Dify + Ertas

Build, deploy, and manage AI applications on Dify using Ertas-trained models for domain-specific intelligence with full visual orchestration.

Overview

Dify is an open-source LLM application development platform that combines a visual workflow editor, prompt IDE, RAG pipeline builder, and agent framework into a single cohesive tool. Unlike simple chatbot builders, Dify provides the full stack for taking an AI application from prototype to production — including API management, usage analytics, annotation and labeling tools, and multi-tenant access control. It supports both cloud-hosted and self-hosted deployments, making it suitable for teams with strict data residency requirements.

Dify's visual orchestration engine lets teams build complex AI workflows by connecting model nodes, knowledge retrieval nodes, conditional logic, and external tool calls on a canvas. Each workflow can be published as a standalone web app, embedded chatbot, or REST API with built-in rate limiting and authentication. The platform also includes a prompt IDE with variable management and version history, which makes iterating on prompt engineering significantly faster than editing code files in a traditional development workflow.

How Ertas Integrates

Ertas-trained models integrate with Dify through its custom model provider interface. After fine-tuning in Ertas Studio, you deploy your model to an OpenAI-compatible endpoint and register it as a custom model provider in Dify's settings. Once registered, the model appears in Dify's model selector across all features — workflow nodes, chat assistants, completion APIs, and agent configurations. This means your fine-tuned model can power any Dify application type without additional configuration.

The Ertas-Dify combination is especially valuable for teams building internal AI tools. Ertas provides the specialized model that understands your company's terminology, processes, and data formats, while Dify provides the application layer — complete with user management, conversation history, feedback collection, and analytics. Teams can deploy a customer-facing knowledge base, an internal document search tool, and a data extraction pipeline all running on the same Ertas-trained model, managed from a single Dify dashboard. When it is time to improve the model, feedback collected through Dify's annotation tools can be exported as training data and fed back into Ertas Studio.

Getting Started

1
Fine-tune your model in Ertas Studio
Train a domain-specific model using Ertas Studio with your JSONL datasets. Select the appropriate base model and quantization format for your deployment target.
2
Deploy the model to an inference endpoint
Serve your GGUF model through Ollama, vLLM, or Ertas Cloud. Ensure the endpoint exposes an OpenAI-compatible API for Dify integration.
3
Register the model in Dify
In Dify's Settings, add a custom model provider pointing to your inference endpoint. Configure the model name, context window, and capabilities.
4
Build your application in Dify
Create a chatbot, workflow, or agent application in Dify. Select your Ertas-trained model as the LLM provider and configure knowledge bases for RAG.
5
Publish and collect feedback
Deploy your Dify application as a web app or API. Use Dify's built-in annotation tools to collect user feedback for future Ertas retraining cycles.

yaml

# Dify custom model provider configuration
model_providers:
  - provider: ertas-local
    label: "Ertas Fine-Tuned Models"
    provider_type: openai_api_compatible
    credentials:
      api_base: "http://localhost:11434/v1"
      api_key: "not-needed"
    models:
      - model: ertas-support-7b
        model_type: llm
        context_length: 8192
        max_tokens: 2048
      - model: ertas-legal-7b
        model_type: llm
        context_length: 8192
        max_tokens: 2048

Benefits

Full application platform with user management, analytics, and API access control
Visual workflow editor for building complex AI pipelines without code
Built-in RAG pipeline with knowledge base management and chunking controls
Self-hosted deployment option keeps all data within your infrastructure
Annotation and feedback tools create a direct loop back to Ertas Studio for retraining
Multi-tenant support lets you serve different teams with different model configurations