Choose your preferred language to see content for your region.

Zortex
Zortex

Zortex Vision: Your second intelligence.

Your second intelligence.

Your life is spread across a dozen apps. Your email knows what you promised, your calendar knows where you'll be next week, and your notes, files and messages each hold part of the same story. No AI you use can see all of it.

Zortex Vision puts those pieces together on your own computer. It links the apps you already use and learns how the people, projects and plans in them connect. Then any AI you choose can work from that context. Ask about your own life and get an answer with its sources, or ask it to take the next step.

Zortex Vision is coming to Windows and macOS. Early access is free.

Ask about your own life.

Ask the way you'd ask a colleague who has been there all along. Vision finds the email, the message and the meeting that hold the answer, and shows you where each part came from.

The examples below use a demo account for a made-up person.

“How did I spend this week?” Vision adds up the hours from your calendar and time tracker and shows which client each one went to.

“What did I promise and haven’t done?” Vision finds the promises you made in messages and email and shows how long each one has waited.

“Who haven’t I spoken to in a month?” Vision checks every inbox and chat for each client and lists the ones that went quiet.

It acts for you.

When the answer is a task, Vision can take the next step in the apps you've linked: draft the follow-up, open the file you need. It acts only when you ask, and every action leaves a receipt that says what it did and what it was based on.

“Draft a check-in to all three.” Vision writes each message from your last conversation with that person, ready for you to send.

“Open the Harbour Point letter.” Vision finds the latest draft and opens it, with the reason it came up: the client has asked twice.

Every action is listed with what was done and the records it came from.

Every app, one world.

Most assistants start from zero each time. Vision keeps a living map of your world: who you work with, what you're working on, what was decided and what is still open. Each app you link adds to the same map, and when something changes in one app, the map changes with it.

Each app you link adds its people, projects and conversations to one world.

Yours to hold.

Your world lives on your computer. You decide what each AI can see, and you can remove anything at any time. Cloud access, for reaching your world while your computer is off, is your choice. Your data is never used to train AI models.

Benchmarks

Every good answer starts with finding the right pieces. We measured both steps on five public question-answering benchmarks, 1,000 questions each, and set Vision beside other methods on the same questions.

Vision's answers score highest of the methods listed on all five. At finding the evidence it is ahead on four, and on NQ it is 0.11 points behind the best.

Answer quality (QA F1) · 2WikiMultiHopQA
0102030405060708090100 85.28 Vision 61.5 NV-Embed-v2 71.0 HippoRAG 2 75.3 PropRAG 78.06 SAG

Questions that need facts from two Wikipedia articles.

Answer quality (QA F1) · NQ
0102030405060708090100 68.96 Vision 61.9 NV-Embed-v2 63.3 HippoRAG 2 62.5 PropRAG

Real questions people typed into Google search.

Answer quality (QA F1) · PopQA
0102030405060708090100 71.01 Vision 55.7 NV-Embed-v2 56.2 HippoRAG 2 56.4 PropRAG

Questions about people, places and things, many of them little known.

Answer quality (QA F1) · HotpotQA
0102030405060708090100 83.71 Vision 75.3 NV-Embed-v2 75.5 HippoRAG 2 76.1 PropRAG 79.66 SAG

Questions that need facts from two or more articles.

Answer quality (QA F1) · MuSiQue
0102030405060708090100 73.07 Vision 45.7 NV-Embed-v2 48.6 HippoRAG 2 53.9 PropRAG 61.15 SAG

Questions that chain two to four facts together.

Finding the evidence (Recall@5) · 2WikiMultiHopQA
0%10%20%30%40%50%60%70%80%90%100% 96.57% Vision 76.5% NV-Embed-v2 90.4% HippoRAG 2 94.1% PropRAG 93.34% SAG

Questions that need facts from two Wikipedia articles.

Finding the evidence (Recall@5) · NQ
0%10%20%30%40%50%60%70%80%90%100% 77.89% Vision 75.4% NV-Embed-v2 78.0% HippoRAG 2 77.9% PropRAG

Real questions people typed into Google search.

Finding the evidence (Recall@5) · PopQA
0%10%20%30%40%50%60%70%80%90%100% 70.70% Vision 51.0% NV-Embed-v2 51.7% HippoRAG 2 56.2% PropRAG

Questions about people, places and things, many of them little known.

Finding the evidence (Recall@5) · HotpotQA
0%10%20%30%40%50%60%70%80%90%100% 97.95% Vision 94.5% NV-Embed-v2 96.3% HippoRAG 2 97.4% PropRAG 96.50% SAG

Questions that need facts from two or more articles.

Finding the evidence (Recall@5) · MuSiQue
0%10%20%30%40%50%60%70%80%90%100% 82.54% Vision 69.7% NV-Embed-v2 74.7% HippoRAG 2 78.3% PropRAG 80.36% SAG

Questions that chain two to four facts together.

Answer quality, QA F1 (higher is better)

Method NQ PopQA MuSiQue 2Wiki HotpotQA
Vision 68.96 71.01 73.07 85.28 83.71
NV-Embed-v2 61.9 55.7 45.7 61.5 75.3
HippoRAG 2 63.3 56.2 48.6 71.0 75.5
PropRAG 62.5 56.4 53.9 75.3 76.1
SAG — — 61.15 78.06 79.66

Retrieval, Recall@5 (%)

Method NQ PopQA MuSiQue 2Wiki HotpotQA
Vision 77.89 70.70 82.54 96.57 97.95
NV-Embed-v2 75.4 51.0 69.7 76.5 94.5
HippoRAG 2 78.0 51.7 74.7 90.4 96.3
PropRAG 77.9 56.2 78.3 94.1 97.4
SAG — — 80.36 93.34 96.50
2026

Footnotes

Zortex

Select language