DeepFellow Infra

The deepfellow infra group manages the DeepFellow Infra installation running in Docker.

deepfellow infra [OPTIONS] COMMAND [ARGS]...

deepfellow infra Commands

Install Templates

--template accepts the built-in workspace name, or a path to your own YAML template file. A template has two parts: config pre-fills the install settings (fields it sets are applied without prompting), and post_start_actions runs after Infra starts — use it to install services and models automatically.

config:
  port: 8086
  infra_name: infra
  infra_url: http://infra:8086
  docker_network: deepfellow-infra-net

# Runs after Infra starts, in order.
post_start_actions:
  # Start the ollama service (GPU spec, 250k context, idle models unloaded after 5 min).
  - function: infra.service.install
    kwargs:
      name: ollama
      spec: '{"hardware": "GPU", "keep_alive": "5m", "is_flash_attention": true, "context_length": 250000}'
  # Chat model.
  - function: infra.model.install
    kwargs:
      service_name: ollama
      model_name: gemma4:e4b
  # Embedding model.
  - function: infra.model.install
    kwargs:
      service_name: ollama
      model_name: mxbai-embed-large
  # Start the speaches-ai service. Infra picks GPU or CPU itself.
  - function: infra.service.install
    kwargs:
      name: speaches-ai
      spec: '{}'
  # Speech-to-text model Doc Chunker uses to transcribe audio and video.
  - function: infra.model.install
    kwargs:
      service_name: speaches-ai
      model_name: Systran/faster-whisper-large-v3
  # Start the custom service.
  - function: infra.service.install
    kwargs:
      name: custom
  # Doc Chunker with one worker.
  - function: infra.model.install
    kwargs:
      service_name: custom
      model_name: doc_chunker
      spec: '{"max_concurrent": 1}'

Valid config keys:

• port
• infra_name
• infra_url
• docker_network

Valid post_start_actions functions:

• infra.service.install
• infra.model.install

Their kwargs must match the infra service install / infra model install command parameters. Do not set a server kwarg — infra install fills it in itself from the port it actually resolved.

The spec kwarg of infra.model.install is optional. It takes the same JSON object as the --spec option of infra model install. Fields you leave out take Infra's defaults.

The built-in workspace template is equivalent to the example above: it installs the ollama service with the same spec, plus the gemma4:e4b (chat) and mxbai-embed-large (embedding) models. It also installs the speaches-ai service with the Systran/faster-whisper-large-v3 model, and the custom service with the doc_chunker model, limited to one worker (max_concurrent: 1). Doc Chunker describes images with gemma4:e4b and transcribes audio and video with Systran/faster-whisper-large-v3. They match the default doc_chunker_picture_description_model and doc_chunker_stt_model settings of DeepFellow Server, so no further configuration is needed. A full copy of this example ships with the CLI source at deepfellow/infra/templates/example.yaml.

deepfellow infra env

Manage Infra environment variables. For what each variable means and what it defaults to, read Configuration.

deepfellow infra env [OPTIONS] COMMAND [ARGS]...

deepfellow infra config

Manage Infra's dynamic runtime configuration.

deepfellow infra config [OPTIONS] COMMAND [ARGS]...

deepfellow infra service

Manage DeepFellow Infra services.

deepfellow infra service [OPTIONS] COMMAND [ARGS]...

deepfellow infra model

Manage DeepFellow Infra models.

deepfellow infra model [OPTIONS] COMMAND [ARGS]...

deepfellow infra mcp

Manage MCP servers.

deepfellow infra mcp [OPTIONS] COMMAND [ARGS]...