Dify
Open-source platform for agentic workflow development combining Backend-as-a-Service and LLMOps with a visual canvas and 50+ tools.
Delv Safety Grade: B
Score 72/100 · assessed 2026-04-18
Dify is an open-source no-code platform for building AI agents and workflows, maintained by LangGenius with strong community engagement (38k+ GitHub stars). The platform offers extensive capabilities including RAG, workflow orchestration, and integration with 50+ tools. Transparency is excellent with active development, comprehensive documentation, and open issue tracking. Supply chain is solid via Docker and standard deployment methods. However, the permissions footprint is substantial: as a self-hosted platform, it requires database access, filesystem operations, network connectivity, environment variable access for API keys, and can execute arbitrary LLM calls to external services. The broad scope of capabilities (messaging, payments, repo access through integrations) creates a large attack surface. No known security incidents, but the platform's power means careful configuration and secret management are essential.
Green flags
- Fully open source with 38k+ stars and active community
- Comprehensive documentation and security guidelines provided
- Standard Docker deployment with clear configuration
- Active maintenance with frequent releases and issue resolution
- No known security incidents or CVEs
Red flags
- Requires access to sensitive API keys and credentials for 50+ integrations
- Self-hosted deployment means full filesystem and database access
- Can execute arbitrary code through workflow nodes and tool integrations
- Broad integration scope increases attack surface significantly
Permissions requested
Pricing
Platforms
Review
Pay for the cloud tier if you're prototyping customer-facing agents and want RAG without infrastructure headaches. Self-host the open-source version if you have DevOps capacity and need full control. Skip it if you need true open-ended autonomy or already have a LangChain pipeline you're happy with.
Good at
- Visual workflow editor genuinely speeds up prototyping vs code-first frameworks
- Built-in RAG pipeline with document chunking and vector search
- LLMOps features (versioning, A/B testing, analytics) included, not bolted on
- 50+ tool integrations cover most common use cases without custom code
- Self-hosting option available for teams that need it
Watch out
- Workflow editor becomes unwieldy on complex multi-branch agents
- Debugging is primitive, mostly log inspection and manual tracing
- Free tier rate limits hit fast under production-like load
- Autonomy is workflow-scripted, not open-ended planning like AutoGPT
- Cloud version creates vendor lock-in despite open-source availability
Use cases
- chat apps
- RAG pipelines
- agent workflows