Overview
The Chat with Flow feature lets users interact with a deployed flow through a conversational chat interface directly in the platform UI. It is also used internally for other platform AI features such as generating automation descriptions and improving Studio prompts.
This feature requires a built-in LLM to be configured via Helm chart values. Without it, users will see a generic error message when attempting to chat with a flow:
“Sorry, I’m having trouble answering your question. Please try again later.”
This is separate from Studio V2, which has its own setup process. See the Studio V2 guide for that configuration. Prerequisites
- CrewAI Platform is deployed and running
- An API key for a supported LLM provider (OpenAI or Anthropic)
Configuration
Three Helm chart values control the built-in LLM. Two are non-sensitive environment variables and one is a secret:
Environment Variables (non-sensitive)
Set these under env in your values.yaml:
env:
# Built-in LLM configuration
# Used for chatting with flows, generating automation descriptions,
# improving Studio prompts, etc.
BUILT_IN_LLM_MODEL: "gpt-4.1-mini" # Model name
BUILT_IN_LLM_PROVIDER: "openai" # "openai" or "anthropic"
| Variable | Description | Default | Required |
|---|
BUILT_IN_LLM_PROVIDER | LLM provider to use. Supported values: openai, anthropic. | openai | Yes |
BUILT_IN_LLM_MODEL | Model identifier for the chosen provider (e.g. gpt-4.1-mini, claude-sonnet-4-20250514). | gpt-4.1-mini | Yes |
Secrets (sensitive)
Set this under secrets in your values.yaml:
secrets:
BUILT_IN_LLM_API_KEY: "sk-..." # Your LLM provider API key
| Variable | Description | Required |
|---|
BUILT_IN_LLM_API_KEY | API key for the configured LLM provider. | Yes |
If BUILT_IN_LLM_API_KEY is not set or is invalid, all Chat with Flow attempts will fail. A 401 error in the logs typically means the API key is missing, expired, or incorrect.
Apply the Configuration
After updating your values.yaml, upgrade the Helm release:
helm upgrade crewai <chart-path> -f values.yaml -n <namespace>
Verify
- Deploy a flow (or use an existing deployed flow)
- Click Chat with this crew from the deployment dashboard
- Send a test message (e.g., “Hi”)
- You should receive a response from the flow instead of an error
Troubleshooting
| Symptom | Likely Cause | Fix |
|---|
| ”Sorry, I’m having trouble answering your question” | BUILT_IN_LLM_API_KEY not set or invalid | Set a valid API key in secrets.BUILT_IN_LLM_API_KEY and upgrade the release |
| 401 Unauthorized in logs | Expired or incorrect API key | Rotate the API key with your LLM provider and update the secret |
| Chat option not visible | Flow is not deployed or not in “Online” status | Check the deployment dashboard — the flow must be Online |
| Model not found errors | BUILT_IN_LLM_MODEL set to an unsupported model | Use a valid model name for the configured provider |
