Dify
Open source and hosted platform for building agentic and RAG apps with a visual flow builder, evaluations, observability and team governance.
Overview
Dify provides an end to end stack for LLM applications without stitching dozens of services. Builders design prompts and multi step agent graphs on a drag and drop canvas, attach retrieval over private knowledge bases, and connect external tools with function calling. Experiments are first class, teams run A B tests, collect traces and user feedback, then compare versions with offline and online metrics.
The open source edition targets self hosting and air gapped environments while Dify Cloud removes infrastructure overhead with SSO, roles and secrets management. Connectors span major model providers and vector stores, making it easy to standardize across OpenAI, Anthropic, Google and open weights. Templates include support assistants, research copilots and workflow bots so product teams can ship quickly, then scale with rate limits, caching and usage analytics.
The goal is repeatability, a single place to design, evaluate and operate apps that answer safely, cite sources and improve over time.
Key features
- Visual builder for prompts tools and agents
- Hosted cloud with SSO roles and secrets
- Evaluation and A B testing with feedback
- Tracing metrics and session replays for QA
- Knowledge base and vector retrieval nodes
- Function calling for external tool use
- Connectors for major LLMs and vector DBs
- Templates for support research and workflow apps
Best for
- Support bot with grounded answers over docs
- Research copilot to summarize sources with cites
- Internal knowledge Q and A for employees
- Ticket triage and suggested replies for CS
- Sales enrichment agent for CRM entries
- Operations automation with tool calling
- Compliance review assistants with audit logs
- POC to production handoff with versioned flows
Capabilities
Visual Flows
Compose prompts retrieval and tool calls on a canvas so reviewers can understand and approve each step before deployment.
Traces and Metrics
Collect spans feedback and outcomes, compare versions and spot regressions to keep quality and safety stable in production.
Knowledge Bases
Attach private documents, chunk and embed with preferred models, then ground answers with citations and filters.
Cloud Teams
Use hosted plans for SSO roles secrets and rate limits, or self host the OSS edition for private environments.
Frequently Asked Questions
What is the entry pricing for the hosted edition?
A free sandbox is available, paid cloud plans commonly start around the mid two digit monthly range for individuals and a higher team tier adds collaboration and controls.
Can it be self hosted for private deployments?
Yes, the open source edition supports full self hosting, useful for regulated data and air gapped networks.
Does Dify work with multiple model providers?
Connectors support leading commercial and open models and you can choose models per step or per environment.
How are experiments and evaluations handled?
Versioning A B tests offline metrics and trace comparisons help teams select prompts graphs and models with evidence.
Is there retrieval over my documents?
You can build knowledge bases, manage chunking and embeddings, and enforce citations and filters at query time.
Can agents call external tools?
Function calling enables integration with APIs and internal systems, results feed back into the conversation state.
What governance features exist?
Hosted plans include roles SSO secrets, audit friendly logs and usage analytics, OSS can be combined with your own controls.
How fast can we move from POC to production?
Templates and managed observability shorten the path, while rate limits caching and alerts help stabilize live workloads.
