Skip to content

Autonomous Media: From data to a published video. Every morning, fully automatic.

A pipeline that turns verified data into a finished video, with script, voice, rendering, quality checks and publishing. Its first show is SOLLE’s Daily Radar, published every morning.

What happens when publishing becomes a system?

Most AI content is still a person typing prompts. We wanted to know how to take that person out of the daily routine and still keep the output true.

The answer: strict inputs, fixed facts, a script that cannot improvise, and quality checks that stop the run instead of guessing.

First implementation · SOLLE Daily Radar

Every morning, a timer starts a newsroom.

Daily Radar pipeline

01 Select Three tokens, picked by fixed rules.
02 Verify Market and holder data measured and locked in.
03 Script The template, filled with those facts.
04 Voice One continuous take.
05 Render Lip-sync, charts, holder maps, captions.
06 Check Automated tests. Any doubt stops the run.
07 Publish Counts once the platform confirms.
08 Distribute Website, community, social.

One real morning · 27 September 2026 Times in UTC · no manual rendering or upload

05:00The timer starts the run.
05:03Three tokens picked, facts locked in.
05:16Published. YouTube confirms the video is processed.
07:26Announced on X.

Episode · 29 September 2026 SOLLE Daily Radar: Top 3 Picks of the Day · 2026-09-29
A real episode, produced and published by the pipeline.
Nothing loads from YouTube until you press play. Then the video loads from YouTube (Google); see Privacy.

Four frames of a vertical Daily Radar episode: the title card with the day’s top three picks over the purple host, holder maps for two of the tokens, and the host signing off. — Four moments from the Daily Radar of 3 October 2026: the day’s three picks, maps of who holds two of them, and the sign-off. Video frames youtube.com/shorts/LRatSJG88rE

01 · No guessing

No fact, no episode.

If a number can’t be verified or the data is out of date, the run stops. On days with fewer than three qualifying tokens, there is simply no show.

02 · Confirmed

Done means the platform says done.

A video counts as published only once YouTube confirms it. Then the website, community and social posts follow. If something fails, a person gets an alert.

03 · Research

Open video models, put to the test.

We test open-weight models for lip-sync and video on our dedicated GPU server. A model goes into production only when it holds up every morning.

Beyond SOLLE

Any field with structured data can have a daily show.

The same approach works wherever there is data people want explained every day: markets, operations, product changes, research.

If your company has data like that, talk to us.

Spec sheet

Cadence: Every morning at 06:00 Canary time, started by a timer.
Selection: Three Solana tokens a day, picked by fixed rules from a live watchlist.
Script: A fixed template filled with verified facts. No language model improvising.
Voice: One dedicated character voice, made with neural text-to-speech.
Picture: Lip-sync, charts, holder maps and captions, rendered on our own setup.
Format: Vertical video, about one minute long.
Distribution: YouTube, solle-ai.com, Telegram and X.

What this is not

A language model making things up. The script only says what the data says.
Unattended. People set the rules and get an alert if something fails.

Next arm · 05 Autonomous Web The website as software, built around the business.

Have data people want explained every day?