Configuration

All settings are managed via .env. No code changes needed to switch providers.

AI Provider

LARAI_AI_PROVIDER=openai     # openai, anthropic, gemini

Provider	Required Key	Models
`openai`	`OPENAI_API_KEY`	gpt-4o, gpt-4o-mini
`anthropic`	`ANTHROPIC_API_KEY`	claude-sonnet-4-20250514
`gemini`	`GEMINI_API_KEY`	gemini-2.0-flash

LARAI_CHAT_MODEL=gpt-4o-mini
LARAI_EMBEDDING_MODEL=text-embedding-3-small

Vector Store

LARAI_VECTOR_STORE=pinecone  # pinecone, pgvector, none

Pinecone (default, works with any database)

PINECONE_API_KEY=pcsk_your-key
PINECONE_INDEX_HOST=https://your-index.svc.pinecone.io

pgvector (requires PostgreSQL)

DB_CONNECTION=pgsql
LARAI_VECTOR_STORE=pgvector

None (disables RAG)

LARAI_VECTOR_STORE=none

Embedding

LARAI_EMBEDDING_DIMENSIONS=1536
LARAI_SIMILARITY_THRESHOLD=0.4
LARAI_RAG_TOP_K=5

Chunking

LARAI_CHUNK_SIZE=512
LARAI_CHUNK_OVERLAP=50

File Upload

LARAI_MAX_FILE_MB=20
LARAI_STORAGE_DISK=public    # public or s3

Retry / Backoff

Controls retry behavior for rate-limited (429) and server error (5xx) API responses from Pinecone:

LARAI_RETRY_MAX=3            # Max retry attempts
LARAI_RETRY_DELAY_MS=1000    # Base delay in ms (doubles each attempt with jitter)

Conversation

LARAI_HISTORY_TURNS=10

Health Endpoint

Exposes a JSON health check at the configured path. Disabled by default — enable in production for uptime monitoring and alerting.

LARAI_HEALTH_ENABLED=true
LARAI_HEALTH_PATH=_larai/health
LARAI_HEALTH_MIDDLEWARE=auth              # pipe-separated; empty = public

Variable	Default	Notes
`LARAI_HEALTH_ENABLED`	`false`	Route is not registered until this is true
`LARAI_HEALTH_PATH`	`_larai/health`	No leading slash
`LARAI_HEALTH_MIDDLEWARE`	`auth`	Pipe-separated middleware stack

Middleware stack syntax

Use | to separate middleware, not , — commas are reserved for middleware parameters (e.g. Laravel's rate-limit syntax throttle:60,1 = 60 requests per minute).

LARAI_HEALTH_MIDDLEWARE=auth|throttle:60,1   # authenticated + rate-limited
LARAI_HEALTH_MIDDLEWARE=throttle:10,1         # public but rate-limited
LARAI_HEALTH_MIDDLEWARE=                      # fully public (use behind ingress allowlist)

Hit the endpoint with ?deep=true to run live API probes (embedding + vector store) instead of just config validation.

Full `.env` Example

DB_CONNECTION=mysql
DB_HOST=127.0.0.1
DB_PORT=3306
DB_DATABASE=myapp
DB_USERNAME=root
DB_PASSWORD=

LARAI_AI_PROVIDER=openai
OPENAI_API_KEY=sk-proj-...
LARAI_CHAT_MODEL=gpt-4o-mini
LARAI_EMBEDDING_MODEL=text-embedding-3-small

LARAI_VECTOR_STORE=pinecone
PINECONE_API_KEY=pcsk_...
PINECONE_INDEX_HOST=https://my-index.svc.pinecone.io

# All optional — shown with defaults
LARAI_EMBEDDING_DIMENSIONS=1536
LARAI_SIMILARITY_THRESHOLD=0.4
LARAI_RAG_TOP_K=5
LARAI_CHUNK_SIZE=512
LARAI_CHUNK_OVERLAP=50
LARAI_MAX_FILE_MB=20
LARAI_HISTORY_TURNS=10
LARAI_RETRY_MAX=3
LARAI_RETRY_DELAY_MS=1000

Switching Providers

Change .env, then clear config:

php artisan config:clear
php artisan larai:doctor

Larai Kit Laravel Package