Both halves, linked onto one timeline.
A video has two tracks: what is said and what is shown. Lirovo lines them up second by second, then connects everything into a knowledge graph of the people, topics, numbers, and decisions inside the video.
What is in the graph
Nodes are the things Lirovo finds in a video. These are the kinds it can place, each one connected to the others through the edges that hold the conversation together.
Who is talking, tracked across the whole conversation.
People named in the video, beyond the speakers themselves.
Companies and institutions that come up, like Mistral or ASML.
Subjects the conversation moves through.
The larger arcs that group several points together.
Numbers that matter, like a valuation or a capacity target.
Money committed, with the amount and what it is for.
Products and platforms mentioned by name.
Assertions made, kept distinct from opinions.
Notable lines, attributed to who said them.
Choices stated on the record.
A figure called out as a headline number.
Text and data read off a slide or lower third on screen.
From a timeline to a graph
First, alignment. Lirovo transcribes the audio and reads the frames, then places both on the same timeline. A spoken sentence and the slide that was on screen at that moment end up at the same second, so the two tracks can be read together instead of in separate files.
Then, connection. From that aligned timeline Lirovo builds the knowledge graph: nodes for the entities it finds and edges for how they relate. A speaker discusses a topic; a topic carries a metric; an organization makes an investment. The structure that was implicit in the conversation becomes something you can hold.
A transcript is a flat list of words. The graph is the structure underneath the conversation. Every node carries where it appears, one or more evidence spans across audio and vision, so you can always trace it back to the source. See how that is kept honest on the evidence page.
Not a screenshot, a graph you can query
The graph is queryable, not static. Explore it, jump from any node to the exact second in the video, and read it back in your terminal or from your agent harness over MCP.
“CEO of Mistral.”
The graph view from the product, on a real run. Every node links to its moment in the source video.
See how every value is proven
Every node in the graph points back to the moment it came from. Follow that thread to the evidence, or run a video through Lirovo and watch the graph build live.