Skip to content

OpenAI

Configure HolmesGPT to use OpenAI's GPT models.

Setup

Get a paid OpenAI API key.

Note

Requires a paid OpenAI API key, not a ChatGPT Plus subscription.

Configuration

export OPENAI_API_KEY="your-openai-api-key"
holmes ask "what pods are failing?"

Create Kubernetes Secret: 

kubectl create secret generic holmes-secrets \
  --from-literal=openai-api-key="sk-..." \
  -n <namespace>

Configure Helm Values: 

# values.yaml
additionalEnvVars:
  - name: OPENAI_API_KEY
    valueFrom:
      secretKeyRef:
        name: holmes-secrets
        key: openai-api-key

# Configure at least one model using modelList
modelList:
  gpt-4o:
    api_key: "{{ env.OPENAI_API_KEY }}"
    model: openai/gpt-4o
    temperature: 0

  gpt-4o-mini:
    api_key: "{{ env.OPENAI_API_KEY }}"
    model: openai/gpt-4o-mini
    temperature: 0

  gpt-5:
    api_key: "{{ env.OPENAI_API_KEY }}"
    model: openai/gpt-5
    temperature: 1
    reasoning_effort: medium

# Optional: Set default model (use modelList key name, not the model path)
config:
  model: "gpt-4o"  # This refers to the key name in modelList above

Create Kubernetes Secret: 

kubectl create secret generic robusta-holmes-secret \
  --from-literal=openai-api-key="sk-..." \
  -n <namespace>

Configure Helm Values: ```yaml

values.yaml

holmes: additionalEnvVars: - name: OPENAI_API_KEY valueFrom: secretKeyRef: name: robusta-holmes-secret key: openai-api-key

# Configure at least one model using modelList modelList: gpt-4.1: api_key: "{{ env.OPENAI_API_KEY }}" model: openai/gpt-4.1 temperature: 0

gpt-4o-mini:
  api_key: "{{ env.OPENAI_API_KEY }}"
  model: openai/gpt-4o-mini
  temperature: 0

gpt-5:
  api_key: "{{ env.OPENAI_API_KEY }}"
  model: openai/gpt-5
  temperature: 1
  reasoning_effort: medium

# Optional: Set default model (use modelList key name, not the model path) config: model: "gpt-4.1" # This refers to the key name in modelList above

Using CLI Parameters

You can also pass the API key directly as a command-line parameter:

holmes ask "what pods are failing?" --api-key="your-api-key"

Available Models

# GPT-4.1 (default) - fast and capable
holmes ask "what pods are failing?"

# GPT-5 (more accurate but much slower)
holmes ask "what pods are failing?" --model="gpt-5"

Best Results

For optimal investigation quality, consider using Anthropic's Claude models:

  • Claude Opus 4.1: Most powerful for complex investigations
  • Claude Sonnet 4.5 or 4: Best balance of speed and quality

GPT-4.1 provides a good alternative with faster response times.

See benchmark results for detailed model performance comparisons.

GPT-5 Reasoning Effort

When using GPT-5 models, you can control the reasoning effort level by setting the REASONING_EFFORT environment variable. This allows you to balance between response quality and processing time/cost.

# Use minimal reasoning effort for faster responses
export REASONING_EFFORT="minimal"
holmes ask "what pods are failing?" --model="gpt-5"

# Use default reasoning effort
export REASONING_EFFORT="medium"
holmes ask "what pods are failing?" --model="gpt-5"

# Use high reasoning effort for complex investigations
export REASONING_EFFORT="high"
holmes ask "what pods are failing?" --model="gpt-5"

Available reasoning effort levels:

  • minimal - Fastest responses, suitable for simple queries
  • low - Balance between speed and quality
  • medium - Standard reasoning depth (default)
  • high - Deeper reasoning for complex problems

For more details on reasoning effort levels, refer to the OpenAI documentation.

Additional Resources

HolmesGPT uses the LiteLLM API to support OpenAI provider. Refer to LiteLLM OpenAI docs for more details.