You watched the video. Now do the thing.

Drop in a recording and say what you are after. It listens, or looks at the picture, and writes back a plan you can execute, the steps behind how something was made, or an honest briefing — with a timestamp behind every specific claim.

Read a videoஇலவசமாக முயற்சிக்கலாம். நிறுவுவதற்கு எதுவும் இல்லை.

இது எவ்வாறு இயங்குகிறது

மூன்று படிகள், அதில் எதுவும் “கட்டமைத்தல்” இல்லை.

01
Drop the video in
Drag in an mp4, a screen recording, a lecture capture, an audio file. Say what you are after in a line — the same conference talk becomes a plan, a summary or a decision about whether to watch it, depending on why you brought it.
02
It listens, or it looks
Transcribing the audio is right when the value is in what is said. Looking at the picture is the only way to catch what is demonstrated silently. It picks, tells you which and why, and you can overrule it.
03
You get a file, not a chat
The output is written to the workspace as plan.md, how-it-was-made.md or briefing.md. The conversation is where you argue with it; the file is what you keep.

இது என்ன செய்ய முடியும்

It watches, not just listens
A transcript loses everything a video shows: the menu that was clicked, the setting that was changed, the joint that was cut at an angle. When the information is on screen it hands the file to a model that can see, and reports what is there rather than what was narrated.
It writes what the video supports
A tutorial becomes numbered steps with the settings named. A how-it-is-made becomes how it was done and what it would take to do it yourself. A talk becomes a briefing. It does not force a music video into a project plan to fill a template.
It says what it could not tell
A blurred terminal, an inaudible passage, a step performed off camera — these are findings and it reports them. "The config is on screen at 4:12 but unreadable" is more useful than a confident guess at what it said.

இது எவ்வாறு அமைக்கப்பட்டுள்ளது

நீங்கள் உள்நுழைவதற்கு முன் எதைப் பெறுகிறீர்கள் என்பதை அறிந்துகொள்வதற்கான செயல்பாட்டு முறைகள்.

சூழல்: A Linux sandbox with a network. The video you drop in stays in the session workspace.
கருவிகள்: Transcription for the audio, a video-capable model for the picture, a shell, and file tools for what it writes.
மாடல்: Any chat model in the catalogue, and you can change it mid-conversation. Looking at the picture uses a video-capable model regardless of your choice.
தொடக்கக் கோப்புகள்: A brief to fill in — what you want out of it, and optionally how closely to read it.
எது சேமிக்கப்படும்: The video and everything written about it stay in the workspace, so a later session can pick up where this one stopped.

மக்கள் இதனிடம் கேட்கும் விஷயங்கள்

“Turn this outbound workflow recording into a plan I can run this week”
“How was this made, and could I do it myself?”
“What is in this talk? I have ten minutes, not fifty”
“Watch this UI walkthrough and write down every setting they changed”

இது எதைச் செய்யாது

இவை ஒவ்வொன்றும் நாங்கள் எதிர்கொண்ட உண்மையான கட்டுப்பாடே தவிர, திட்டமிடப்பட்ட அம்சம் அல்ல.

It usually cannot download from a link. YouTube refuses servers, so pasting a URL will often get you the title and an apology rather than a plan. It will tell you which video you linked and ask for the file. Drop the file in and everything works.
Uploads stop at 20 MB, which is about seven minutes of video. A long talk does not fit yet — upload the audio on its own and you get roughly twenty minutes, which is the honest workaround until the ceiling moves.
A long video is a slow read and a real cost. Past about forty minutes it will say roughly what it will take before spending it, and offer to work from the section you care about instead.
It is not a substitute for watching something you enjoy. It is for the forty-minute talk you were told to watch and the tutorial you need the settings out of.

எந்த மாடல் இதைச் சிறப்பாகச் செய்கிறது

Measured over real runs of this agent, per model. A run counts as finished when it produced the answer on its own — nothing failed, nobody was asked to approve anything, and it did not run out of steps. Read all three columns together: a model that finishes fast by giving up scores badly on the first, and one that completes everything by grinding scores badly on the second.

Model	நிறைவு விகிதம்	இடைநிலைச் செலவு	இடைநிலை நேரம்	Runs
kimi-latest~moonshotai	100%	$0.217	3m 16s	3

நிறைவு விகிதம்: நீங்கள் தலையிட வேண்டிய அவசியமின்றி வேலையை முடித்த இயக்கங்கள்.
இடைநிலைச் செலவு: ஒரு சாதாரண இயக்கத்திற்கு ஆகும் செலவு, கிரெடிட்களில்.
இடைநிலை நேரம்: முதல் செய்தியிலிருந்து பதில் வரும் வரையிலான கணக்கிடப்பட்ட நேரம்.

Medians over the runs behind each row. A model appears once it has 3 runs on this agent, and the run count is shown so you can judge how much a figure rests on.