20% OFFon credit limits — Cohort-03 exclusiveClaim Offer →
k.
kapavamoda.buzz
private inference fabric
Sign In
Cohort-03 is now closed — limited to invited developers onlyNew applications are waitlisted. Operator review required.
Private Inference Fabric • Invite-Only • Curated

The inference fabric
for builders, not browsers.

kapavamoda.buzz is a private, operator-provisioned platform. We host 47 production models across Text, Image, Video and Audio — with optimized routing on H100x8 NVLink and enterprise-grade reliability.

Designed for engineers who understand MoE sparsity, MLA, TTFT/TPOT and KV-cache residency. If you need screenshots, this fabric is not for you. Admission is manual and limited.

Explore Catalog — 47 models →View Documentation
Bearer + x-kapa-key Invite-only Enterprise Ready
Auth
Bearer + x-kapa-key
scope: inference:execute
Admission
Cohort-03 closed
operator manual
Context
1M residency
MLA 37B active
Infrastructure
Enterprise
global edge
Live Fabric Snapshot● LIVE
# POST /v1/inference/execute — invite key required
curl -X POST https://kapavamoda.buzz/v1/inference/execute \
  -H "Authorization: Bearer $KAPA_KEY" \
  -H "x-kapa-cohort: 03" \
  -d '{ "model": "claude-sonnet-5", "prompt": "ship it" }'
// 200 OK — routed via admission-control
{ "decision": { "targetRef": "h100x8-nvlink" }, "time_info": { "prefill_ms": 186, "decode_ms": 842 } }
// TTFT 186ms p50 • TPOT 14.2ms • Goodput 99.2%
Inference Path
prefill → decode → SynthID
Provenance
Vertex + 1.04T MoE
TTFT p50 • Prefill
186msSLO <500ms
Goodput 99.2% • 2.6k TPS • 8xH100
TPOT • Decode
14.2ms
KV-cache
Context
1M
MLA residency
Throughput
2.6k
TPS • H100x8
Cohort
03 • Closed
invite-only
Simple pricing

Pay as you go. No monthly subscription.

Purchase prepaid credits and use them across eligible text, image, video, and audio endpoints. Usage is deducted according to the selected model. Approved developers can review their balance, spend limits, and transaction history inside the console.

View pricingRequest access

Model Catalog — 4 Surfaces • 49 Endpoints

View all catalog →
TEXT GENERATION
LLMs • Reasoning • MoE • 1M KV-cache • GRPO • MLA
24 curated endpoints
  • GPT-5.6 SolOpenAI
  • GPT-5.6 TerraOpenAI
  • GPT-5.6 LunaOpenAI
  • GPT-5.4OpenAI
  • GPT-5.4 MiniOpenAI
  • +19 more models →
Enter text generation →
IMAGE GENERATION
Diffusion • Rectified Flow • Autoregressive • ControlNet
10 curated endpoints
  • Imagen 4 UltraGoogle
  • GPT Image 1.5OpenAI
  • Flux 2 ProBlack Forest Labs
  • Flux 1.1 Pro UltraBlack Forest Labs
  • Stable Diffusion 3.5 LargeStability AI
  • +5 more models →
Enter image generation →
VIDEO GENERATION
Diffusion + Synced Audio • Motion Brush • HDR • Omni
9 curated endpoints
  • Veo 3.1Google
  • Runway Gen-4.5Runway
  • Kling 3.0Kuaishou
  • Luma Ray 3.2Luma
  • Wan 2.7Alibaba
  • +4 more models →
Enter video generation →
AUDIO GENERATION
TTS • Music • SFX • 48kHz • Dubbing • Voice Cloning
6 curated endpoints
  • ElevenLabs TTS v2ElevenLabs
  • Fish Audio S2 ProFish Audio
  • Suno v5.5Suno
  • Udio v2Udio
  • Stable Audio 3 LargeStability AI
  • +1 more models →
Enter audio generation →
Why Limited Access

Built for a small cohort of serious builders.

We keep the cohort small so every developer gets low-latency, high-residency inference. No queue dilution, no over-subscription. When the limit is reached, onboarding pauses until the next cohort.

Limited to invited developersManual review • 24-48hNo public self-serve
MLA 37B / 671BGQA 96 headsEAGLE-3FlashAttention-3
Admission Status
Cohort-03 — Closed
Onboarding limit reached. New requests are waitlisted for Cohort-04. You will still get a response within 48 hours.

Approved developers receive an x-kapa-key scoped to their cohort. All models are actively maintained and versioned.

Join Waitlist →Developer Sign In
⚠Note on Legacy: GPT-4o/4.1/o3 and Sora 2 are deprecated - listed only for reference. Gemini 2.5 sunsets Oct 16 2026. See /docs for migration.