A computer that can do things for you.
Lightcone works in its own persistent environment, so you can start tasks, come back to previous work, and keep moving.
01
One persistent environment
02
Come back to previous work
03
Rerun and share completed tasks
04
Schedule work to happen later
Handles useful tasks across the tools you already use
Order food for delivery
Reorder something you buy often
Book a stay that fits these constraints
Find a good movie or a show
Work you can return to. Past tasks, saved context, scheduled runs, and shareable results — all in one place.
Create a Linear issue to fix the slow loading on Tzafon.com
Just nowDownload the best 3 movies and shows
2h agoUpdate Terms & Policy page
YesterdayOrder sushi for 2 people
ThuFind me the best Thunderbolt 4 cable under $80
Disable click-through link on stripe
Mar 19Weekly task: Remove unused features
Mar 17Show me some good stays for my Japan trip in july
Mar 6Fix cache on initial loading page in Lightcone
Feb 23Weekly task: Remove unused features
Feb 16Review V1 consumer scope
Jan 29Try it yourself.
$ pip install tzafon
from tzafon import Lightcone
client = Lightcone()
Full stack. Open source.
30x lower cost.
Lightcone
General purpose LLMs
Browser infrastructure
CUA model
Northstar (4B, purpose-built)
General model + tool use
Bring your own
Browser infrastructure
50K+ concurrent sessions
Not included
Yes
Desktop / VM
Full GUI (X11, persistent)
No
Limited
Price
$0.50 / M tokens
~$15 / M tokens
Infra cost + model cost
Open-source model
Apache 2.0
No
n/a
On-prem / air-gapped
Yes
No
Partial
Custom model training
Yes
No
n/a
Single API (model + infra)
Yes
Model only
Infra only
A foundation model lab for computer use. We train vision-language models and build infrastructure for autonomous computer use.
Founded by researchers and engineers from DeepMind, Jane Street, Citadel, ETH Zurich, MIT, Google, and Amazon AGI. Backed by HV Capital, OpenAI, xAI, Streamlined Ventures, and Kakao.