juturna-nodes

The officially supported node collection for Juturna, the Python library for building real-time, modular data pipelines.

Juturna pipelines are graphs of nodes: sources produce data (audio and video streams, HTTP requests, files, ...), processing nodes transform it (transcription, voice activity detection, object detection, ...), and sinks deliver the results (HTTP endpoints, databases, UDP sockets, ...). This package provides a set of ready-to-use nodes, maintained by the Juturna team, that install into the juturna.nodes namespace.

Installation

juturna-nodes requires Python 3.12 or newer, and installs the Juturna core library as a dependency.

pip install juturna-nodes

Nodes with heavy dependencies (media libraries, machine learning models) are available through optional extras, so that you only install what you need:

Extra	Dependencies	Nodes
av	av	AudioFile, AudioRtp, VideoRtp
audio-ml	faster-whisper, silero-vad	TranscriberWhisper, VadSilero
vision	ultralytics	DetectorYolo
warp	juturna[warp] (gRPC)	Warp
all	all of the above	all nodes

The vision extra installs ultralytics, which is distributed under the AGPL-3.0 license.

pip install "juturna-nodes[av,audio-ml]"
pip install "juturna-nodes[all]"

To install from source:

git clone https://github.com/meetecho/juturna_nodes.git
cd juturna_nodes
pip install -e ".[all]"

Available nodes

Sources

Node	Description	Extra
AudioFile	Read an audio file, and chunk it into audio messages	av
AudioRtp	Consume an RTP audio stream	av
VideoRtp	Consume an RTP video stream	av
FileWatcher	Watch a filesystem location, and emit filesystem events
JsonHTTP	Receive JSON content from POST requests on an HTTP server
JsonWebsocket	Expose a websocket server, and receive data from it
BrowserInput	Receive messages from the web page hosting a pipeline

Processing

Node	Description	Extra
VadSilero	Voice activity detection with the Silero VAD model	audio-ml
TranscriberWhisper	Transcribe audio with Whisper models	audio-ml
TranscriberKroko	Transcribe audio with a Kroko server (experimental)
DetectorYolo	Detect and annotate objects in frames with YOLO	vision
PassthroughIdentity	Forward input messages after a delay (for testing)
Warp	Run a node remotely, over gRPC	warp

Sinks

Node	Description	Extra
NotifierHTTP	Send messages to an HTTP endpoint
NotifierMongo	Store messages in a MongoDB collection
NotifierUDP	Send messages to a UDP socket
BrowserOutput	Send messages to the web page hosting it

Every node folder in src/juturna/nodes/ includes a config.toml file with the node arguments and their default values, and a README.md file with further details.

Usage

Installed nodes are used in a pipeline configuration by class name. The node configuration overrides the default arguments of the node:

{
  "version": "3.0.0",
  "pipeline": {
    "name": "live_transcription",
    "id": "transcription_01",
    "folder": "./running_pipelines",
    "nodes": [
      {
        "name": "0_src",
        "type": "AudioRtp",
        "configuration": { "port": 8888 }
      },
      {
        "name": "1_vad",
        "type": "VadSilero",
        "configuration": {}
      },
      {
        "name": "2_transcriber",
        "type": "TranscriberWhisper",
        "configuration": { "model_name": "small", "language": "en" }
      },
      {
        "name": "3_dst",
        "type": "NotifierHTTP",
        "configuration": { "endpoint": "http://localhost:8080/transcripts" }
      }
    ],
    "links": [
      { "from": "0_src", "to": "1_vad" },
      { "from": "1_vad", "to": "2_transcriber" },
      { "from": "2_transcriber", "to": "3_dst" }
    ]
  }
}

The pipeline can be validated and launched with the Juturna CLI:

juturna validate -c live_transcription.json
juturna launch -c live_transcription.json

Alternatively, juturna create interactively builds a pipeline configuration, listing all the installed nodes.

Nodes can also be imported directly, for instance to use them in code or to extend them:

from juturna.nodes import AudioRtp, NotifierHTTP

Nodes are imported lazily, so importing a node only requires the dependencies of that node.

Contributing

Contributions are welcome! See CONTRIBUTING.md for the development setup, the code quality checks, and a guide to writing and submitting a new node.

To report a security vulnerability, please follow the process described in SECURITY.md, and do not open a public issue.

License

juturna-nodes is released under the MIT License. Copyright © Meetecho s.r.l.