# Jan Server

This is a list of Problem Statements that we will be working towards

Long-term, Jan Server is competing with:
- "Self-Hosted AI"
- ChatGPT Enterprise
- Cohere, Mistral etc
- OpenWebUI, AnythingLLM Server
- LLM Gateway/Router
- LiteLLM, Portkey, etc

## User 1: Jan Cloud

### Jan's Cloud

- Jan Cloud version: https://ask.jan.ai
- Jan Desktop can connect to Jan Cloud

### Jan Cloud is open source and self-hostable

- Jan Server powers Jan Cloud
- Apache 2.0 licensed outside of `/ee` folder

- /ee licensed
    - Billing

### Jan Cloud to collect Model Training data

- LMArena-style comparison of 2 models or agents
- RLHF (vote up/down)
- Comparison (similar to Yupp or LMArena)

Note: Code for Data Collection should be Enterprise Licensed (i.e. `/ee`)

### Jan Cloud support Teams

- Jan for Teams
    - Signing up allows to create subdomain
    - URL: https://menlo.jan.ai
    - SaaS
- Easier than self-hosting

Note: Code for Teams should be Enterprise Licensed (i.e. `/ee`)

### Jan Cloud to do billing

- Dan's note: Enterprise Licensed feature

- Consumer subscription plan
- Pay-as-you-go
    - Similar to OpenRouter

Note: Billing Code should be Enterprise Licensed (i.e. `/ee`)

### Jan Cloud to offer MCP-as-a-Service

- Offer MCP Servers as APIs
- Jan Cloud run (e.g. we self-host Firecrawl)
- Possible: route to existing MCP Providers
- "OpenRouter for MCPs?"
- e.g. Exa (passthrough billing)

- Run popular MCP servers as-a-Service
    - e.g. Firecrawl or Opencrawl
    - e.g. Browser Use
    - e.g. Search
- Route to popular MCP services
    - e.g. Exa
- Focus on powering https://ask.jan.ai first

## User 2: Enterprise Self-Host

- Primary: AWS/GCP/Azure/Oracle
- Secondary: On-Premise

### "I want to give my staff ChatGPT"

- 50 person org that wants to share access to Gemini, GPT-5, and Local Models
    - GPT-5
    - Local Kimi K2
- Provision accounts for your staff, but not pay per-seat licensing
    - ChatGPT for Teams: $20/head
    - Claude for Teams
    - Typing Mind
- Dogfood this internally within Menlo
- Enterprise-friendly Features
    - Audit trails
    - PII Detection
- Smart Routing
    - If PII detected, route to local model

Customer: Kuok Group

### "I want to deploy my own custom Agent"

- Startup might choose to deploy their model on Jan Server
    - Local or Remote model
    - Custom MCP or Tools
    - Collect Evals
- Similar to how people use OpenWebUI

- Dan's instinct: the Agent composes on top of Jan Server APIs
- Jan Server provides Models and MCP Servers
- i.e. a Repo
- Dify-style Visual Designers have not worked well
- Agents move to code, after the MVP
- n8n etc are great, but fragile
- Better for Jan Server to complement n8n, vs. compete
- MCP, Models and Evals

Customer: Kuok Group

### "I want a scalable Ollama"

- Basic MLOps
    - vLLM based (vs. llama.cpp)
- Model Management
    - `/models/pull` (pull from HF)
    - `/models/delete`
    - `/models/load`
 - Autoscaling of models on GPUs
 - "Serverless"-style scaling

Customer: Kuok Group