# Jan

## What

Jan is a distributed computing project to develop an Open Superintelligence.

We see Superintelligence as a moving target. As of 2025, we define this as AI's ability to complete [economically valuable tasks at a human-level](https://openai.com/index/gdpval/), as measured by [GDPval](https://openai.com/index/gdpval/).

Unsure: Jan deprioritizes consumer AI (e.g. roleplay) to focus on economically valuable domains in science and engineering.

_Fig 1: Impact of Industrial Revolution on Humanity_

## How?

_TODO: replace_

Jan is inspired by precursor projects:
- [Minecraft](https://www.minecraft.net/en-us) servers and ["mod servers" culture](https://www.curseforge.com/servers/minecraft)
- [Folding@Home](https://en.wikipedia.org/wiki/Folding@home#Patterns_of_participation)
- [SETI@Home](https://en.wikipedia.org/wiki/SETI@home)

Jan is a pay-per-use superintelligence run by a heterogeneous network of sovereign AI nodes:
- Available via app or API
- Geographically diverse
- Sovereign

Jan relies on nodes that run Jan:
- We make it super easy to run a Jan node, offering it as a desktop client with 0 setup
- Jan also has a more powerful datacenter node that is scalable
- Heteregenous network that has different capabilities, availability

Jan routes requests to the optimal available node in the network:
- Nodes compete to mod Jan to offer the smartest intelligence at every price point
- Economic incentives to improve and train superintelligence
- Nodes can be closed source
- Intelligence/watt vs. willingness to pay?

Jan continuously improves through the generation of open datasets:(actually lets not have this here)(maybe add it below somewhere)
- TBD

## Why?

At a societal level, it is desirable for humanity to...

Jan creates opportunities for more people to be involved in AI's development...

Jan provides economic incentives for distributed development of AI, by defining a reward function...

# Appendix

- 2000 AI infra operators that don't exist yet due to high barrier of entry to getting demand
- 2000 AI labs that don't exist yet due to lack of economic incentive/jobs at big labs
- Pooling together compute, models, from sovereign AI nodes

## Problem

Cutting-edge AI is locked in capital-heavy clouds controlling data, compute, and economics. Most labs can't reach AGI due to unaffordable infrastructure and lack of control over data and values. Thousands of capable researchers lack production-grade compute.

AI requires enormous compute & data, which is expensive and limited, giving users little control & profit over the data & AI systems they help create. Only a handful of well-funded corporations possess the infrastructure, data, and scale to advance state-of-the-art AI.

As we approach AGI, those controlling its development not only reap outsized economic benefits, but also shape the future in ways that may not reflect the interests, values, or needs of the wider population.

## Alt Description

Jan started as a local AI application to run foundation models on consumer hardware, and has been downloaded 4+ million times. We recently released a Jan v1 model, an early model prototype capable of agentic, long-horizon tasks on consumer hardware.

## Resources
- [Nguyen's concerns](https://hackmd.io/93nZxo5-SpuXi_kY0dw2xQ)

## Our Solution

- Jan Node: Open-source, local or self-hosted (Desktop Light or Server Full).
- Jan Platform: Routes requests, handles payments, ensures transparency.


Superintelligence via agentic AI, with unified APIs and optimized routing.

- **Separation of Concerns**: The inference engine handles model execution and computation, while the node API layer manages external interfaces, routing, and payments. The inference stack passes completed inferences to the API layer for distribution and billing, ensuring modularity and scalability.


# How Routing Takes Place

Routing optimizes requests to nodes based on scoring for performance, cost, and reliability.


## How People Participate

**Node operators**: Install CLI/app, declare capabilities, earn Jan Cash for uptime and performance.

**Developers & researchers**: Launch agents, fine-tune models, ship MCP integrations on shared compute.

**Infrastructure providers**: Sell GPUs/CPUs into the marketplace with demand-based pricing.
**End users**: Browse agents, switch providers seamlessly, contribute evaluations.

## Economics & Value Creation

**Value flow**

- Usage metered in compute-seconds with tariffs and multipliers.
- Clients pay nodes minus 10-20% platform fee.
- Multipliers and leaderboards reward top nodes.


**Stakeholder benefits**
- **Node operators**: Transparent earnings tied to performance.
- **Developers**: Global compute without running infrastructure.
- **Users**: Competitive pricing, privacy, and optionality.
- **Platform**: Sustainable revenue through network volume fees.

# Daniel's Explanation

## Our Mission

Jan is a:
1. Distributed computing project, aimed to
2. Incentivize a global community of independent AI researchers and engineers
3. To develop an Open Superintelligence

## Our Product(s)

- A "Jan node" is a open superintelligence that you can run/self-host
- Jan is a full open-source alternative to OpenAI
- Run locally or self-hosted for sovereignty/privacy
- Jan Network is a distributed network of Jan nodes,
- Incoming requests to the "smartest" node
- Pay-as-you-go

## Demand

- Switchability of major players (e.g. Google, ChatGPT, etc)
- DeepResearch or Agent Mode across the different players
- Private Memory/Knowledgebases that are vendor agnostic
- Stored at the edge, retrieved only to send into AI

## Supply

Core problem: There are about 2,000 AI labs that currently don't exist yet.

Solution

### Node Types

## Key Concerns

### Privacy

- Nodes' privacy policies are up to them

### Architecture

- Jan Nodes
- Jan Light Node (i.e. Desktop)
- Jan Full Node (i.e. Server)
- Can be run independently & privately, as an open superintelligence
- Similar to Minecraft servers, can be modded and improved
- Jan Network
- Nodes can optionally be connected to the Jan Network
- Jan Coordinator monitors nodes
- Similar to Folding@Home
- Centralized at first, can explore decentralized later on

## The problem

## Mission
Jan is a distributed computing project aimed to help a global community of independent AI researchers and engineers develop an Open Superintelligence.

## Goals
1. Let more people enter the AI game and compete
2. AI is not a digital god that only Elon/Sam/Mark can build
3. AI is a tool that can be wielded by the individual

## How people participate
(concretely land the idea)

### Nodes
- developers can launch agents
- companies can provide mcps
- researchers can train models
- infrastructure providers can provide compute
all served through a /completions endpoint

### Users
- get access to 10000 agents, niche things
- indirectly provide evails/verifiers, to improve the product.