Dify
Open-source LLM app builder: workflows, agents, RAG, one canvas
Overview
Dify bundles a visual workflow canvas, RAG pipeline, agent tooling, 100+ model providers and observability into one open-source platform. Self-host it or use the hosted Sandbox — the API is exposed for every app you build.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 11 Jul 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- 1 workspace
- 3 team members
- 5 apps
- 50 knowledge docs
- 5,000 msg credits/mo
- 50 apps
- 500 knowledge docs
- 5GB storage
- 10,000 msg credits/mo
- 200 apps
- 1,000 knowledge docs
- 20GB storage
- Multiple workspaces
- SSO
- Commercial license
- Dedicated support
What you can produce with Dify
- Build a production RAG knowledge base that ingests internal documents (PDFs, Notion pages, web URLs) and exposes a chatbot interface, using Dify's configurable chunking, embedding, and hybrid retrieval settings — entirely without writing retrieval code.
- Compose a multi-step agentic workflow on the visual canvas that calls external APIs, branches on LLM-evaluated conditions, and escalates to a human reviewer via the Human Input node before final output.
- Deploy a self-hosted AI application on your own infrastructure (Docker Compose or Kubernetes) with full control over data residency and no per-message cloud billing, switching underlying LLM providers (OpenAI, Anthropic, local Llama) purely through config.
- Create and publish a customer-facing chatbot with a branded embed widget, complete with conversation history, system prompt guardrails, and model fallback rules — ready for a production domain within hours.
- Wire together a Supervisor multi-agent system where a coordinator agent delegates subtasks (web search, code execution, document retrieval) to specialist sub-agents and aggregates their outputs into a final response.
- Prototype and iterate on an LLM application using Dify's prompt IDE with A/B testing across model versions, then inspect per-node token usage and latency traces to identify bottlenecks before scaling.
- Extend Dify's capabilities by installing plugins from the marketplace (custom tools, retrieval strategies, or model connectors) or publishing your own plugin — made available via the plugin ecosystem launched in 2025.
ASEAN Perspective
Dify in Southeast Asia
ASEAN-region availability and pricing notes coming soon. Drop the editorial team a note via /contact/ if you can supply local context (Singapore/Malaysia/Indonesia/Thailand/Vietnam).
Dify's pitch has held up: a visual canvas for workflows, agents and RAG, backed by a fully open codebase you can self-host, with LangGenius's own $30M Series Pre-A in March 2026 proof the model works commercially too. The API-first design means anything built visually is instantly usable as a real endpoint, which is why it keeps coming up as the default recommendation in open-source LLM-tooling threads.
Self-hosting for real scale takes actual DevOps effort — this isn't a managed SaaS pretending otherwise — and the free Sandbox's 200 one-time credits barely cover a test run. Complex branching logic can outgrow the visual canvas, pushing teams toward code. Still, for anyone who wants to own their data and avoid platform lock-in, it's the strongest default in the category.
What people say
146,000 GitHub stars and counting — Dify's repo has become the reference implementation for what an open-source LLM app platform should look like. LangGenius, the company behind it, closed a $30 million Series Pre-A in March 2026 at a reported $180 million valuation (HSG led, with GL Ventures and others), and its own figures claim 1.4 million self-hosted deployments across 175+ countries plus enterprise users like Maersk and Novartis running it in production.
Practitioners rate the RAG pipeline and the visual workflow debugger highly — stepping through a node, inspecting its input and output, and seeing exactly where a chain broke beats guessing from stack traces. That debuggability is the thing reviewers mention most when comparing Dify to code-first frameworks.
The complaints are consistent too. Large canvases with heavy branching get unwieldy fast, and teams doing anything genuinely complex often end up migrating logic to code. Error messages inside failed nodes are frequently vague — "node execution failed" with nothing else — which drags out debugging. Custom code nodes run in a locked-down sandbox that blocks a lot of standard libraries, forcing API-call workarounds. And the free Sandbox tier's 200 credits are one-time, not monthly, with the jump to the $59/month Professional plan feeling steep for hobbyists. Third-party tutorials and plugins still lag the more established frameworks, which shows in how often Dify users end up reading the source instead of the docs.
Summary of public user & expert reviews, compiled by RECATOOLS.
About this listing
This entry was compiled from publicly available data including Dify's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Dify unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Dify directly →
Spotted something out of date? Suggest an update →
Dify in the news
Alternatives to Dify
More in Agents & Automation
