NetShow: the platform for agentic AI

Not a harness. A platform.

A harness runs one agent’s loop. Many companies have one. NetShow is where agents live, work, remember, talk and sell, and where people meet them.

We bring together the frontier models, with open-weight models built in, and a router that can use any harness, other companies’ and our own. In front of it all we put a living face and voice. The agent greets people first, says it is an AI, and does the work.

The platform is built in layers. Each layer is open through hooks our own team and outside developers build on, the layers work together in one conversation, and they all serve one thing: the person’s goal. If one layer cannot finish a task, the next one takes it, up to a person on the owner’s team.

LIVE Agents that speak first with a face and a voice
LIVE Runs on the frontier models, chosen by the job
BUILT Open-weight models as cheaper brains
LIVE Uses the harness a team already runs

The one agent you come home to

People try many AI apps and harnesses. NetShow.ai is built to be the one agent they come home to: a live host that talks with them and looks, sounds and feels the way they want.

Free to talk, pay for the work. Talking with the host carries no per-minute charge; the work it does runs on credits or on your own key.

Through that one host you reach your own harnesses, your own model keys and the newest voice models. Close to six thousand capabilities stand behind it.

LIVE Talk with a live host on the web, with no sign-in and no per-minute charge to the visitor
LIVE Bring your own model key
LIVE The newest realtime voice through the NetShow voice proxy, one voice per host
LIVE Pair your own harness to your host

The layers, in plain words

The doors. Our site, a business’s own site with one pasted line, a lobby TV, a phone call.

The alive presence. A face and a voice that say hello first, move around the page, point, and draw what they mean.

The models. The best brain for the job, swapped the day a better one ships.

Memory. It remembers what it should, where it should, and forgets a visitor who asks.

Tools and hooks. It does things: looks up, books, makes images and video, drives the page.

The harness router. It can hand work to any agent team, a team’s own or one of ours.

Computer use. When nothing connects, an agent can use a website the way a person does.

A person. When it should not finish alone, it hands over with everything done so far.

Across every layer: credits and spend ceilings, and trust: the AI label, consent, forget-me.

LIVE One pasted line puts the host on any site
LIVE Forget me: a visitor is erased on request
LIVE Browser and computer-use tools behind a checked gateway
BUILT The agent’s own sandbox computer through a paired harness
LIVE Hands a visitor to a person with the reason and contact

Built to be built on

Developers reach NetShow the ways they already work: MCP in both directions, agent-to-agent (A2A), webhooks, the pasted widget, the harness connector, and tools that drop in as one file.

An outside harness can act on the stage itself: say a line, show progress, show a card, gesture or ask, as plain validated data.

Skills are sold in a marketplace, so a builder’s work can earn.

LIVE NetShow runs its own MCP server with 9 tools
LIVE Machine doors for agents: llms.txt, agent card, MCP card, catalog
LIVE A2A door: agent card and JSON-RPC message/send
LIVE The harness stage language: say, progress, card, gesture, ask
LIVE Tool providers drop in as one file
LIVE Skill marketplace
NEXT One page for developers that lists every hook

The harness router, one layer

The owner talks to one agent. That agent runs the rest: “ask my Hermes to do this” goes to that team’s own agent, and the answer comes back in the same voice.

The router is how people bring in what they already set up elsewhere, the harnesses and agent teams they configured on other platforms, and put it behind one alive face.

The router picks which harness answers and says why. When a paired harness goes quiet, it falls back to our own and records the reason.

eleven harness recipes are ready for an owner to pair today; more are drafted and are not named until reviewed.

LIVE The owner’s agent conducts harnesses by voice
LIVE The router picks with named reasons and falls back when a harness is silent
LIVE Which harness answered each turn is recorded

The fallback ladder that finishes a task

A task should finish, or end honestly with a person. Never most of the way, then silence.

Each step has a time limit, a paid step never starts without a daily spend ceiling, the host says which step is working, and a stuck task always ends with a person and everything done so far.

LIVE A silent paired harness falls back to our own, with the reason recorded
NEXT One ladder for a task: tool, other harness, computer use, person, and the visitor sees which is working

Step	What tries	What the person hears	Today
First	The agent’s own brain and memory	Nothing extra: she just answers	Live
Second	A NetShow tool	“Let me check that for you.”	Live
Third	An outside tool through MCP, A2A or a webhook	“Asking your calendar.”	Partly built
Fourth	Another harness: the owner’s paired one, then a borrowed one	“Your Hermes agent is on it.”	Live: the router and its fallback
Fifth	Computer use in a sandbox, under a daily ceiling	“No direct way in, so I’m using the website myself. You can watch.”	Built, not wired
Last	A person	“I got this far. I’ve passed the rest to the team with everything so far.”	Live for site sales

Why the face matters

Most AI still waits behind an empty box. Our agents greet first, by voice, in the visitor’s language, and say they are an AI.

The face is a small live drawing, not a video: a live text face on a tiny button, a sketch, then 2D and 3D as the room and the connection allow. A photoreal face is a separate paid look, and never a real person’s face without that person’s consent.

The visual connection is what we are cracking: on the page the host sits beside what you read, and on the stage she steps out of her box, moves over the page and shows what she means.

The same agent can leave the page: a lobby TV today, a desk robot in simulation next.

LIVE Greets first, in the visitor’s language, with the “you can interrupt me anytime” line
LIVE Owns the stage: steps out of its box and moves over the page
LIVE A TV in your lobby
NEXT An open-source desk robot

NetShow.com: the entertainment center

NetShow.com is where the same alive hosts entertain: the Theater, with shows and stories led by a live host. Watching is free; making your own show or story is pay to play.

Later come virtual worlds where people meet and vibe: clubs such as ClubSpecies.com, date night, and places to meet and chat, each with an alive host who says it is an AI.

LIVE The NetShow Theater with a live host, free to watch
NEXT Virtual worlds: clubs, date night, places to meet and chat

What is already built (measured)

NetShow has been built since November 2023. Our own atlas reads the code and records each capability with its state.

The September atlas snapshot records 5,355 distinct capabilities; 2,779 of them are classified LIVE. This is a dated classification of the code, not an uptime check.

By area in that snapshot: 618 capabilities for live avatars and looks, 601 for the agent runtime, models and tools, 317 for the harness, router and outside agents, and 671 for the doors, widget and marketing pages.

1,151 business sites are built; each is still greeted by Mrs. NetShow, and giving each business its own agent is the next step.

LIVE The September atlas snapshot records 5,355 distinct capabilities
LIVE 2,779 classified LIVE on that date

Pay to play: how it makes money

AI employees are monthly: Front Desk from $699 a month, an AI Team from $4,999 a month and an Enterprise AI Workforce from $9,999 a month; the exact price is set in writing after a build call.

Free to talk, pay for the work. Everything the agent does that costs money runs on credits or on the owner’s own keys, and every paid lane has a daily ceiling.

Stories and movies are one-time purchases; watching is free. Skills sell in the marketplace.

So the platform earns in plain lines: monthly AI employees, usage credits, enterprise builds, pay-to-play stories and shows, and the skill marketplace.

LIVE AI employee ladder, monthly
LIVE Bring your own model key
LIVE Credits and spend lanes with daily ceilings

Why a platform, why now

Models can already do far more than people use. The gap is not intelligence; it is putting it in people’s hands. A face and a voice that speak first close that gap for people of every age.

Harnesses are becoming common. The value moves to what sits above them: the place people meet the agent, the memory, the trust, and the router that lets any harness do the work.

We build a few months ahead of the models: every brain is a lane we switch, every harness a recipe we add, so the platform gets better when the models do.

What comes next

NEXT The fallback ladder: the task finishes, or says where it stopped
NEXT The host as onboarding: it sets up a newcomer’s first agent while they watch
NEXT Tell the host what it got wrong; it becomes a test the next version must pass
NEXT A developers’ page with every hook

Next step

Ask Mrs. NetShow about any layer, or read the growth case.

Open the growth case

What we have measured

Only numbers the platform counted from its own records. Anything not yet measured is left out.

Serving cost against ceilings USD 0.13 Today, UTC; the fence keeps one day. 80 of 80 lanes with a ceiling are under it; 0 lanes spent with no ceiling. Today, UTC, as of 2026-09-27