Visual workflows
Non-engineer accessible; API-first for developers
AI & MLFrom $29/app/month
Managed Dify Hosting
LLM app development platform
What is Dify on ManageStacks?
Dify on ManageStacks is the open-source LLM application platform deployed to your own AWS, Azure, or GCP region — priced flat at $29 per instance per month regardless of user or workflow count, with visual workflow builder, RAG pipeline, agent tooling, prompt management, and API-first design. Materially cheaper than LangSmith or Vercel AI SDK Cloud for team-based AI app development, and prompts + RAG data stay in your cloud. Pairs with Ollama, Qdrant, and LiteLLM (also on ManageStacks) for a fully self-hosted GenAI stack.
Dify on ManageStacks is the open-source LLM application platform deployed to your own AWS, Azure, or GCP region — priced flat at $29 per instance per month regardless of user or workflow count, with visual workflow builder, RAG pipeline, agent tooling, prompt management, and API-first design. Materially cheaper than LangSmith or Vercel AI SDK Cloud for team-based AI app development, and prompts + RAG data stay in your cloud. Pairs with Ollama, Qdrant, and LiteLLM (also on ManageStacks) for a fully self-hosted GenAI stack.
Last updated Official siteSource on GitHubDocumentation
About Dify
What Dify does, and why teams deploy it.
Dify is an open-source LLM application platform that combines workflow orchestration, RAG pipelines, AI agent capabilities, prompt management, and model routing into one product. It's what you reach for when your team is building 5-50 LLM-powered features and doesn't want to hand-roll LangChain glue for every one.
Dify supports every major LLM provider — OpenAI, Anthropic, Google, Cohere, Azure OpenAI, AWS Bedrock — and every major open-source model via Ollama, vLLM, or Hugging Face TGI. Its visual workflow builder lets non-engineers assemble chat apps, agents, and RAG pipelines; developers get an API-first design with SDKs for Python and Node.
On ManageStacks, Dify pairs naturally with Ollama (local models), Qdrant (vector store), and LiteLLM (unified LLM proxy) — all in the same account, same region, no cross-cloud data flow. Self-hosted end-to-end GenAI without prompts, embeddings, or user queries leaving your infrastructure.
DIY vs ManageStacks
What running Dify yourself looks like — and what it looks like with us.
DIY self-hosting
- Write LangChain glue code for every workflow; version and test manually
- Wire in vector DB, embedding calls, prompt versioning, and cost tracking separately
- Build UI for non-engineers to iterate on prompts + workflows — or don't
- Track LLM API spend across providers with a spreadsheet
- Deploy each workflow separately as a service
On ManageStacks
- Subscribe through your AWS, Azure, or GCP marketplace
- Dify comes up with Postgres, Redis, vector-DB integration, and workflow UI
- Non-engineers build workflows visually; developers consume via REST API
- Pair with Ollama (local LLMs), Qdrant (vector DB), LiteLLM (provider proxy) on same account
- Rolling version upgrades handled by us
Dify on ManageStacks — key numbers
$29/mo
Flat per instance, unlimited workflows + users
Any LLM
OpenAI, Anthropic, Bedrock, Ollama, vLLM — all supported
RAG built-in
Chunking, embedding, retrieval, re-ranking, hybrid search
Key features
Everything Dify ships with, running on our stack.
- Visual AI workflow builder with drag-and-drop nodes
- RAG pipeline with vector database integration (Qdrant, pgvector, more)
- Model routing across OpenAI, Anthropic, Google, Bedrock, Ollama, vLLM
- Agent framework with function calling and tool integration
- Prompt management with versioning and A/B testing
- Dataset management for RAG knowledge bases
- API-first: consume Dify apps as REST endpoints in your product
- Team collaboration with roles and workspace management
- Observability: token usage, latency, and quality metrics per workflow
- SSO integration (SAML, OAuth, LDAP) on Business+ with Keycloak
How it deploys
From subscribe to live in minutes.
Subscribe
Subscribe to ManageStacks through your AWS, Azure, or GCP marketplace.
Provision
Dify spins up with Postgres, Redis, and vector-DB integration — typically 3-5 minutes.
Configure models
Add API keys for OpenAI, Anthropic, or point at your Ollama endpoint on the same account. Configure LiteLLM for unified routing.
Build + deploy apps
Create workflows visually or via YAML. Upload RAG datasets. Consume as REST endpoints from your product.
Who this is for
Built for teams that want Dify to just work.
Product teams shipping AI features
You've got 3-10 LLM-powered features in flight (support-bot, doc-search, summarisation, etc.). Dify gives your team a shared platform to build, version, and observe them.
Enterprises consolidating AI experimentation
Different teams are hand-rolling LangChain apps. Dify becomes the platform layer — one place for prompts, RAG datasets, and cost tracking.
Regulated / air-gapped AI
Prompts + RAG data + user queries can't leave your cloud. Dify + Ollama + Qdrant on ManageStacks is a fully self-hosted GenAI stack.
Compliance & compatibility
What we handle, what Dify runs on.
Compliance & operations
- TLS-encrypted UI + API traffic
- Prompts, datasets, and user queries stay in your cloud region
- Per-workflow rate limiting + cost tracking
- GDPR data-residency — deployment stays in your chosen cloud region
- OS-level and Dify security patches applied during your maintenance window
- SSO integration (SAML, OAuth, LDAP) on Business+ via Keycloak on same account
Compatibility
- Version
- Latest Dify stable (validated before release)
- Runtime
- Python + Node.js on containerised infrastructure
- Dependencies
- PostgreSQL 15, Redis 7, vector DB (Qdrant or pgvector on Postgres)
- Min. resources
- 2 vCPU / 4 GB RAM (dedicated); scale with workflow volume
How ManageStacks helps
We handle the parts you shouldn't be writing yourself.
ManageStacks deploys Dify with Postgres, Redis, and a vector database backend (Qdrant on the same account for the natural pairing). We handle scaling, dataset storage, and version upgrades. Combine with Ollama for local LLMs and LiteLLM for provider routing to build production GenAI apps entirely within your cloud region.
How it compares
Dify on ManageStacks vs the alternatives.
How Dify on ManageStacks compares to the vendor-hosted LLM development platforms and hand-rolled LangChain.
| Property | Dify on ManageStacksUs | LangSmith | Vercel AI SDK Cloud | Hand-rolled LangChain / LlamaIndex |
|---|---|---|---|---|
| Deployment | Managed on your AWS, Azure, or GCP | Vendor-hosted | Vendor-hosted | You deploy + operate |
| Data residency | Your cloud region | Vendor infrastructure | Vendor infrastructure | Your cloud region |
| Pricing basis | Flat per instance | Per team + per trace | Per token routed | Your compute cost |
| UI for non-engineers | Yes (visual builder) | Partial (prompt playground) | No (code-first) | No (code-only) |
| Open source | Yes (Apache 2.0) | No (proprietary) | SDK is open (MIT) | Yes |
| Model choice | Any (OpenAI, Anthropic, Ollama, more) | Any | Any | Any |
Comparison focuses on architectural properties (deployment model, pricing basis, open-source status) that don't change with vendor pricing pages. Verify current pricing on each vendor's own site.
FAQ
Common questions about Dify on ManageStacks.
How does this compare to LangSmith or Vercel AI SDK?
LangSmith is LangChain's hosted observability + prompt-management platform, priced per team + per trace. Vercel AI SDK is a code-first LLM abstraction (not a UI platform). ManageStacks Dify is flat $29 per instance and gives you the full UI platform (workflow builder, RAG, agents, prompt mgmt) — closer to LangSmith + LangGraph combined. LangSmith wins on deep LangChain integration and vendor-managed traces; Dify wins on flat cost, self-hosting, and non-engineer accessibility.
Can I use both cloud LLMs and self-hosted models with Dify?
Yes. Dify supports OpenAI, Anthropic, Google, Azure OpenAI, AWS Bedrock, Cohere, and any Ollama/vLLM/TGI endpoint. Route different steps in a workflow to different models. Pair with LiteLLM (on ManageStacks) for unified auth, rate limiting, and cost tracking across providers.
How does Dify handle RAG?
Upload documents (PDF, DOCX, Markdown, text) or crawl URLs; Dify chunks, embeds, and stores in a vector database (Qdrant on ManageStacks is the natural pairing). At query time, Dify retrieves relevant chunks and passes them to the LLM. Advanced re-ranking, hybrid search (BM25 + vector), and metadata filtering are all supported.
Is Dify safe for production customer-facing AI?
Yes. Dify apps expose stable REST endpoints; your product consumes them like any API. Observability shows token usage, latency, and quality per workflow. Rate limiting, per-workflow quotas, and cost tracking prevent runaway spend on cloud LLM APIs.
Does ManageStacks handle Dify version upgrades?
Yes. Dify ships releases frequently; we test each one against your workflows before rolling forward. Database migrations handled.
What if I want to move off Dify later?
Export workflow definitions as JSON, export datasets as text/documents, export prompts as templates. Rebuild in LangChain, LlamaIndex, or roll your own. Dify is Apache-2.0. Migration off is a supported operation.
Deploy Dify in under 5 minutes.
Subscribe through your AWS, Azure, or GCP marketplace. We handle provisioning, SSL, monitoring, backups, updates, and security. From $29/app/month.