Research preview · v0.1

Realtime 3D avatars,
streamed to any browser.

Train a photoreal avatar from a 5-minute phone capture. Speak to it in real time. Embed it anywhere a video tag works.

Get early access How it works Read the spec
5 min to train your avatar
1.18s to respond
30 fps in any browser

See the engineering details ↗

Fig. 01 · The Laika conversation surface. Avatar stream is server-rendered and arrives over WebRTC; the transcript and per-stage HUD are driven by gateway events.

Built on the open stack

  • Bun
    runtime
  • Hono
    gateway
  • Svelte
    frontend
  • Anthropic
    LLM
  • Whisper
    STT
  • F5-TTS
    synthesis
  • NVIDIA
    animation
  • WebRTC
    transport

See it in action

One avatar, four jobs.

Same trained avatar, four contexts. Hover any tile to see it come alive.

Creators

Voice-cloned avatars for streaming.

Capture once, broadcast as a photoreal 3D character driven by your live voice.

5 min to train

Agencies

Branded characters as a hosted endpoint.

Train a client's spokesperson, IP-licensable mascot, or virtual presenter.

1 GPU per session

Research

Rapid avatar capture for HCI studies.

Reproducible Gaussian-splat captures from monocular phone footage.

$5 per run

Customer ops

Realtime avatars in front of an LLM.

Plug any OpenRouter model into the gateway and front it with a face.

<1.2s round-trip

Release timeline

From skeleton to product.

Currently · Phase 1
  1. Phase 0 · Mar 2026 Shipped

    Walking skeleton

    Repo + service stubs running end-to-end on CPU with mock STT/LLM/TTS.

  2. Phase 1 · May 2026 In progress

    Real models, rented GPUs

    Mocks swapped one at a time. Inference on Vast.ai A10s, training on RunPod H100s.

  3. Phase 2 · Q3 2026 Next

    Local inference

    5070 Ti arrives. Renderer, animator, STT, TTS all run locally inside ~13 GB.

  4. Phase 3 · Q4 2026 Later

    Productisation

    Multi-tenant sessions, optional local LLM, auth + billing + observability.

Backers

Get early access. Help fund the build.

Laika is a solo project running through alpha. Three ways to get in early — and three secondary channels for one-off support.

Early access

$25 / month

For developers and creators who want to use Laika in their own projects as it ships.

  • Private alpha builds
  • Priority access to the SDK
  • Monthly build notes
  • Cancel anytime
Join via Patreon

Founding member

Recommended
$250 one-time

Back the Kickstarter. Reserve your trained avatar and lock founding-member pricing for life.

  • One trained avatar (≈45 min H100)
  • 30 minutes of session credit
  • Founding-member rate, locked
  • Name in the credits
Notify me at launch

Design partner

Talk to us

For studios, agencies, and research groups who want a co-developed deployment.

  • Co-developed deployment
  • Custom training pipeline
  • Direct line to engineering
  • Quarterly roadmap input
Email Joe

Kickstarter · Q3 2026

Get notified when we launch.

One email · no spam

Other channels

Ready when you are

Ship a face this quarter.

Get your trained avatar, the SDK, and direct access to engineering — for less than a week of cloud inference would cost.

Reserve early access Talk to us