You don't need a podcast host. Turn on your camera. Weaver interviews you.

The interviewer is heard but never seen. That is intentional: you remain the visual focus, and the audience never watches a fake avatar. The AI hears your microphone, reasons, and speaks back with low latency. Meanwhile Weaver records your camera and microphone locally, verifies the footage to R2, and captures the transcript. After you end, the analysis finds standalone moments: stories, strong opinions, surprising numbers, emotional beats — not just the loudest second. Each moment can be rendered as (1) full exchange with the AI question audible, (2) question turned into a hook card, or (3) your answer alone. The transcript hours you chose to process and the renders still bill like any recording — interview minutes only paid for the live interviewer.

Brilliant makes connections across earlier answers. Urban keeps it contemporary, not corporate. Investigative pushes for dates and dollars. Provocative asks what others avoid. Story Hunter hunts anecdotes and turning points. Content Director hunts viral standalone lines. The 5-step intensity slider — Easygoing to "Don't Let Me Off Easy" — dramatically changes the experience: at 1 it moves on, at 5 it challenges vague answers until you give a number. Energy (Laid Back → Intense) and interruption style (Never → Aggressive) tune the feel. The custom field (up to 1,000 chars) is normalized before it reaches the model — not dumped raw — so "don't kiss my ass" becomes a real instruction without prompt injection risk. Save a combination as "The Thinker" and say next time: "Interview me with The Culture Host."

A perfect interviewer asks short, intelligent questions that cause long answers. Creator should speak 75–85% of the time. That is better content and better economics — less generated audio to bill. The system prompt encodes that: one short question at a time, no monologues, no "that's fascinating" filler, follow the interesting answer, ask what happened next, what it felt like, what changed after, challenge vague claims, remember previous statements, and bring them back 20 minutes later. The host talk-time ratio is visible in analytics so you know whether you actually got to talk.

OpenAI's Realtime session caps around 60 minutes. Weaver's backend keeps structured interview memory separate from the raw Realtime context: guest facts, people, open threads, notable statements, last 3–5 questions, emotional tone, stories in pursuit, topics remaining, desired ending. Two minutes before the cap, Weaver writes a handoff packet; the next session receives it immediately. A 120-minute interview uses ~3 Realtime sessions but remains one Weaver interview — the guest hears "Before we move on, you said something a few minutes ago I don't want to let you get away with..." not "Hello, what are we discussing?"

Whenever possible Weaver saves three tracks: your mic, the AI, and the mix. That lets you remove the question, replace it with a hook card, or keep the exchange — without re-recording. Interview time runs only while the AI session is active. Pausing, reconnecting after a network drop, voice previews, and the setup screen don't count. At 5 minutes left you get a subtle notice; at 1 minute the host wraps naturally and never cuts you mid-sentence. If the network drops, your recording continues locally and the host says "Interviewer reconnecting..." Restored from the last memory checkpoint, it resumes without asking what you're talking about again. And the footage still needs storage and a transcript hour to become searchable — interview minutes don't replace those allowances, they sit beside them.

The room is a large camera feed, not a chat transcript. Side rail: interview time remaining, recording indicator, current question, topic progress. Controls: Pause, Camera, Microphone, End Interview, Mark Moment. When the host asks "When did you realize the company might work?" the canvas briefly shows WHEN DID YOU KNOW IT WAS REAL? — kinetic typography, not an avatar. Vertical export puts the question on top, you in the center, captions at the safe 0.74 baseline — the signature Weaver Interview look without spending engineering on a fake face that would fight you for attention.