


fileConfig:
  serverFileSizeLimit: 25  # 25MB limit per file
  avatarSizeLimit: 2
  endpoints:
    default:
      fileLimit: 10
      fileSizeLimit: 10  # Maximum size for an individual file in MB
      totalSizeLimit: 50  # Maximum total size for all files in a single request in MB
      supportedMimeTypes:
        - "image/.*"
        - "application/pdf"


modelSpecs:
  enforce: false
  prioritize: true
  addedEndpoints:
    - "openAI"
    - "AI Studio"
    - "agents"
  list:
    # a list of model specs
    - name: "Librechat Agent"
      label: "Librechat Agent"
      description: "the Librechat Agent assistant"
      iconURL: "https://librechat-dev.kindredgroup.com/images/691c608fbe748ac988502acf/agent-agent_TFHmfU-FaFpOu2Z2ys_bv-avatar-1765203949991.png?manual=false"
      default: true
      preset:
        endpoint: "agents"
        agent_id: "agent_5EbTBXiMBgNPkBFokWHy0"
        greeting: "You are the Librechat Agent assistant. You are here to help with the Librechat platform."
    - name: "Claude Haiku 4.5"
      label: "Claude Haiku 4.5"
      description: "Claude Haiku 4.5 assistant"
      preset:
        endpoint: "anthropic"
        model: "claude-haiku-4-5-20251001"
    
    - name: "OpenAI GPT-4o"
      label: "OpenAI GPT-4o"
      description: "OpenAI GPT-4o assistant"
      preset:
        endpoint: "openAI"
        model: "gpt-4o"

    - name: "OpenAI GPT-4o-mini"
      label: "OpenAI GPT-4o-mini"
      description: "OpenAI GPT-4o-mini assistant"
      preset:
        endpoint: "openAI"
        model: "gpt-4o-mini"

    - name: "bedrock-claude-haiku-4-5"
      label: "Claude Haiku 4.5 (bedrock)"
      description: "Claude Haiku 4.5 assistant"
      preset:
        endpoint: "bedrock"
        model: "arn:aws:bedrock:eu-central-1:029109262688:inference-profile/eu.anthropic.claude-haiku-4-5-20251001-v1:0"

[dustinhealy]

Hi @romy651,

Thanks for raising this, just opened a PR #11253 that appears to address the cause of this behavior. I was able to reproduce your issue, and after testing it for a couple minutes, it seems like the allowed models are now being respected consistently. The root cause was the query hook grabbing a default list from our config as initialData and the contained lists getting cached and served to the frontend until they had a chance to be invalidated and the proper model lists served instead.

[dustinhealy]

Note: a more focused PR #11260 was prioritized to fix this behavior