How to use
- Copy the installation command and review the source package before installing it in your own agent environment.
- Check the source document for required tools, credentials and supported hosts.
- Describe your input and desired deliverable; review the output against the source material.
npx skills add https://github.com/atmanity/atmee-toolkit --skill atmee-productBefore you start
- Installation does not include a model subscription, API allowance or permission to access your files.
- Repository stars and installation counts describe activity, not an independent quality or security test.
Rendered from the source repository. Headings, examples, code, tables, links, and referenced images are preserved.
Atmee: the product map
You know what Atmee is, what each part of it does and how it is paid for, so you can answer a question about the product and steer an idea to the right path before anything is built. This file holds what changes slowly: what exists and how it fits together. Numbers change often (prices, plan limits, the balance), so they always come from a live source (Pricing, below), never from this file or from memory.
What Atmee is
Atmee (atmee.ai) makes live video avatars: a face, a voice, a personality and knowledge that hold a real-time spoken conversation with a person over video, lips in sync. In the website's words: build your own live video avatars, with any voice, personality and knowledge, then launch them on Atmee or integrate them anywhere. People make avatars of themselves (a digital twin), a brand's concierge, an expert, a teacher, a companion, a historical figure or an original character.
Atmee v2 also makes video clips: from a photo and a prompt, a short mp4 of that person or character moving as the prompt says, speaking a voice line when one is given (silent without), with no avatar needed (a teaser, a social post, an ad, a product demo). Whenever someone asks what Atmee can do, or wants a video rather than a conversation, say that it makes clips too; create-avatar makes them. Clips need Atmee V2 on the account (see Two engines), and a server offers them only when get_account's pricing.generation has a clip price.
The company was called Atmanity, and the name stays in infrastructure: the API host (api.atmanity.us), the GitHub organisation and some ids. In anything a person reads, say Atmee.
The surfaces
| Surface | For | What happens there | Where |
|---|---|---|---|
| atmee.ai | people who talk to avatars | explore avatars, visit creators' profiles, talk to an avatar from its share link | https://www.atmee.ai |
| Atmee Studio | creators, no code | make and edit v1 avatars, set visibility, share, analytics, API keys (Studio → API & Embeds), plans, top-ups, sponsoring and earnings (Studio → Billing) | https://www.atmee.ai/studio, signed in |
| Avatar API | developers | put an avatar in their own product: start conversations from their backend, create and manage avatars and keys | docs https://www.atmee.ai/docs/api |
| Agent tooling | anyone working through an agent | the skills create-avatar, plan-avatar-app, build-avatar-app, atmee-setup and this one, the avatars server (with generation), the atmee CLI | the atmee plugin |
What each surface offers in detail: references/features.md.
Website pages
When the user asks for a page on the website, give the link the avatars server returns, never one you build: an avatar's links (get_avatar, list_avatars, atmee avatars get --json) are chat, profile and studio; get_account's links are explore, avatars, create, billing, apiKeys, analytics and docs. For a link to talk to an avatar, give links.chat from get_avatar: it carries the avatar's language.
Only when a response has no links (a server that predates them), build the link: https://<host><path>, where <host> is www.atmee.ai; never an API host, which serves no website pages.
| They ask for | Path |
|---|---|
| a link to talk to an avatar | /avatars/<avatar id>?lang=<language> |
| its public profile | /avatars/<avatar id>/profile |
| editing it in Studio | /studio/avatars/<avatar id>/edit |
| Explore | /studio/explore |
| their avatars | /studio/manage |
| making an avatar in Studio | /studio/create |
| credits, plan, top-ups, earnings | /studio/billing |
| API keys | /studio/api-keys |
| analytics | /studio/analytics |
| a creator's page | /@<handle> |
| the API docs | /docs/api |
?lang= is the avatar's language, so the call is in it. A private avatar's pages open only for its owner, signed in. A link to talk to an avatar is always its chat page: never start a session for it (start_session, session chat --open) unless the user asks for a LiveKit room by name.
Ideas worth knowing
- An avatar is a portrait, a voice (cloned from a sample), a persona
(who it is, what it does with the visitor, greeting, guardrails, names) and optional knowledge files. It speaks one language; a session can ask for another.
- Two engines. v1 animates a talking head from one portrait: sharp and
cheaper. v2 adds emotion, expression and motion. Atmee V2 comes with Growth and up; below Growth, an invite code unlocks it (the agent redeems it with redeem_code, atmee redeem CODE in a terminal, or Creator Studio's Billing page), or an email invite does (waitlist: https://www.atmee.ai/v2signup). Building spends no credits but needs some balance usable on Atmee V2; viewers of someone else's V2 avatar need Atmee V2 too unless it is sponsored. get_balance's v2SpendableMicrocredits is how much of the balance is usable on Atmee V2, and get_account's v2Access whether an invite code unlocked it (where V2 is still invite-only, that is what decides). A v2 avatar is captured several times; each capture is a seed (the server says candidate or take, and the API's seed field is an informational number, not the clip) and the creator picks the one that serves. Studio, the Avatar API and the agent tooling (create-avatar) build v2 for an account that can, and v1 for any account. Agents go to v2 by default: without Atmee V2 on the account they suggest Growth and up or ask for an invite code first, and make v1 only if there is neither and the creator agrees.
- Atmee V2 HD (engine
v2hd) is a larger v2 model on its own GPUs, at
larger frames (640×640 by default; limits.v2ProResolutions lists the others). It builds, picks seeds and serves exactly like v2, through the same tools and sessions, at its own per-minute rates (get_account pricing, sessionV2HdPerMinuteMicrocredits and the like). It needs what v2 needs, Atmee V2 on the account, and only some servers offer it yet: get_account's limits.v2Pro is true where the account can make one, false where the plan lacks it (where billing still decides V2 HD by plan), and absent where the server doesn't offer it (then don't mention it). limits.v2ProResolutions lists its sizes.
- `/v1` in a path is the API version, not the engine. Every endpoint
starts with /v1 (POST /v1/sessions, /v1/avatars), and v1 and v2 avatars use the same ones. Never infer an avatar's engine from a path or call a /v1 endpoint "the v1 API"; read engine on the avatar. To a person, name the engines Atmee V1, Atmee V2 and Atmee V2 HD.
- Visibility: private, unlisted (anyone with the link) or public (listed
on atmee.ai). Through the API an avatar starts private or unlisted and goes public by an update once it is ready.
- Status: building, then ready, or failed, or
moderation_hold(a person
reviews it; never retry). Talk only when readyToChat is true.
- Credits pay for everything: conversations by the minute, generation
per call. The API counts microcredits: divide by 100 for credits.
- Monetized and sponsored are opposites, never one setting. A
monetized (Signature) avatar charges the visitor a higher per-minute rate on atmee.ai and the creator earns from those minutes. A sponsored avatar lets visitors talk on the creator's credits from its sponsor pool, and earns nothing. An avatar can't be both. Rates come from get_account, never from memory (references/pricing.md).
- An avatar can talk without being spoken to. An app makes it remark on
something with an event, an agenda's steps carry it on by itself, and it speaks up after a silence. Don't tell anyone it needs the user to speak first, or claim limits these files don't state: say you'll check, and read build-avatar-app's limits.
- One conversation at a time per avatar, one user in it. More users at
once need more avatars.
- The conversation runtime. v2 conversations on the plugin's servers run on Causeur,
which gives apps their controls: room messages (state, event, typed turns), app tools (the avatar calls the app) and agendas (a conductor graph of steps). A session the older behavior module runs accepts tools and never calls them.
Which path fits
Match the idea to a row, name the path in one line, then hand over to the last column.
| They want | Path | Then |
|---|---|---|
| An avatar people can talk to: themselves, a brand, a character | Studio for a v1 avatar on atmee.ai; for v2 (early access), the avatars server | Studio needs no agent; create-avatar for v2 |
| A talking avatar in their website or app, with Atmee running the conversation | Session API: their backend starts the session, their page joins the LiveKit room (there is no embeddable widget) | build-avatar-app |
| The avatar to act on their app: turn a slide, look up an order, keep score | Session API with app tools | build-avatar-app |
| A conversation with steps: an interview, a lesson, an onboarding | Session API with an agenda | build-avatar-app |
| A face for a voice agent they already run on LiveKit Agents | Render path: livekit-plugins-atmee (v1 and v2 avatars from 0.2.0), or the API's avatar sessions for any engine (V2 HD only this way); a portrait-only avatar is enough | https://www.atmee.ai/docs/livekit-plugin/livekit-agents-plugin |
| Visitors talking free on the owner's credits | a sponsor pool on the avatar | Studio → Billing, or directly through the agent or the CLI: signed-in is enough (references/pricing.md, Sponsoring) |
| An avatar live in front of an audience: a stream, a webinar | the broadcast-host shape on the Session API, restreamed (Current limits) | build-avatar-app |
| To earn from their avatar | the verified creator program on atmee.ai | Studio; nothing to build (references/pricing.md, Earnings) |
| A portrait, a voice or a short video they don't have | the avatars server's generation tools, paid per call; a clip of 1 to 15 seconds is priced by its size (480×480 up to 1280×720 or 720×1280; 1080p coming soon) and its length, charged at the next length up (up to 5, 10 or 15 s), with or without sound, and needs Atmee V2 (Growth and up, or an invite code) | create-avatar |
| A voice conversation over a WebSocket, with no video | Speech session (start_speech_session, atmee speech start); the avatar needs a voice | build-avatar-app |
| To manage avatars from code or CI | the Avatar API or the atmee CLI | docs/cli.md in the GitHub repo Atmanity/atmee-toolkit |
Proven app shapes, the sample apps that proved them, and what they taught: references/use-cases.md.
Current limits
Say up front when an idea needs what Atmee does not do yet, with the closest path that works:
- An audience watching one conversation: one user token per session, no
viewer tokens; a broadcast restreams the avatar's video. atmee.ai's own live streaming is not open to everyone.
- Long sessions: the plan caps a session's length; longer use chains
sessions, with a visible cut.
- A different persona per session: the persona belongs to the avatar;
per-app behaviour comes from state and the agenda, or another avatar.
- Eyes through the API: the app runs its own vision model and sends what
it sees as state and event.
- A transcript from the server: the room's captions are the record; the
app keeps them if it needs one.
- Built but not open to everyone: live streams and the avatar profile
page on atmee.ai, mini-apps and apps inside a call, and voice design in Studio (the generation server has it). Name these as coming, never promise them; references/features.md marks each where its surface is described.
build-avatar-app's "Current limits" has the full list for builders.
Pricing
Prices and plans change: quote them only from a live source, with the date you read them.
- Plans and credit packs: Studio → Billing,
https://www.atmee.ai/studio/billing (sign-in needed). There is no public pricing page; /pricing sends visitors to sign in.
- API costs, public:
https://www.atmee.ai/docs/api/overview#costs:
credits per minute for the Session API and for the render path, what a credit costs depending on how it is bought, and what the Free plan allows through the API.
- The account in hand, from the avatars server:
get_account(`atmee
account) for the plan, its limits and per-minute pricing; getbalance` (`atmee balance`) for what is spendable now. Each generation tool's price is in `getaccount under pricing.generation`.
Credits, buckets, what costs what, spend caps, sponsoring and earnings: references/pricing.md.
Rules
- Real people. An avatar may be the creator or someone who authorised
them, and the creator keeps a record of that consent (Terms of Use). Their own photo and voice sample are fine once permission is confirmed; never generate a real person's face or design a voice to sound like them.
- Restricted content (
https://www.atmee.ai/terms-of-use): hate,
sexual content, anything involving minors, graphic violence, extremism, illegal activity, defamation, unauthorised endorsements, dangerous behaviour such as self-harm, and medical, legal or financial advice. An avatar in health, law or money informs and points to a professional.
- Public only on request, after the creator approves the title,
description and tags.
References
| File | Read when |
|---|---|
references/features.md | a question about what one surface does: atmee.ai, Studio, the API, the agent tooling |
references/use-cases.md | matching an idea to a proven shape, or showing what others built |
references/pricing.md | how credits, plans, packs, spend caps, sponsoring and earnings work |

