Local Models

Use Dagu with local model servers such as Ollama, llama.cpp, vLLM, or LM Studio when they expose an OpenAI-compatible chat API.

This guide covers workflow steps that use action: chat.completion with a local model provider.

What Dagu Expects

For local models, Dagu expects an OpenAI-compatible /v1 base URL.

Typical value:

http://localhost:11434/v1

Dagu then calls the normal chat-completions route under that base URL.

WARNING

Enter the base URL, not a full endpoint. Do not paste vendor-native endpoints such as /api/generate.

Valid and Invalid Base URLs

Value entered in Dagu	Result
http://localhost:11434/v1	Correct
http://localhost:8080/v1	Correct for a default llama.cpp llama-server
empty	Correct for a local Ollama server on the same machine as the Dagu process
http://localhost:11434/api/generate	Wrong
http://localhost:11434/v1/chat/completions	Wrong

Workflow Example

steps:
  - action: chat.completion
    with:
      provider: local
      model: llama3.2
      base_url: http://localhost:11434/v1
      messages:
        - role: user
          content: |
            Summarize this repository in one paragraph.
    output: RESULT

Dagu also accepts aliases such as ollama, vllm, and llama in workflow YAML, but they all follow the same local-model path.

llama.cpp

llama-server from llama.cpp exposes the same OpenAI-compatible API, serving the model it was started with:

llama-server -m ./model.gguf --port 8080

steps:
  - action: chat.completion
    with:
      provider: llama
      model: model
      base_url: http://localhost:8080/v1
      prompt: |
        Summarize this repository in one paragraph.

provider: llama is an alias for local. The server reports the name it serves at /v1/models; use that as the model value.

Important Networking Note

localhost means the machine or container running Dagu, not the browser.

Examples:

• if Dagu runs directly on your laptop, localhost usually means your laptop
• if Dagu runs in Docker or Kubernetes, localhost means that container or pod
• if a remote node runs the step, localhost means that remote node

Current Limitations

The local-model path is intended for standard text chat.

Consider these constraints:

• Dagu expects the OpenAI-compatible route, not vendor-native endpoints.
• Provider-specific options, such as Ollama native think settings, cannot be configured here.
• Multimodal or image message content is not supported on this path.

If you need vendor-specific behavior, place the service behind a compatible proxy or invoke the tool directly outside this path.

Troubleshooting

404 From Ollama

Most often the base URL is wrong. Use:

http://localhost:11434/v1

Do not use:

• http://localhost:11434/api/generate
• http://localhost:11434/v1/chat/completions

model not found

The model name in Dagu must exactly match the model tag available in your local server.

Works On My Laptop, Fails In Dagu

Check where Dagu is actually running. If Dagu runs in Docker, Kubernetes, or on a remote node, the local server must also be reachable from there.

External References

• Ollama Quickstart
• Ollama OpenAI Compatibility
• llama.cpp server

Related Pages

• LLM Completion
• Providers & Endpoints