# Jan Todo

_Mockup_

## 1-liner

> Explain your product, without using the words "AI" or "Agent".

Jan is a Todo app that works on tasks for you.

## Overview

Jan is raising a $Xm seed via SAFE for product development to wrap Jan's AI Agent research into a consumer-focused agentic Todo app.

Jan has trained two popular open-source agentic models with >100k downloads, and also built one of the more popular local AI apps that have >4 million downloads.

We see a fundamental gap between autonomous AI Agents and mainstream users. We believe a simple todo list is the correct cross-the-chasm interface.

We are running an early-access pilot with power users, with a goal of producing >70% task completion automation and NPS in the high 60s.

## Problem

My todo list is packed with small errands and micro-tasks that theoretically can be done with modern AI agents.

However, there is a friction between AI and existing productivity tools, which requires me to take initiative, provide agency and persistence as a coordinator:
- Stop procastinating, open my todo app, xt on task
- Prompt ChatGPT, paste context, come up with a plan
- Take action, by hopping between email, phone, messenger and browser tabs

## Solution & Product

Jan is an agentic Todo app, that wraps Jan's proprietary AI agent(s) and models.

We believe the best interface for AI agents is a simple todo list:
- You add a task
- An agent takes ownership of the task
- The agent has its own phone, browser and can take actions
- It runs asynchronously in the background, asks clarifying questions, and only checks in for approvals on irreversible actions
- The agent can access your systems (e.g. calendar) with your approval

A few examples of tasks Jan can do:

- "Call DBS bank to waive my credit card fee"
- "Fix callus in foot by finding doctor with earliest appointment"
- "Plan trip to Lisbon to pick up visa on 15th march 2026"

Jan's agentic Todo app are powered by frontier models and Jan v2, our own agentic task planning model. Jan has a Bitwarden-style open-source approach, with a Jan Server that can be self-hosted.

Long-term, Jan's agentic todo app allows us to develop a best-of-class [Jan Agentic model](https://x.com/arvindh__a/status/1991090271400964458), as our app UX allows us to continually generate a human-guided task execution dataset to train a better model[^1]. We see this as Jan's eventual long-term product with use-cases that expand far beyond simple todos.

## Market Opportunity

We estimate the global market for getting personal and administrative errands done (i.e. tasks that people would pay to offload via a todo list) is in the range of $30b annually.

- 50-100m people who pay for todo apps
- Assume people value 2h/month of time at $50
- Assume people would be willing to trade money for time if friction drops (i.e. which Jan as an agentic todo app would do)

Todo apps are already a $10-20b TAM for consumer and prosumer, driven by 200-300m "digitally organized" knowledge workers who plausibly pay $5-7 for a todo app.

The bigger, scary market opportunity is if Jan's models improve to be able to take on more difficult, long-horizon tasks. This expands the market to $200-400b, where we would compete with frontier AI labs.

## Why Now?

AI Agents are now capable of solving errand-level tasks at an acceptable success rates, cost-effective price point:
- Tasks can now be completed for pennies
- LLM pricing is now cheap enough to let AI run autonomously
- e.g. GPT-5.1 Nano: $0.05/1M input, $0.40/1M output

This means for most mass affluent consumers, it's possible to trade money for time by paying a few cents to offload a task from your todo list.

From a technology maturity perspective, AI Agents, Task Planning models, Voice Agents, and MCP Connectors ecosystems are sufficiently performant for production.

## Business Model

Jan costs $10/month, with pay-as-you-go Usage:

- $10 credit for AI usage
- Pay-per-use for AI models, with a 20% markup
- Pay-per-use for tools (e.g. Browser, Computer, Telephony, Search)

## Traction & Metrics

> TODO: fill in with real numbers after pilot

- Traction
- Y DAU
- X waitlist signups
- Y active weekly pilot users completing ~N tasks each.
- Agent Success Rate
- A% of submitted tasks successfully completed
- B% fully automously completed
- C% partially autonomously completed (requires human-in-the-loop)
- Connectors live:
- Gmail, Outlook, Notion, Google Drive, Slack
- 15 additional connectors in development.

## Go-to-Market Strategy

Jan already has a large community and following on Discord, Twitter, and r/localllama.
- Existing app has 4 million downloads
- 10k Discord community
- Twitter

Our primary motion is product-led self-serve onboarding, a $10/mth entry point, and positive NPS from individual users who earn extra credit via inviting others to Jan.

We later layer on marketing-led growth (influencers, content, community) to broaden top-of-funnel once core product loop is working.

## Competitive Landscape

Existing todo apps like Todoist, Ticktick, Microsoft Todo:
- A few of them have begun to add AI copilots
- However, none take end-to-end ownership of tasks
- Agentic AI capabilities are non-trivial to build, and even tougher to achieve high task completion success rates.

General purpose AI agent platforms like ChatGPT Agent Mode, Perplexity, Manus and Genspark:
- Typically start from chat or developer consoles rather than a consumer-grade todo app
- Their current focus is also on high token consumption enterprise tasks (e.g. coding, slide generation, spreadsheets).
- Manus and Genspark "wrap" existing AI models, which limits their ability long-term to outperform baselines

## Technology & Differentiation

We believe that [The Model is the Product](https://vintagedata.org/blog/posts/model-is-the-product), and have invested in training Jan's own long-horizon task planning models:
- Jan v1, Jan v2 agentic models with long-horizon capability
- Data feedback loop via Todo app, for RL datasets
- Continuous training pipeline
- Task-based evals, that increase in difficulty with capability

Jan builds with a full-stack approach, and tight coupling of our Jan models and Jan agent stack results in best-of-class outcomes[^2]:
- Jan Models are e2e RL-optimized for Jan's Agent Stack
- Agent orchestration layer (e.g. workflow, state management)
- Agentic infrastructure layer (e.g. Browsers, VMs, Telephony)

## Team

Jan team previously built:
- Jan, a desktop-based local AI app (4 million downloads)
- Jan models, an agentic model capable of long-horizon tasks

## Financials & Forecast

| Year | Revenue | Paying Customers | Key Assumptions |
| ---- | ------- | ---------------- | --------------------- |
| 2026 | $XM | 2,000 | Paid beta, $Y/mo ARPU |
| 2027 | $10M | 20,000 | $50/mo ARPU |
| 2028 | $100M | 84,000 | $A/mo ARPU |

- Gross margin path:
- 55% → 70% as Jan's in-house models get optimized and cheaper to run
- We currently run Jan in our own datacenter and do not give up margins to the cloud
- Burn dominated by engineering and GTM hires

## Funding Ask

- **Raise:** $XM seed at a $XXm-capped SAFE
- **Runway:** 12 months to reach $10M ARR exit velocity for Series A.
- **Use of funds:**
- 60% product & engineering (agent reliability, connectors, infra)
- 20% GTM (marketing-led sales, growth experiments, community)
- 10% operations & compliance (security reviews, SOC2)
- 10% contingency / working capital

## Risks & Mitigation

1. **Agent reliability:**
- Risk: Agent success rate is <70%, affecting user adoption
- Solution: Invest in evaluation harnesses, continuous training pipeline, and create datasets from failure cases
2. **Data security & compliance:**
- Risk: Todo apps hold a lot of private, confidential information
- Encrypt data at rest, pursue SOC2, invest in threat defense
3. **Customer education:**
- Autonomous agents are new for mainstream users.
- Invest in marketing, customer education, and customer success

# Appendix

## Jan's Desktop App

Jan's main repo (with 38k stars) will continue to be public:
- Grows into a multi-repo product
- `janhq/jan` will be an informational repo
- `janhq/server` will be the more actively maintained server
- `janhq/desktop-server` will be the current Jan Desktop

## Jan's Repo Structure

Jan's will have a new repo, that

- `janhq/jan`: our main repo with 36k stars

Jan's AI Agent is delivered via Jan's servers:

- `janhq/server`: Recommended infrastructure/backend (API, Agent), AGPLv3 license
- `janhq/desktop-server`: Lightweight desktop-based server, MIT license

Jan is used by consumers through clients that connect to Jan Servers:
- `janhq/mobile`: Jan mobile apps (iOS, Android), closed source
- `janhq/clients`: Jan web, browser extension, etc

Jan's model training pipeline is closed-source:

- `janhq/models`: Jan's Agentic models
- `janhq/training`: Jan's Training Pipeline
- `janhq/datasets`: Jan's Dataset (and crown jewels)

We may open-source Jan's smaller or last-generation models, but keep high-capability models proprietary.

_Example: Bitwarden_

[^1]: https://vintagedata.org/blog/posts/model-is-the-product
[^2]: https://x.com/MatternJustus/status/1876829656483086601