Dify

Dify is a platform for building applications around language models. Its visual Workflow and Chatflow editors combine model calls, knowledge retrieval, tools, code, and conditional logic. Workflow suits a bounded request such as extracting fields from a document. Chatflow adds conversational interaction for an assistant.

A practical workflow

For an internal policy assistant, load approved documents, configure retrieval, connect a model, and publish an application. Evaluate questions with known answers before opening access. Include missing information and conflicting documents in the evaluation. A retrieved passage is context for a model, not proof that its answer is correct.

Dify supports cloud hosting and self-deployment. Its repository uses a modified Apache 2.0 license with additional conditions. Operating a multi-tenant environment requires written authorization. Removing or changing frontend logos or copyright notices also raises commercial licensing requirements. Review those conditions before offering a hosted service or a branded derivative.

Free access and paid boundaries

Cloud Sandbox currently includes 200 message credits, one member, five applications, and 50 MB of knowledge storage. Those credits are model-dependent trial capacity, not 200 guaranteed OpenAI calls or a monthly token allowance. Professional and Team raise workspace limits. Your own model API key can incur separate provider fees.

Check document quotas, retrieval request rates, trigger-event allowances, logging, and team permissions against your workload. Self-hosted deployments need storage, model connections, upgrades, and backups. For a production assistant, define who maintains the source documents and how unsupported answers reach a person.

Official sources checked on October 23, 2026. Pricing, Workflow and Chatflow, and license.