morphik.toml.example

# Example configuration file for Morphik Core
# Copy this file to morphik.toml and fill in your API keys and settings
# DO NOT commit morphik.toml with real API keys to version control

[api]
host = "0.0.0.0"
port = 8000
reload = true

[auth]
jwt_algorithm = "HS256"
dev_mode = false  # Set to true for easier local development
dev_entity_id = "dev_user"  # Default dev user ID
dev_entity_type = "developer"  # Default dev entity type
dev_permissions = ["read", "write", "admin"]  # Default dev permissions

#### Registered models
[registered_models]
# OpenAI models
openai_gpt4-1 = { model_name = "gpt-4.1" }
openai_gpt4-1-mini = { model_name = "gpt-4.1-mini" }

# Azure OpenAI models
azure_gpt4 = { model_name = "gpt-4", api_base = "YOUR_AZURE_URL_HERE", api_version = "2023-05-15", deployment_id = "gpt-4-deployment" }
azure_gpt35 = { model_name = "gpt-3.5-turbo", api_base = "YOUR_AZURE_URL_HERE", api_version = "2023-05-15", deployment_id = "gpt-35-turbo-deployment" }

# Anthropic models
claude_opus = { model_name = "claude-3-opus-20240229" }
claude_sonnet = { model_name = "claude-3-7-sonnet-latest" }

# Google Gemini models via OpenRouter
gemini_flash_preview = { model_name = "google/gemini-2.5-flash-preview-09-2025", api_base = "https://openrouter.ai/api/v1", api_key="YOUR_OPENROUTER_API_KEY"}

# Jina Embeddings
jina_embed = {model_name = "jina-embeddings-v3", api_base = "https://api.jina.ai/v1", api_key = "YOUR_JINA_API_KEY"}

# Ollama models (modify api_base based on your deployment)
# - Local Ollama: "http://localhost:11434" (default)
# - Morphik in Docker, Ollama local: "http://host.docker.internal:11434"
# - Both in Docker: "http://ollama:11434"
ollama_qwen_vision = { model_name = "ollama_chat/qwen2.5vl:latest", api_base = "http://localhost:11434", vision = true }
ollama_llama_vision = { model_name = "ollama_chat/llama3.2-vision", api_base = "http://localhost:11434", vision = true }
ollama_embedding = { model_name = "ollama/nomic-embed-text", api_base = "http://localhost:11434" }

# Lemonade models (for AMD GPU and NPU support)
# - Local: "http://localhost:8020/api/v1"
# - Morphik in Docker: "http://host.docker.internal:8020/api/v1"
lemonade_qwen = { model_name = "openai/Qwen2.5-VL-7B-Instruct-GGUF", api_base = "http://host.docker.internal:8020/api/v1", vision = true }
lemonade_embedding = { model_name = "openai/nomic-embed-text-v1-GGUF", api_base = "http://host.docker.internal:8020/api/v1" }

openai_embedding = { model_name = "text-embedding-3-small" }
openai_embedding_large = { model_name = "text-embedding-3-large" }
azure_embedding = { model_name = "text-embedding-ada-002", api_base = "YOUR_AZURE_URL_HERE", api_version = "2023-05-15", deployment_id = "embedding-ada-002" }

#### Component configurations ####

[agent]
model = "gemini_flash_preview" # Model for the agent logic

[completion]
model = "gemini_flash_preview"
default_max_tokens = "2048"
default_temperature = 0.3

[database]
provider = "postgres"
# Connection pool settings
pool_size = 10           # Maximum number of connections in the pool
max_overflow = 15        # Maximum number of connections that can be created beyond pool_size
pool_recycle = 3600      # Time in seconds after which a connection is recycled (1 hour)
pool_timeout = 10        # Seconds to wait for a connection from the pool
pool_pre_ping = true     # Check connection viability before using it from the pool
max_retries = 3          # Number of retries for database operations
retry_delay = 1.0        # Initial delay between retries in seconds

[embedding]
model = "jina_embed"  # Reference to registered model
dimensions = 1024
similarity_metric = "cosine"

[parser]
chunk_size = 6000
chunk_overlap = 300
use_unstructured_api = false
use_contextual_chunking = true
contextual_chunking_model = "gemini_flash_preview"  # Reference to a key in registered_models

[parser.xml]
max_tokens = 350
preferred_unit_tags = ["SECTION", "Section", "Article", "clause", "Секция", "Раздел"]
ignore_tags = ["TOC", "INDEX"]

[document_analysis]
model = "gemini_flash_preview"  # Reference to a key in registered_models

[parser.vision]
model = "gemini_flash_preview"  # Reference to a key in registered_models
frame_sample_rate = -1  # Set to -1 to disable frame captioning

[reranker]
use_reranker = false
provider = "flag"
model_name = "BAAI/bge-reranker-large"
query_max_length = 256
passage_max_length = 512
use_fp16 = true
device = "cuda" # use "cpu" if on docker and using a mac, "cuda" if cuda enabled device

[storage]
provider = "local"
storage_path = "./storage"

# [storage]
# provider = "aws-s3"
# region = "us-east-2"
# bucket_name = "morphik-s3-storage"

[vector_store]
provider = "pgvector"

[multivector_store]
provider = "postgres"  # "postgres" or "morphik" for fast implementation

[rules]
model = "gemini_flash_preview"
batch_size = 4096

[morphik]
enable_colpali = true
mode = "self_hosted"  # "cloud" or "self_hosted"
api_domain = "api.morphik.ai"  # API domain for cloud URIs
# Only call the embedding API if colpali_mode is "api"
morphik_embedding_api_domain = "http://colpali-service:8001"  # endpoint for multivector embedding service
colpali_mode = "api" # "off", "local", or "api"

[pdf_viewer]
frontend_url = "http://localhost:3000/api/pdf"

[workflows]
model = "gemini_flash_preview"  # Model for workflow extraction tasks

[graph]
model = "gemini_flash_preview"
enable_entity_resolution = true

# [graph]
# mode="api"
# base_url="https://graph-api.morphik.ai"

[telemetry]
enabled = true
honeycomb_enabled = true
honeycomb_endpoint = "https://api.honeycomb.io"
honeycomb_proxy_endpoint = "https://otel-proxy.onrender.com"
service_name = "databridge-core"
otlp_timeout = 10
otlp_max_retries = 3
otlp_retry_delay = 1
otlp_max_export_batch_size = 512
otlp_schedule_delay_millis = 5000
otlp_max_queue_size = 2048