​
Morphik Configuration Guide

Morphik’s behavior can be fully customized through the morphik.toml configuration file. This file is the central place to configure all aspects of the system, from API settings to model providers, database connections, and processing options.

​
Configuration Basics

The morphik.toml file uses the TOML format (Tom’s Obvious Minimal Language) to organize settings into logical sections. Each section controls a specific component of the Morphik system:

[section_name]
setting1 = "value1"
setting2 = 123

When you first set up Morphik, you can:

  • Run python quick_setup.py for an interactive setup process
  • Copy and modify the example morphik.toml file
  • Create your own configuration file manually

​
Registered Models Approach

Morphik uses a registered models approach which allows you to define hundreds of different models and reference them throughout your configuration. This makes it easy to:

  • Use different models for different tasks
  • Mix and match models based on your needs (e.g. smaller models for simpler tasks)
  • Configure model-specific settings in one place
  • Switch between providers without changing the rest of your configuration

​
Defining Registered Models

First, register your models in the [registered_models] section:

[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" }

# 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" }

# Embedding models
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" }

​
Using Registered Models

Then reference these models by their key in different sections:

[completion]
model = "ollama_qwen_vision"  # Reference to a key in registered_models
default_max_tokens = "1000"
default_temperature = 0.5

[embedding]
model = "ollama_embedding"  # Reference to registered model
dimensions = 768
similarity_metric = "cosine"

[parser.vision]
model = "ollama_qwen_vision"  # Reference to a key in registered_models

[rules]
model = "ollama_qwen_vision"

[graph]
model = "ollama_qwen_vision"

This approach gives you maximum flexibility to use different models for different components, and even switch between local and cloud models as needed.

​
Ollama Deployment Configurations

When using Ollama models, you need to configure the api_base parameter based on your deployment setup:

​
Local Development

If both Morphik and Ollama are running locally:

ollama_qwen_vision = { model_name = "ollama_chat/qwen2.5vl:latest", api_base = "http://localhost:11434", vision = true }
ollama_embedding = { model_name = "ollama/nomic-embed-text", api_base = "http://localhost:11434" }

​
Morphik in Docker, Ollama Local

If you’re running Morphik in Docker but Ollama locally on your machine:

ollama_qwen_vision = { model_name = "ollama_chat/qwen2.5vl:latest", api_base = "http://host.docker.internal:11434", vision = true }
ollama_embedding = { model_name = "ollama/nomic-embed-text", api_base = "http://host.docker.internal:11434" }

​
Both in Docker

If both Morphik and Ollama are running in Docker containers:

ollama_qwen_vision = { model_name = "ollama_chat/qwen2.5vl:latest", api_base = "http://ollama:11434", vision = true }
ollama_embedding = { model_name = "ollama/nomic-embed-text", api_base = "http://ollama:11434" }

Simply modify the api_base parameter in your existing model definitions rather than creating separate model entries for each deployment scenario.

​
Core Configuration Sections

​
API Server Settings

Controls how the API server runs:

[api]
host = "localhost"  # Use "0.0.0.0" for Docker
port = 8000
reload = true       # Hot reload for development

​
Authentication

Controls user authentication and permissions:

[auth]
jwt_algorithm = "HS256"
dev_mode = true  # Enables simplified auth for local development
dev_entity_id = "dev_user"
dev_entity_type = "developer"
dev_permissions = ["read", "write", "admin"]

​
LLM Completion Settings

Configure the model used for generating text:

[completion]
model = "ollama_qwen_vision"  # Reference to a key in registered_models
default_max_tokens = "1000"
default_temperature = 0.5

​
Database Connection

Choose your document metadata database:

[database]
provider = "postgres"

# Or for MongoDB
[database]
provider = "mongodb"
database_name = "morphik"
collection_name = "documents"

​
Embedding Configuration

Configure vector embeddings generation:

[embedding]
model = "ollama_embedding"  # Reference to registered model
dimensions = 768
similarity_metric = "cosine"

​
Document Parsing

Control how documents are processed and chunked:

[parser]
chunk_size = 1000        # Characters per chunk
chunk_overlap = 200      # Overlap between chunks
use_unstructured_api = false
use_contextual_chunking = false
contextual_chunking_model = "ollama_qwen_vision"  # Reference to a key in registered_models

​
Vision Processing

Configure multimodal processing for images and videos:

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

​
Reranking

Configure the cross-encoder reranker for improved retrieval:

# Enable reranking
[reranker]
use_reranker = true
provider = "flag"
model_name = "BAAI/bge-reranker-large"
query_max_length = 256
passage_max_length = 512
use_fp16 = true
device = "mps"  # "cpu", "cuda", or "mps"

# Disable reranking
[reranker]
use_reranker = false

​
Document Storage

Choose where to store your document files:

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

# AWS S3 storage
[storage]
provider = "aws-s3"
region = "us-east-2"
bucket_name = "morphik-storage"

​
Vector Database

Configure where vector embeddings are stored:

# PostgreSQL with pgvector
[vector_store]
provider = "pgvector"

# MongoDB Atlas
[vector_store]
provider = "mongodb"
database_name = "morphik"
collection_name = "document_chunks"

​
Rules Engine

Configure document processing with rules:

[rules]
model = "ollama_qwen_vision"  # Reference to a key in registered_models
batch_size = 4096

​
Knowledge Graph

Configure entity extraction and graph building:

[graph]
model = "ollama_qwen_vision"  # Reference to a key in registered_models
enable_entity_resolution = true

​
General Morphik Settings

Control core platform features:

[morphik]
enable_colpali = true    # Enable advanced retrieval
mode = "self_hosted"     # "cloud" or "self_hosted"

​
Telemetry

Configure observability settings:

[telemetry]
enabled = true
honeycomb_enabled = true
honeycomb_endpoint = "https://api.honeycomb.io"
honeycomb_proxy_endpoint = "https://otel-proxy.onrender.com"
service_name = "morphik-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

​
Environment Variables

Morphik uses environment variables for sensitive credentials and configuration that shouldn’t be stored in the configuration file:

  • OPENAI_API_KEY: Your OpenAI API key
  • ANTHROPIC_API_KEY: Your Anthropic API key
  • JWT_SECRET_KEY: Secret for authentication tokens
  • POSTGRES_URI: PostgreSQL connection string
  • MONGODB_URI: MongoDB connection string
  • AWS_ACCESS_KEY and AWS_SECRET_ACCESS_KEY: For S3 storage
  • ASSEMBLYAI_API_KEY: For video transcription
  • UNSTRUCTURED_API_KEY: For enhanced document parsing (optional)
  • HONEYCOMB_API_KEY: For telemetry (optional)

​
Complete Configuration Example

Here’s a complete example showing the structure of a morphik.toml file with registered models:

[api]
host = "localhost"  # Use "0.0.0.0" for Docker
port = 8000
reload = true

[auth]
jwt_algorithm = "HS256"
dev_mode = true
dev_entity_id = "dev_user"
dev_entity_type = "developer"
dev_permissions = ["read", "write", "admin"]

[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" }

# 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" }

# Embedding models
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" }

[completion]
model = "ollama_qwen_vision"  # Reference to a key in registered_models
default_max_tokens = "1000"
default_temperature = 0.5

[database]
provider = "postgres"  # or "mongodb"

[embedding]
model = "ollama_embedding"  # Reference to registered model
dimensions = 768
similarity_metric = "cosine"

[parser]
chunk_size = 1000
chunk_overlap = 200
use_unstructured_api = false
use_contextual_chunking = false
contextual_chunking_model = "ollama_qwen_vision"  # Reference to a key in registered_models

[parser.vision]
model = "ollama_qwen_vision"  # Reference to a key in registered_models
frame_sample_rate = -1

[reranker]
use_reranker = true
provider = "flag"
model_name = "BAAI/bge-reranker-large"
query_max_length = 256
passage_max_length = 512
use_fp16 = true
device = "mps"  # "cpu", "cuda", or "mps"

[storage]
provider = "local"  # or "aws-s3"
storage_path = "./storage"

[vector_store]
provider = "pgvector"  # or "mongodb"

[rules]
model = "ollama_qwen_vision"
batch_size = 4096

[morphik]
enable_colpali = true
mode = "self_hosted"  # "cloud" or "self_hosted"

[graph]
model = "ollama_qwen_vision"
enable_entity_resolution = true

[telemetry]
enabled = true
honeycomb_enabled = true
honeycomb_endpoint = "https://api.honeycomb.io"
honeycomb_proxy_endpoint = "https://otel-proxy.onrender.com"
service_name = "morphik-core"

​
Mixing and Matching Models

With the registered models approach, you can easily optimize your configuration:

  • Use powerful models like Claude Opus for knowledge graph generation
  • Use smaller, faster models for simple parsing tasks
  • Use OpenAI models for some tasks and Anthropic models for others
  • Use Ollama for local development and cloud models for production

For example:

[parser.vision]
model = "ollama_llama_vision"  # Use Llama vision model for image analysis

[rules]
model = "claude_sonnet"  # Use Claude Sonnet for rules processing

[completion]
model = "ollama_qwen_vision"  # Use Qwen vision model for general completion

​
Docker Configuration Adjustments

When using Docker, make these changes:

  1. Set host = "0.0.0.0" in the [api] section
  2. For Ollama integration, use models with api_base set to "http://ollama:11434"
  3. Use container names in your database connection strings

For more details on Docker deployment, check out the DOCKER.md file in the repository.

​
Need Help?

If you’re having trouble with your configuration:

  1. Check the logs for error messages
  2. Run python quick_setup.py for guided setup
  3. Join our Discord community
  4. Open an issue on our GitHub repository