AI Decision System
Replayable, offline-verifiable records for AI-assisted go/no-go decisions: experiment promotions, typed model answers and human-approved actions.
翻譯準備中
尚未偵測到更新。GitHub 發布版本後會顯示在這裡。
個人 GitHub 帳號下公開儲存庫的擁有者可以登入並驗證所有權,自行管理介紹,無需管理員核准。
登入並產生驗證檔案,在 30 分鐘內將其提交至此儲存庫的預設分支。其他分支或分叉儲存庫中的檔案無效。然後在下方檢查檔案,即可無需管理員核准直接管理專案。請保留檔案以維持管理權限。
如需登入,將前往登入頁面。登入後請回到此處,再次點擊所選的驗證按鈕。
在專案資料夾中執行 /nomorevibe verify,即可驗證所有權並自行管理介紹。
一分鐘完成驗證正在載入完整 README。 以下為已儲存的摘錄。
Real-time communication of any data for web browsers, native apps, IP cameras, USB cameras, robots, drones and AI agents. One publish/subscribe API with a simple architecture, high-performance QUIC transport, priority control and caching, so you no longer have to pick between WebSocket, MQTT, HLS, DASH and WebRTC for each use case. Implements Media over QUIC Transport draft-ietf-moq-transport-14 in Rust (drafts in docs/spec/). Try the demos hosted at. They connect to the cloud relay by default, so nothing needs to be installed. [README 설치·사용 부분 발췌] Getting Started檢視完整 README ↗
Real-time communication of any data for web browsers, native apps, IP cameras, USB cameras, robots, drones and AI agents. One publish/subscribe API with a simple architecture, high-performance QUIC transport, priority control and caching, so you no longer have to pick between WebSocket, MQTT, HLS, DASH and WebRTC for each use case.
Implements Media over QUIC Transport
draft-ietf-moq-transport-14
in Rust (drafts in docs/spec/).
Try the demos hosted at https://nttcom.github.io/moq-wasm/. They connect to the cloud relay by default, so nothing needs to be installed.
| Use case | Demo |
|---|---|
| Video meetings | Meeting |
| Large-scale live streaming | Live Viewer: live playback with rewind and rendition switching |
| Video on demand | Live Viewer: publish an MP4 from the browser |
| Remote monitoring and control of IP cameras | ONVIF: PTZ control and monitoring |
| Remote monitoring from USB cameras | Remote Monitoring |
| Real-time chat moderation by an AI agent | MoQ Chat Moderation |
| Real-time video analysis by an AI agent | MoQ Camera Detection: answers questions about a camera feed |
| Camera control by an AI agent | MoQ PTZ Tracking: pans and tilts an ONVIF camera to keep a target centered |
| Remote monitoring, data collection and control of drones | |
| Remote monitoring, data collection and control of robots | |
| Remote monitoring, data collection and control of autonomous vehicles |
Publishers outside the browser: crates/live-ingest (RTMP/SRT),
crates/gst-plugin-moqt (moqtsink),
examples/rust/moq-cli and the
pipecat bots. Details of every browser demo are in
examples/browser/README.md.
nix develop
npm --prefix examples/browser ci
Run a relay, publish a test stream and watch it, in four terminals:
make relay # 1. relay and a local VTS on https://127.0.0.1:4433
make live-ingest # 2. RTMP/SRT ingest publishing to the relay
make ffmpeg-srt # 3. test pattern over SRT into the ingest
make browser # 4. wasm build and the example hub on http://localhost:5173
Then make chrome (make chrome:linux on Linux) opens Chrome trusting the
relay's self-signed certificate. Open Live Viewer from the hub and watch
anon/live/test.
docker compose up -d
Starts relay-a (https://127.0.0.1:4433) and relay-b (:4434) sharing a
VTS and a Redis route registry, the topology the cascading, auth and meeting
E2E use. The make targets above publish to relay-a and resolve the Docker
Desktop bridge host for native clients on macOS.
make test # Rust unit tests
make browser-e2e-media # browser publish/subscribe
./tests/auth-e2e/run.sh # relay scenarios, see tests/README.md
cargo test -p vts # VTS
flowchart LR
subgraph Publishers
B1[Browser<br/>wasm]
LI[live-ingest<br/>RTMP / SRT]
GS[moqtsink<br/>GStreamer]
ON[onvif bridge]
CLI[moq-cli]
end
subgraph Relays
RA[relay-a] <--> RB[relay-b]
VTS[VTS<br/>JWT verification]
RA -.-> VTS
RB -.-> VTS
end
subgraph Subscribers
B2[Browser<br/>Live Viewer, Meeting, ...]
BOT[AI agents<br/>pipecat bots]
CLI2[moq-cli]
end
Publishers -- PUBLISH / SUBSCRIBE / FETCH<br/>QUIC or WebTransport --> RA
RB --> Subscribers
Every client speaks MoQT to a relay over QUIC or WebTransport. Relays cache
objects for FETCH, forward subscriptions to each other and authenticate
sessions through the VTS. The moqt crate is the protocol library every other
component builds on. Every Rust crate lives in crates/<package name>/.
| Path | Component |
|---|---|
crates/moqt/ | Protocol library: wire format, sessions, publisher and subscriber API |
crates/relay/ | Relay: QUIC and WebTransport on one port, cache, FETCH, cascading, JWT auth |
crates/wasm/ | WebAssembly bindings used by the browser examples |
crates/gst-plugin-moqt/ | GStreamer plugin with the moqtsink element |
crates/live-ingest/ | RTMP and SRT ingest into MoQT |
crates/onvif-ingest/ | ONVIF/RTSP cameras with PTZ control into MoQT |
crates/mediapack/, msf/, broadcast/, transcode/ | Media crates shared by the publishers: containers, MSF catalog, the broadcast's track layout, re-encoding |
| , |
Architecture documents:
moqt,
relay,
broadcast,
Live Player.
Contributor rules: AGENTS.md.
Relays always authenticate. A session without a token is limited to
anon/**, which every demo uses by default. Tokens are HS256 JWTs verified
by the VTS; sign one with the auth-token CLI
(crates/auth-token/README.md). See
crates/vts/README.md.
crates/vts/| Verify Token Service: checks client JWTs for the relay; the shared JWT claims type |
examples/ | Browser examples, moq-cli, pipecat bots |
tests/ | Relay E2E scenarios: auth, FETCH, cache eviction, cascading, dedup |
docs/architecture/ | Architecture and dependency decisions per crate |