You + AI agents = a media company.
An open, co-built media network. Every node is a co-builder. One platform solves production (skills + models), distribution (apex curation), revenue (70/15/15), and even fundraising (LaunchPad-style token issuance — coming).
Your device IS a node in the media network.
Each publisher is a node with self-sovereign identity (wallet / DecentLAN ID). Listeners can play on the interview.network site, on your subdomain, or inside our Beagle Chat App (where they can also DM speakers). Subdomains and the DecentLAN subnet inside Decent Network exist only to stay compatible with the browser + HTTPS + DNS stack we all use. Underneath, Decent Network is an identity-based secure overlay among trusted nodes — end-to-end encrypted, peer-to-peer, no public IP, no open ports exposed to strangers.
┌──────────────────────────────────────────────────────────────────┐
│ ① YOUR MACHINE (Mac / Linux / +laptop GPU) │
│ install: skills, scripts, ollama, mlx-whisper, seed-vc │
│ generate Chinese audio from a YouTube source │
│ runs Next.js on localhost:3004 → your subdomain │
└────────────────────────────┬─────────────────────────────────────┘
│
ASR-QC quality gate
(zh_fidelity ≥ 0.90)
│
▼
┌──────────────────────────────────────────────────────────────────┐
│ ② YOUR SUBDOMAIN <your-name>.interview.network │
│ Decent Network (secure friend overlay) → localhost:3004 │
│ listeners discover & play here first │
│ access logs accumulate on your machine (/local-stats) │
└────────────────────────────┬─────────────────────────────────────┘
│
14 days + 100 unique IPs + ≥ 40%
listen-completion proxy
│
▼
┌──────────────────────────────────────────────────────────────────┐
│ ③ APEX interview.network │
│ qualifying episodes cached + listed globally │
│ beta.interview.network = explorer + leaderboard │
│ revenue split: publisher 70% · speaker 15% · protocol 15% │
└──────────────────────────────────────────────────────────────────┘What you need to become a publisher
- Apple Silicon Mac (M1+, 16 GB+) OR Linux box with 6 GB+ NVIDIA GPU
- Install the DecentLAN client inside Decent Network — join the secure friend-overlay so trusted nodes can reach each other end-to-end-encrypted, no public IP or open ports
- A GitHub account to clone the repo
- Familiarity with terminal commands (zsh/bash); no front-end coding needed
- Optional but recommended: a Gemini or OpenAI API key for cloud-fallback generation
Drop a URL. Get a Chinese episode.
The animation below mirrors the real pipeline (publish.py + the clone path). On a 30-min English video this takes ~25 min wall-time on a single Mac, ~4 hr on pure 3060, or ~45 min on Mac+3060 combined.
Three tiers — which one fits you?
This is a poll, not a signup. Tap one — we use the signal to decide what to launch. (We log only the tier you pick + timestamp; no PII.)
- Full source code, all skills, monthly updates
- <name>.interview.network subdomain
- Tech support via shared chat
- Runs free on your own machine (Mac MPS / NVIDIA GPU)
- If you want cloud TTS, you sign up + pay the vendor directly
- ~$30/month of cloud-generation credits, managed by us
- No need to open Gemini/OpenAI accounts yourself
- Cloud queue when your local machine is busy / asleep
- Roughly 60 hours of generated audio per month
- Standard support response
- ~$80/month of cloud credits — ~160 hours of audio
- Priority support + early-access to new features
- Custom voice fine-tuning when available
- Direct access to the team for ops issues
Get on the list — we'll reach out when slots open.
We're launching publisher slots in waves. Email is fastest; social follow keeps you in the announcement loop.
70 / 15 / 15.
In that order.
The publisher running the OPC takes the largest share — they took the risk and made the call. The speaker gets paid every time their voice appears, in any language. The protocol takes the smallest cut to fund the rails.
Token layer on the roadmap.
Things people ask before signing up
What do I need to run this on my own machine?＋
A Mac with Apple Silicon (M1/M2/M3/M4, 16 GB+ RAM) OR a Linux box with a 6 GB+ GPU. We install Ollama + a small Python venv. If you have both Mac and a Windows laptop with NVIDIA, you can split the work for ~10x speed on voice synthesis. DIY tier requires you bring your own Gemini/OpenAI key.
Do I need to know how to code?＋
You need to be comfortable running a Mac terminal and reading short docs. The pipeline IS scripts — we provide them, you invoke them. No web GUI editor by design (Brief 13b — auth is via wallet signature, not a web form).
Why a federated subdomain instead of uploading to interview.network?＋
Your subdomain = your source of truth. You control your content. Apex curates the best onto interview.network after audience signal (14 days + 100 unique IPs + 40% completion). You never lose your content if the apex changes policy.
What about copyright on source material?＋
Publish only what you have the right to. Every episode credits the original creator and links back, and the Chinese audio is clearly labelled as an AI-generated dub.
Whose voice is used?＋
The original speaker's voice via voice conversion (Plan C, seed-vc) or direct cloning (Plan B, Qwen3-TTS). Output is labelled AI-generated and can be watermarked.