Skip to main content
The assistant requires a provider and model to be configured before use. All other settings have sensible defaults.

​
Provider and Model

Set the provider and model via CLI: 
qai config set assist.provider ollama
qai config set assist.model llama3.1
Or use the Settings page in the web UI — the Assistant section has provider and model selectors that save to the same configuration store. The assist.model value is the model name without a provider prefix. The runtime builds the full provider/model string from assist.provider and assist.model automatically. See LLM Provider Configuration for the full list of supported providers.

​
Environment Variable Override

You can set provider and model via environment variables instead of the config store: 
export QAI_ASSIST_PROVIDER="ollama"
export QAI_ASSIST_MODEL="llama3.1"
Environment variables take precedence over stored settings.

​
Base URL

If your assistant’s LLM provider runs on a remote host or non-default port, set a custom base URL. This is useful when targeting a remote Ollama instance, LM Studio on another machine, or any custom OpenAI-compatible endpoint. 
qai config set assist.base_url http://inference-server:11434
Or via environment variable: 
export QAI_ASSIST_BASE_URL="http://inference-server:11434"
You can also set it in the Settings page under the Assistant section.
assist.base_url controls only the assistant’s LLM endpoint. It is completely independent of the base URLs used for testing targets (e.g., ollama.base_url). You can point the assistant at one Ollama instance while running inject campaigns against a different one.
If not set, litellm uses the provider’s default endpoint (e.g., http://localhost:11434 for Ollama).

​
Credentials

Cloud providers require an API key. Store credentials using the same keyring/environment variable pattern as other qai providers: 
# Store in OS keyring (masked input)
qai config set-credential anthropic

# Or set as environment variable
export ANTHROPIC_API_KEY="sk-ant-..."
Local providers — ollama, lmstudio, and custom — skip the credential check entirely. If you’re using Ollama, no API key is needed; just ensure Ollama is running (ollama serve). See LLM Provider Configuration for credential management details.

​
Embedding Model

The knowledge base uses a sentence-transformers model to generate embeddings for retrieval. The default is all-MiniLM-L6-v2, which runs locally and requires no API key. To override: 
qai config set assist.embedding_model "all-mpnet-base-v2"
Or via environment variable: 
export QAI_ASSIST_EMBEDDING_MODEL="all-mpnet-base-v2"
The embedding model runs locally via sentence-transformers regardless of which LLM provider you use. If you change the embedding model, run qai assist reindex to rebuild the knowledge base with the new embeddings.

​
User Knowledge Directory

The assistant indexes user-provided reference files from ~/.qai/knowledge/ by default. To use a different directory: 
qai config set assist.knowledge_dir "/path/to/your/docs"
Or via environment variable: 
export QAI_ASSIST_KNOWLEDGE_DIR="/path/to/your/docs"
See Knowledge Base for what to put in this directory.

​
Configuration Summary

SettingCLIEnv VarDefault
Providerqai config set assist.providerQAI_ASSIST_PROVIDERNone (required)
Modelqai config set assist.modelQAI_ASSIST_MODELNone (required)
Base URLqai config set assist.base_urlQAI_ASSIST_BASE_URLNone (provider default)
Embedding modelqai config set assist.embedding_modelQAI_ASSIST_EMBEDDING_MODELall-MiniLM-L6-v2
Knowledge directoryqai config set assist.knowledge_dirQAI_ASSIST_KNOWLEDGE_DIR~/.qai/knowledge/
Credentialsqai config set-credential <provider>{PROVIDER}_API_KEYNone (required for cloud)

​
Quick Start

Three commands to get started with a local model: 
qai config set assist.provider ollama
qai config set assist.model llama3.1
qai assist "what can qai test?"
For a cloud provider, add a credential first: 
qai config set assist.provider anthropic
qai config set assist.model claude-sonnet-4-20250514
qai config set-credential anthropic
qai assist "how do I scan an MCP server?"