Project Config

The .vai.json file stores project-level defaults so you don't need to repeat flags on every command. Create one with:

vai init

This launches an interactive wizard. For a non-interactive setup with sensible defaults:

vai init --yes

Schema

{
  "model": "voyage-4-large",
  "db": "myapp",
  "collection": "knowledge",
  "field": "embedding",
  "dimensions": 1024,
  "chunk": {
    "strategy": "recursive",
    "size": 512,
    "overlap": 50
  },
  "chat": {
    "provider": "anthropic",
    "model": "claude-sonnet-4-20250514",
    "maxDocs": 5,
    "maxTurns": 10,
    "systemPrompt": "You are a helpful assistant.",
    "history": true
  }
}

Fields

Field	Type	Default	Description
model	string	voyage-4-large	Voyage AI embedding model
db	string	—	MongoDB database name
collection	string	—	MongoDB collection name
field	string	embedding	Field name for storing vectors
dimensions	number	1024	Embedding dimensions
chunk.strategy	string	recursive	Chunking strategy: fixed, sentence, paragraph, recursive, markdown
chunk.size	number	512	Target chunk size in characters
chunk.overlap	number	50	Overlap between adjacent chunks
chat.provider	string	—	LLM provider: anthropic, openai, ollama
chat.model	string	—	LLM model name
chat.maxDocs	number	5	Max documents per query
chat.maxTurns	number	10	Max conversation turns as context
chat.systemPrompt	string	—	System prompt for the LLM
chat.history	boolean	true	Persist chat history to MongoDB

Override Precedence

CLI flags always take priority over .vai.json:

# Uses .vai.json defaults
vai query "how does auth work?"

# Overrides model and collection for this command only
vai query "how does auth work?" --model voyage-4 --collection archive

Next Steps

• Shell Completions: Tab completion for commands and flags
• Configuration: Environment variables and config store