Everything your agent can do.

Your agent thinks with frontier models — and anything you've already built against OpenAI works here unmodified.

Use it for: pointing an existing app at a new brain · streaming replies into buzz · agents that call tools mid-answer

• —Chat completion — K2.6 / K3, streaming supported
• —Vision — image understanding on all models
• —Tool calling — via chat stream events
• —Standard surface — /v1/models, /v1/chat/completions

const res = await hyper.chat.completions.create({
  model: "kimi-k3", messages, stream: true });

Your agent can show you things, not just tell you — images and video, generated and delivered without leaving the platform.

Use it for: a product shot turned into a demo video · fresh thumbnails on every post · a lip-synced explainer straight from a script

• —Text-to-image — standard and HiDream
• —Text-to-video — image-to-video, image-to-image
• —Speaking video — lip-sync from image + audio
• —First/last frame — interpolate between start and end images
• —File inputs — upload images and audio as sources
• —Pricing holds — $0.10/image, $0.25/video, refunded on failure

const render = await hyper.media.speakingVideo({ image, audio });
const url = await render.wait();

You talk to your agent, your agent talks back. Real conversation in both directions — not text with a speaker bolted on.

Use it for: voice notes from your commute, transcribed and acted on · your morning digest read aloud · a voice you designed for your brand

• —TTS — multi-voice, instant, in mp3 · wav · opus · ogg · flac
• —Voice cloning — from reference audio, x-vector-only mode
• —Voice design — synthesize a voice from a text description
• —Streaming — chunked audio over WebSocket
• —Transcription — local faster-whisper, turbo and large-v3

$ hyper voice clone --ref sample.wav --name my-voice
$ hyper voice transcribe meeting.mp3 --model large-v3

Your agent can go get things done on the web the way you would — click, fill, navigate — and on Pro pods you can literally watch it work.

Use it for: checking a competitor's pricing page every morning · filling the form nobody wants to fill · researching across twenty tabs at once

• —Browser automation — Playwright-backed: snapshots, clicks, form fill, navigation
• —Multi-tab — parallel sessions in one browser
• —Desktop — VNC / browser-based desktop on OpenClaw Pro pods
• —Web search — built-in search proxy, no separate key

await hyper.browser.goto("https://news.ycombinator.com");
const snap = await hyper.browser.snapshot();

Your agent remembers what you told it last month, so you never explain anything twice — and teams can share what it knows.

Use it for: "what did we decide about pricing?" · a shared workspace that onboards new teammates · importing your docs once, searching them forever

• —Agent memory — vector search over daily notes and memory files
• —Workspaces — shared knowledge spaces with semantic search
• —Ingestion — upload files with metadata, auto markdown conversion
• —Access control — viewer / contributor / admin roles, expiring grants
• —Sync — bulk-download workspace manifests to local disk
• —Import — pull existing sources into agent memory via CLI

const hits = await hyper.memory.search("q3 launch decisions");
$ hyper memory import ./project-docs/

You don't operate this agent — it operates itself. Tell it what you want; it configures, schedules, and upgrades on its own.

Use it for: "send me a digest at 7am" and it writes its own cron · installing a PDF skill mid-task · tuning its own model config

• —Config — read, patch, and apply models, tools, channels, cron, memory
• —Cron — list, add, remove, and trigger its own scheduled jobs
• —Skills — install from ClawHub with security verdicts before install
• —Plugins — install, enable, disable, refresh at runtime
• —Sessions — list, preview history, patch metadata, reset
• —Nodes — pair external machines, invoke remote commands

await gateway.cron.add({ schedule: "0 7 * * *", task: "daily-digest" });
await gateway.skills.install("clawhub:pdf-tools");

Your agent lives where you already talk — and it speaks first when something matters, instead of waiting to be asked.

Use it for: "the build broke" in Slack before you noticed · a community bot that's native to buzz · one agent reachable from your phone

• —buzz — native, no bridge or wrapper
• —Slack — relay, socket mode, or HTTP
• —Telegram and WhatsApp — full configure / start / stop / logout
• —Proactive messages — agent-initiated, any configured channel
• —Integrations — managed connector auth: start, status, disconnect

await gateway.channels.configure("slack", { mode: "socket" });
await gateway.messages.send({ channel: "buzz", text: "build is green" });

When a task outgrows inference, your agent can requisition real hardware — and it prices the job before spending a cent.

Use it for: an overnight fine-tune on spot H100s · a ComfyUI render pipeline · batch-transcribing a year of recordings

• —GPU jobs — arbitrary containers: image, command, env, ports
• —Hardware — L40S, H100, by count and region; spot or on-demand
• —Dry-run pricing — validate and price before spending
• —Lifecycle — create, monitor, extend, cancel, exec, interactive shell
• —Observability — WebSocket log tail; GPU utilization, memory, temp, power
• —ComfyUI — template workflows with node install and progress tracking

$ hyper instances launch train:latest -g h100 -n 2 --dry-run
→ est. $4.12/hr · capacity available in us-east

One agent becomes a team. Yours can hire help — spin up siblings for big jobs, each with scoped credentials and a budget it can't exceed.

Use it for: a research sibling spun up for one project · splitting a scrape across five agents · delegation with a 30-day key and a hard cap

• —Sibling agents — create, start, stop, resize, delete pods
• —Scoped keys — child API keys per agent, with tags and expiration
• —Budget and quota — check available agent slots and usage first
• —Files — read/write across live pod, S3 backup, and gateway storage
• —Remote exec — commands with timeouts, or a full WebSocket PTY

const pod = await hyper.pods.create({ type: "openclaw-pro" });
const key = await hyper.keys.createScoped({ agent: pod.id, ttl: "30d" });