Ollama Endpoint Not Showing in LibreChat UI Despite Correct Setup (Windows + Docker)

[neverresting77]

Goal:
I'm trying to integrate my locally running Ollama instance (installed via the Windows installer, running directly on the host) with LibreChat (v0.7.7, running in Docker via docker compose).

Environment:

• OS: Windows 10 (Version 10.0.19045.5608)
• Docker: Docker Desktop 28.0.4
• LibreChat: v0.7.7 (Cloned from GitHub, running via docker compose)
• Ollama: Running directly on the Windows host (installed via official installer), accessible at http://localhost:11434.

Problem:
The "Ollama" endpoint and its models do not appear in the LibreChat model selection menu in the web UI (http://localhost:3080). The LibreChat container starts without apparent errors in the logs related to YAML loading, and network connectivity to the host Ollama server from within the container is confirmed.

Troubleshooting Steps Performed:

1. Initial Setup: Successfully installed LibreChat using Docker Compose, including cloning the repo, copying .env.example to .env, creating docker-compose.yml from the repository source, creating necessary data directories, and mounting the .env file.
2. librechat.yaml Configuration:
   • Created librechat.yaml in the project root.
   • Mounted the file into the api container using docker-compose.yml:

     volumes:
       - type: bind
         source: ./librechat.yaml
         target: /app/librechat.yaml

   • Configured the endpoints.custom section for Ollama. The final, simplified configuration (after several iterations) is:

     version: 1.2.1
     cache: true
     interface:
       customWelcome: "Welcome to LibreChat! Enjoy your experience."
     privacyPolicy:
       externalUrl: 'https://librechat.ai/privacy-policy'
       openNewTab: true
     termsOfService:
       externalUrl: 'https://librechat.ai/terms-of-service'
       openNewTab: true
     endpoints:
       custom:
         - name: "Ollama"
           apiKey: "ollama"          # Re-added based on guides, seems required by LibreChat
           baseURL: "http://host.docker.internal:11434/v1/" # For accessing host from Docker
           models:
             # Updated based on `ollama list` output
             default: ["llama3:8b-instruct-q4_K_M", "mistral:7b"]
             fetch: true
           titleModel: "current_model" # Changed from "self-host"
           modelDisplayLabel: "Ollama"
           order: 10
     mcp:
       enabled: false
       port: 3090
       plugins: {}

   • YAML Errors Resolved: During the process, corrected errors found in logs, including "invalid API key" (initially removed apiKey, then re-added it) and "duplicated mapping key" / "invalid YAML format" (removed duplicate endpoints: section and stray ```).
3. .env Configuration:
   • To isolate the Ollama issue, API keys for other services (OpenAI, Google, Anthropic, Assistants) were commented out (e.g., # OPENAI_API_KEY=user_provided). The ENDPOINTS variable remains commented out.
4. Network Connectivity Test:
   • Successfully confirmed that the LibreChat container can reach the host Ollama server using:

     docker exec LibreChat curl http://host.docker.internal:11434/api/tags

   • This command returned the start of the JSON model list ({"models":[{"name":"llama...), indicating successful communication.
5. Container Management: Consistently restarted containers using docker compose down && docker compose up -d after each configuration change.
6. LibreChat Reinstallation: Performed a full reinstallation by stopping/removing containers and volumes (docker compose down --volumes), deleting the LibreChat directory, re-cloning, and repeating the setup steps.
7. Browser Cache: Cleared browser cache and performed hard refreshes.
8. Log Analysis: Current docker logs LibreChat show no errors related to YAML loading or API keys after the latest corrections. The server starts successfully and listens on port 3080.

Current Status:
Despite all the steps above, the Ollama endpoint is still not visible in the LibreChat UI model selection menu. Only the default "Agents" item might appear, or standard endpoints if their API keys are uncommented in .env.

Request:
I've exhausted the common troubleshooting steps. Could there be a subtle issue with the librechat.yaml configuration for Ollama specifically in v0.7.7 on Windows, a specific Docker Desktop interaction, or another hidden factor preventing the endpoint from registering correctly in the UI? Any insights or suggestions would be greatly appreciated.

[neverresting77]

Finally, I found the soultion by myself.

Troubleshooting Log: Connecting Local Ollama to LibreChat (v0.7.7) on Docker

Goal

Configure a Dockerized LibreChat (v0.7.7) instance to use models hosted on a local Ollama server.

Initial State

• LibreChat (v0.7.7) running via Docker Desktop on a Windows environment.
• Local Ollama server installed with necessary models (e.g., llama3:8b-instruct-q4_K_M, mistral:7b) downloaded.
• LibreChat UI, by default, only shows external API models like OpenAI.

Troubleshooting Process

1. Issue: Ollama Endpoint Configuration Not Applied

• Symptom: Modifying the .env file to add Ollama settings or changing existing endpoint variables has no effect on the UI. Only default models (like OpenAI) are displayed.
• Diagnosis: LibreChat likely prioritizes the ENDPOINTS environment variable or the librechat.yaml file over individual endpoint settings in the .env file. If ENDPOINTS is set to something else (or perhaps even its default), it might ignore other API key settings in .env.
• Attempt 1 (Modify .env):
  • Locate the ENDPOINTS variable in the .env file and set it to use librechat.yaml:

    ENDPOINTS=librechat.yaml

  • Comment out or leave API key variables for other endpoints (OpenAI, Google, etc.) as they are (e.g., user_provided), as they will be ignored when librechat.yaml is used.
• Attempt 2 (Create and Configure librechat.yaml):
  • Create a librechat.yaml file in the LibreChat project root.
  • Add a custom endpoint configuration for Ollama:

    version: 1.2.1 # Specify the config version
    endpoints:
      custom:
        - name: "Ollama"
          apiKey: "ollama"  # Dummy value as Ollama doesn't require an API key
          baseURL: "http://host.docker.internal:11434/v1/" # Address for the Docker container to reach the host's Ollama
          models:
            default: ["llama3:8b-instruct-q4_K_M", "mistral:7b"] # Specify the exact names of the models to use
            fetch: true # Enable dynamic loading of the model list from the Ollama server
          titleModel: "current_model" # Use the current model for generating conversation titles
          modelDisplayLabel: "Ollama" # Label for the endpoint shown in the UI
          order: 10 # Display order in the UI

  • Rebuild and restart Docker containers: docker-compose up -d --build
• Result: All models disappeared from the UI. This is a positive sign, indicating that the ENDPOINTS=librechat.yaml setting is now active, and LibreChat is attempting to read the librechat.yaml file instead of .env. However, the Ollama models are still not displayed.

2. Issue: Models Not Displayed After Loading librechat.yaml

• Symptom: The model selection menu in the UI is empty after setting ENDPOINTS=librechat.yaml.
• Diagnosis 1 (Potential librechat.yaml Content Error):
  • The fetch: true option might be disabled, or the model names specified in models.default might not exactly match the names on the Ollama server.
  • Test: Comment out fetch: true, simplify models.default (e.g., ["llama3"]) -> Failed. Re-enable fetch: true, use exact model names -> Failed.
• Diagnosis 2 (Check LibreChat API Container Logs):
  • Command: docker logs LibreChat --tail 100 (Later corrected to docker logs api --tail 100 after identifying the correct service name).
  • Error Found:

    error: Invalid custom config file at /app/librechat.yaml:
    ...
    message: "Unrecognized key(s) in object: 'privacyPolicy', 'termsOfService', 'tools', 'mcp'"
    ...
    

    The current LibreChat version (v0.7.7) does not recognize certain configuration keys (privacyPolicy, termsOfService, tools, mcp) present in the librechat.yaml file (likely from a newer template). This prevents the file from being parsed correctly.
• Solution: Comment out the entire sections related to the unrecognized keys (privacyPolicy, termsOfService, tools, mcp) in the librechat.yaml file.
• Application:
  1. Modify librechat.yaml (comment out the problematic sections).
  2. Restart the Docker container (using docker-compose restart api after confirming the service name).
• Result: The Ollama models configured in librechat.yaml (including those fetched dynamically via fetch: true, like llama3:8b-instruct-q4_K_M, mistral:7b) are now correctly displayed in the UI's model selection menu.

3. Side Issue: Docker Service Restart Error

• Symptom: Running docker-compose restart LibreChat results in a no such service: LibreChat error.
• Cause: Mismatch between the service name defined in docker-compose.yml and the name used in the command. Checking the file revealed the service name is api, while the container name is LibreChat.
• Solution: Use the correct service name api in the command: docker-compose restart api.

Final Successful Configuration Summary

• .env file: ENDPOINTS=librechat.yaml setting enabled.
• librechat.yaml file:
  • version: 1.2.1 specified.
  • Ollama configuration added under the endpoints.custom section (including baseURL, models.default, fetch: true).
  • Keys unrecognized by v0.7.7 (privacyPolicy, termsOfService, tools, mcp) commented out.
• docker-compose.yml file: No changes needed (aware that the service name is api).
  

[danny-avila]

  * Locate the `ENDPOINTS` variable in the `.env` file and set it to use `librechat.yaml`:
    ```dotenv
    ENDPOINTS=librechat.yaml
    ```

For using Ollama exclusively, you can use ENDPOINTS=custom, that is the correct configuration

[neverresting77]

Hi Danny,

Thanks for the suggestion to use ENDPOINTS=custom in the .env file when exclusively using Ollama via librechat.yaml.
I'm trying to integrate my local Ollama setup with LibreChat v0.7.7 and also utilize the Tools and MCP features defined in librechat.yaml.

However, I'm encountering an issue:

1. When I uncomment the tools: and mcp: sections at the top level of my librechat.yaml file (following the structure in librechat.example.yaml), the LibreChat container fails to load the configuration correctly. The logs show an Invalid custom config file ... Unrecognized key(s) in object: 'tools', 'mcp' error (ZodError). This results in the model selection menu disappearing or becoming empty in the UI.
2. Commenting out the tools: and mcp: sections allows the endpoints.custom.Ollama configuration to load successfully (using either ENDPOINTS=librechat.yaml or ENDPOINTS=custom in .env), and I can use my Ollama models.
3. Setting ENDPOINTS=custom in .env unfortunately doesn't resolve the issue if the tools: or mcp: keys are uncommented in librechat.yaml, as the file parsing fails before the endpoint selection logic seems to apply.

It appears that LibreChat v0.7.7 does not recognize the top-level tools: and mcp: keys in librechat.yaml.
Is there a different way to configure Tools and MCP integration in v0.7.7, specifically when primarily using a local LLM like Ollama defined in the endpoints.custom section? I'd really like to leverage these features alongside my local models.

Thanks for your help!

[tandryukha]

I solved it following these steps:

1. specified local ollama models in librechat.yaml

  
  custom:
    - name: "Ollama"
      apiKey: "ollama"  # Dummy value as Ollama doesn't require an API key
      baseURL: "http://host.docker.internal:11434/v1/" # Address for the Docker container to reach the host's Ollama
      models:
        default: ["deepseek-coder:latest"] # Specify the exact names of the models to use
        fetch: true # Enable dynamic loading of the model list from the Ollama server
      titleModel: "current_model" # Use the current model for generating conversation titles
      modelDisplayLabel: "Ollama" # Label for the endpoint shown in the UI
      order: 10 # Display order in the UI

2. specified both custom and pre-built endpoints in .env
   ENDPOINTS=openAI,ollama-deepseek

3. added the volume in docker-compose.yml for librechat.yaml:

      - type: bind
        source: ./.env
        target: /app/.env
      - type: bind
        source: ./librechat.yaml
        target: /app/librechat.yaml```

[tandryukha]

[CyprienGille]

Your 3rd step is what fixed the same issue for me, thank you!

[kwadwoadu]

Thank you! I've been dealing with this all day!

[thebearking-cmd]

Discussion #6779: Ollama Endpoint Not Showing (Chiming in)

Experiencing the same issue on Garuda Linux (Arch-based) with LibreChat v0.7.9 Docker and native Ollama 0.2.9.
My Setup

LibreChat running in Docker (v0.7.9)
Ollama running natively on host at 0.0.0.0:11434
Firewall port 11434 open

What I've Tried

1) .env configuration:

envENABLE_OLLAMA=true
OLLAMA_BASE_URL=http://172.17.0.1:11434

Also tried: localhost, 127.0.0.1, host.docker.internal

2) Verified connectivity:

From host: curl http://localhost:11434/api/tags (shows all models)
From inside container: wget http://172.17.0.1:11434/api/tags (works)
Ollama is definitely reachable from the container

3) docker-compose.override.yml attempts:

network_mode: host - breaks LibreChat UI entirely
extra_hosts entries - no effect

Result
Same as OP - backend logs show [Provider] Ollama not enabled, skipping model init despite ENABLE_OLLAMA=true

Additional Context
This issue led me to attempt building from source, which revealed a separate bug with Bun builds failing (will file separately). After 6+ hours across two days, still no working Ollama integration. Had to build from source due to binaries being yanked the morning after I nuked a failed repair attempt.

[danny-avila]

Hi @thebearking-cmd

You are referencing environment variables that may be specific to Ollama but not LibreChat, and you haven't shared your custom endpoint config.

You need your librechat.yaml to have something like the following:

endpoints:
  custom:
    - name: "Ollama"
      apiKey: "ollama"
      # `baseURL` should be whatever is accessible in your LC env
      # baseURL: "http://host.docker.internal:11434/v1/"
      baseURL: "http://127.0.0.1:11434/v1/"
      models:
        default: [
          "llama3.3",
          "llama3.2",
          "llama3.2-vision",
          "qwen2.5",
          "qwen2.5-coder",
          "command-r",
          "mixtral",
          "phi3",
          "phi4",
          "deepseek-r1:32b",
          ]
        fetch: true
      titleConvo: true
      titleModel: "current_model"
      modelDisplayLabel: "Ollama"

Please review the following docs:

• https://www.librechat.ai/docs/quick_start/custom_endpoints
• https://www.librechat.ai/docs/configuration/librechat_yaml/ai_endpoints/ollama