// I want to solve · Enterprise IT / Small teams & business
🏢 Private LLMs and knowledge bases for the enterprise
Deploy chat, knowledge-base and agent platforms on your intranet or private cloud: choose models and inference engines, size hardware, set permissions and audit, and check compliance in China.
6 steps5 alternatives5 stacks · 2 problemschecked 2026-10-22
How to do it
- 01Classify data first: what may use public cloud APIs under an enterprise no-training agreement, and what must stay private
- 02Shortlist 2–3 models with our model picker and open-weight lists, then test them on your own tasks
- 03Size GPUs with our VRAM calculator and compute guides; use vLLM or SGLang in production, Ollama for pilots
- 04On top: Open WebUI / LibreChat for chat, Dify / RAGFlow / FastGPT / MaxKB for knowledge bases and apps, wired to company SSO
- 05Route all calls through LiteLLM with departmental budgets, keep audit logs in Langfuse, and test prompt injection and access control before launch
- 06Public-facing generative AI services in mainland China need filing/registration; internal-only use still needs a compliance review
What it costs
Pilot
One workstation with 24–48 GB VRAM plus open-source software is enough to validate (see self-hosted compute for models)
Department
Multi-GPU servers or hourly cloud GPUs; software is open source, operations staff are the main cost
Company-wide
Needs an HA cluster, gateway and audit; compare total cost with public-cloud enterprise plans before deciding
Prices change; the vendor’s official page wins. Subscriptions compared · API prices
Alternatives: which to pick
| Option | Choose it when |
|---|---|
| Dify | Build knowledge Q&A and AI apps visually |
| RAGFlow | Complex documents (tables, scans) where parsing quality matters |
| FastGPT / MaxKB | Active Chinese communities with many local deployments |
| Open WebUI / LibreChat | You only need one internal chat front door |
| 公有云企业版(ChatGPT Enterprise、Gemini for Workspace 等) | Data may leave your network and you want no operations |
✓ Good fit
- Enterprises with sensitive data, intranet requirements, audit and chargeback
✗ Not a good fit
- Small teams without ops staff: start with a public-cloud enterprise plan
Ready-made stacks
All stacksSoftware that helps
All softwareProblems & solutions
All problems- How-toSelf-host Dify on company servers (Docker Compose)Dify can be deployed on your own servers with Docker Compose and connected to private models or cloud APIs to build knowledge-base Q&A and internal AI apps.
- How-toEmployees paste company data into AI: what to doBlanket bans rarely work; people switch to personal accounts. It works better to provide an approved route (enterprise plan or private deployment) plus data classification and auditing.
Needs in this scenario
-
ENT-02Enterprise admission: certifications, data residency, SSO/SCIM, SLA, exit terms
Enterprise IT · People search for: is x soc 2 compliant; self-hosted llm observability; data residency
-
ENT-04Transparent TCO and residency matrix
Enterprise IT · People search for: llm observability pricing; ai gateway tco
-
ENT-05China compliance: filing, domestic IT stacks, bundled tenders
Enterprise IT · People search for: china generative ai filing; ai regulation china
-
ENT-01Choosing the control plane: gateways, observability, security, identity
Enterprise IT · People search for: enterprise llm gateway comparison; ai observability platform
Official references
Compiled from official documentation on 2026-10-22; not reproduced in our own environment. Follow the official docs if they differ.