IntelligenceISIT

Demo · speech clone

Speech

type=speech

One-shot clone jobs on the GPU host: pick a voice model from the Speech catalog, a numbered voice pack, optional pace / instruction, then Generate. Live duplex streaming models are also on the Live tab.

Voice model Voice pack Pace

Text to speak Optional control instruction (style only — not spoken as extra text unless backend injects)

Play when ready

Run a speech job — recent clips stack here (last 4).

Demo

Music

Write a prompt, optional instructions, and lyrics — or generate lyrics from the prompt with a local language model, then generate the track. Some models also accept a sound clip for cover or repaint. Longer lyrics or instructions make a longer track (up to 300 seconds). New tracks play when ready, and the last 4 stay on this page.

Model

Prompt Instructions (arrangement / “instrumental only” — not sung)

Lyrics Write them myself Generate from the prompt

Lyrics (optional — leave empty for instrumental) Sound clip (optional — only if this model accepts a clip)

Model settings

Settings for the selected model.

Play when ready

Run a music job — recent tracks stack here (last 4).

Demo · multi-turn Qwen Code agent

Agent

type=code_agent

Full tools on HQ in session RAM. Multi-turn until the session clock ends — in-flight turns still finish. Tokens & cloud cost estimates sit in the left gutter.

Conversation scrolls here. Composer stays under the latest reply.

Your message

Tools only Show all

Enable thinking Show detected thinking Enter to send

Model specs & coding benchmarks

Loading model catalog…

Create a small hello.py that prints Hello, World! and show me its contents.

Demo · realtime duplex

Live voice

Azure Live

Talk with a local speech stack over a secure cloud relay: your mic is transcribed, a chat model replies, and a clone voice speaks back. Connect, pick models and a voice pack, load, then start the mic. One Live session per RIG — quiet or session limit ends the channel.

Voice model Voice pack

Voice details

Connect to load voice engine and pack specs.

Chat model

Chat model details

Connect to list models and show size, family, quant, and capabilities.

Mic

Speaker Headset

Echo cancel Noise suppress Auto gain Web search

disconnected

Mic

Voice

Live log

Expert settings

Reply length Max tokens 1536 TTS steps 8 TTS CFG 2.0 Prebuffer ms 450 Gap pad on Buffer left 200 Min silence 60 Elongate 40 Silence thr 12 Barge after 800 Speaker hold 400 Search busy max 8000 Search busy vol 12

Last reply playout waiting for first reply…

Buffer

Model RTF = VoxCPM generate only · orange = underrun · magenta = gap pad · sawtooth drain is normal when speech pauses

Doorbell · LAN

Fleet ops

LAN · Wake

GPU load, temp, VRAM, demo readiness — no visitor content. Tab appears when this browser can reach the doorbell (192.168.1.184:8790) or any RIG :8742 — so you can still Wake when both RIGs are asleep. Hidden off-LAN. Prefer the full LAN console for day-to-day ops (no Azure deploy for UI tweaks).

No accounts; prompts are not stored as product data. Bookmark this URL yourself — it is intentionally unlisted.