AI Livestream Selling Glossary

Plain definitions of AI livestream selling terms: AI livestream host, AI avatar, agentic commerce, AI disclosure, voice cloning, stream hours, MCP and GMV.

Plain definitions of the terms that come up around AI avatars, livestream commerce, and the models behind them. One term per entry, defined before it is explained.

Agentic AI

Also known as: AI agent, autonomous AI

Agentic AI is AI that pursues a goal by taking actions, rather than only producing text when prompted. An agentic system decides what to do next, calls tools, and reacts to what comes back. A Luqira avatar is agentic in this sense: it watches live chat, decides which comment is worth answering, pulls the answer from your catalogue, and speaks it, without a human directing each step.

Agentic commerce

Also known as: agentic selling, autonomous commerce

Agentic commerce is retail in which software takes the selling actions rather than assisting a person who takes them. The distinction is authority rather than intelligence: a recommendation engine advises a human, an agentic system decides and acts inside limits you set. A Luqira session is agentic in that sense, and it is the framing enterprise buyers increasingly use for this category.

AI avatar

Also known as: virtual host, AI presenter, digital human

An AI avatar is a computer-generated on-screen person that speaks, moves and reacts without being filmed. It is produced by a model rather than a camera, so it can present at any hour, in any language, for as long as you want. Luqira ships 100+ studio-grade avatars and can build a custom one from a short reference clip with the subject's consent.

AI disclosure

Also known as: AI labelling, synthetic media disclosure

AI disclosure is telling viewers that what they are watching was generated rather than filmed. Most platforms now require it for realistic synthetic content, through an on-screen label, a platform toggle, or both, and several treat a false disclosure as a violation in either direction. Each destination's own rule is quoted on its platform page.

AI livestream host

Also known as: AI host, AI streamer, virtual livestream host

An AI livestream host is the on-screen presenter of a shopping livestream when that presenter is generated rather than filmed. It differs from an AI avatar in scope: the avatar is the face and the voice, the host is the whole job, including reading chat, answering buyers and holding a stream together for hours. It is the term this category has settled on.

AI livestream selling

Also known as: AI live selling, automated livestream selling

AI livestream selling is running a shopping livestream in which an AI avatar, not a human host, presents the products and answers viewers in real time. The economics differ from human livestreaming because the marginal cost of another hour or another channel is compute rather than staffing. It is the category Luqira operates in.

AI whitelisting

Also known as: AI livestream approval

AI whitelisting is a platform granting one seller permission to run AI-hosted livestreams, usually by application and usually before the first stream. It exists so the platform knows who is broadcasting synthetic content, rather than because the technology is restricted. Approval attaches to the seller account, not to the software vendor.

Comment to DM

Also known as: comment-to-message, comment automation

Comment to DM is the selling motion used where a platform has no in-stream checkout: a viewer comments a keyword and receives a private message carrying the product link. It is how live selling works on Instagram and Facebook now, and it changes the host's job from closing the sale on screen to starting a conversation off it.

Concurrent streams

Also known as: parallel streams, simultaneous sessions

Concurrent streams are the number of separate live broadcasts running at the same moment. With human hosts, concurrency is capped by how many people you can put in front of cameras at once. With Luqira, one model drives hundreds of concurrent streams, so concurrency becomes a question of the plan you buy rather than the team you hire.

Conversion rate

Conversion rate is the share of viewers who complete a purchase, expressed as a percentage of everyone who watched. In livestream commerce it is the number that decides whether a stream was worth running at all, which is why answering a viewer's question in under a second matters more here than in recorded video.

Digital twin

Also known as: AI clone, AI twin

A digital twin, in livestream commerce, is an AI avatar built to look and sound like a specific real person. It carries their face, voice and speaking cadence so their audience recognises them. Luqira builds one from a short reference sample, only with the consent of the person being cloned, and it is most often used by creators who want to sell while they sleep.

GMV

Also known as: gross merchandise value

GMV, or gross merchandise value, is the total value of goods sold through a channel before returns, discounts and fees are taken out. It is the headline number livestream commerce is usually measured by, because it captures volume across an entire storefront or campaign rather than the margin on any single order.

Inference stack

An inference stack is the hardware and software that actually runs a trained model to produce output, as opposed to the training that produced the model. Who owns it decides who controls latency, unit cost and quality. Luqira serves its avatar and voice models on its own GPUs, with no third-party API in the serving path, which is why those three can be tuned rather than accepted.

KOL

Also known as: key opinion leader, influencer

A KOL, or key opinion leader, is a person whose recommendation moves purchasing decisions within a specific audience. The term is standard in Asian e-commerce where Western markets tend to say influencer. KOLs are the audience for AI digital twins, because their revenue is capped by the hours they can personally be on camera.

Latency

Also known as: response time

Latency is the delay between an input and the response to it. In a shopping livestream it is measured from a viewer posting a comment to the host beginning to answer out loud. Luqira holds this under 200 milliseconds, the threshold below which an exchange reads as conversation instead of a lag.

Lip sync

Also known as: lip-sync, viseme matching

Lip sync is the alignment of an on-screen mouth with the sounds being spoken. Poor lip sync is the single most common reason a synthetic presenter reads as fake. Luqira generates it frame by frame from the audio, so the mouth follows the words rather than trailing them.

Livestream commerce

Also known as: live commerce, live shopping, shoppable livestream

Livestream commerce is selling products through a live video broadcast in which viewers can ask questions and buy without leaving the stream. It merges demonstration, Q&A and checkout into one session. It is mainstream across Asian e-commerce and growing on TikTok, YouTube, Instagram, Shopee and Whatnot elsewhere.

MCP

Also known as: Model Context Protocol

MCP, the Model Context Protocol, is an open standard for connecting AI assistants to external tools and data over a defined interface, so any compliant assistant can use any compliant server. Luqira runs two: a read-only public server at luqira.com/mcp needing no account, and an account-level one at meet.luqira.com/mcp that lets an assistant set up and run your streams.

Multistreaming

Also known as: simulcasting, multi-livestream

Multistreaming is broadcasting one live session to several platforms at the same time. The difficulty is not the video: it is that each platform has its own separate chat, so a human host cannot follow them all. An AI host can, which is why Luqira drives 32 platforms from one dashboard at once, TikTok, Instagram, Facebook, YouTube, Shopee and AliExpress among them.

Objection handling

Also known as: objection management

Objection handling is answering the reason a buyer gives for not purchasing, instead of repeating the pitch. In a livestream those reasons arrive in chat, are usually about price, sizing, delivery or trust, and repeat all session. It is the part of selling an AI host has to do well to be worth running unattended.

Product grounding

Also known as: catalogue grounding

Product grounding is constraining what an AI says to a verified source of truth, so its answers come from your real catalogue rather than from what the model finds plausible. It is what stops a selling avatar inventing a price, a size or a shipping date. Luqira grounds every answer in the catalogue, FAQs and selling points you import.

Stream hours

Stream hours are the billing unit for AI livestreaming: one hour of one avatar broadcasting. Every Luqira plan is priced in them, at a published rate that falls from $15 an hour to $10 as the plan grows, and running two streams for an hour spends two of them. Buying capacity in hours rather than in seats is what makes concurrency a pricing decision rather than a staffing one.

Text to speech

Also known as: TTS, speech synthesis

Text to speech, or TTS, is the conversion of written text into spoken audio by a model. Modern TTS carries intonation, pacing and emotion rather than reading flatly. Luqira Voices is Luqira's own TTS model, trained in-house from random initialisation on the Apache-2.0 OmniVoice architecture over Qwen3-0.6B rather than fine-tuned from anyone's checkpoint. Luqira owns the weights it trained, rents no model from any provider, and serves 85+ languages with 600+ localized voices on its own GPUs.

Voice cloning

Also known as: voice replication, synthetic voice

Voice cloning is building a synthetic voice that reproduces a specific person's timbre and speaking style from a recording of them. A short sample is now enough. Luqira clones a voice only with the consent of the person it belongs to; cloned voices come free with a yearly plan, while the 600+ public voices are included on every plan.