Frequently Asked Questions

Self-Hosting Documentation Access Granted

Which LlamaCloud services communicate with which database/queue/filestore dependencies?

Backend: Postgres, MongoDB, Redis, Filestore
Jobs Service: Postgres, MongoDB, Filestore
Jobs Worker: RabbitMQ, Redis, MongoDB
Usage: MongoDB and Redis
LlamaParse: Consumes from RabbitMQ, Reads/Writes from Filestore
LlamaParse OCR: None

Which Features Require an LLM and what model?

Chat UI: This feature requires the customer’s OpenAI Key to have access to either the Text-Only models and/or the Multi-Modal model (if multi-modal index)

These keys are set up via the Helm chart:

config:
  llms:
    openAi:
      apiKey: <OPENAI-APIKEY>
      # Name of the existing secret to use for the OpenAI API key
      # secret: ""

  # If you are using Azure OpenAI, you can configure it like this:
  # azureOpenAi:
  #  secret: ""
  #  deployments: []
  #  # - model: "gpt-4o-mini"
  #  #   deploymentName: "gpt-4o-mini"
  #  #   apiKey: ""
  #  #   baseUrl: "https://api.openai.com/v1"
  #  #   apiVersion: "2024-08-06"

Embeddings: Credentials to connect to an embedding model provider are input within the application directly during the Index creation workflow.
LlamaParse Fast: Text extraction only. No LLM.
LlamaParse Accurate: This mode uses the gpt-4o under the hood.

LLM API Rate Limits

There will be many instances where you may run into some kind of rate limit with an LLM provider. The easiest way to debug is to view the logs, and if you see a 429 error, increase your tokens per minute limit.

How do I adjust log levels?

  config:
    ## Log level for the application (DEBUG, INFO, WARNING, ERROR, CRITICAL)
    logLevel: INFO

What auth modes are supported at the moment?

We support both OIDC and Basic Auth for self-hosted deployments. For more information, please refer to the Authentication Modes documentation.

Known Issues

BYOC Port-Forwarding with Custom Helm Release Names

Issue: When testing BYOC deployments without ingress setup (using port-forwarding), the backend service must be reachable at http://llamacloud-backend:8000.

Affected Setup:

BYOC deployments without ingress configuration
Using kubectl port-forward for testing

Workarounds (until permanent fix is available):

Manual Service Creation: Create an additional backend service with the expected name.
Setup Ingress: Configure proper ingress instead of relying on port-forwarding. See the Ingress Configuration documentation for details.

Recommendation: For production deployments, always use proper ingress configuration rather than port-forwarding.

Manual Service Creation

When you do not have ingress properly configured you can use these steps as a workaround.

Create a Kuberentes Service object:

apiVersion: v1
kind: Service
metadata:
  name: llamacloud-backend
  namespace: <your namespace>
spec:
  ports:
  - name: http
    port: 8000
    protocol: TCP
    targetPort: http
  selector:
    app.kubernetes.io/instance: llamacloud
    app.kubernetes.io/name: llamacloud
  type: ClusterIP
status:
  loadBalancer: {}

Apply the object: kubectl apply -f llamacloud-backend-service.yaml -n <your namespace>.

You should now be able to create accounts and log into the LlamaCloud UI. If you would also like to test document parsing, you must tell the browswer how to talk to the llamacloud-backend service. You can do this by adding this line to your /etc/hosts file on your local machine:

127.0.0.1  llamacloud-backend