Gabber is an engine for building real-time AI across all modalities — voice, text, video, and more. It supports graph-based apps with multiple participants and simultaneous media streams. Our goal is to give developers the most powerful, developer-friendly AI app builder available.

If you found this interesting, please consider leaving a star ⭐️. We will be updating this repo frequently with new nodes and functionality.

LiveKit is a WebRTC server that the frontend uses to send/receive media to/from the Gabber Engine.

brew install livekit

UV for python dependency management.

curl -LsSf https://astral.sh/uv/install.sh | sh

make all

docker compose up

If you are using the LocalLLM node, you'll need to start an LLM server. LLM scripts are provided for convenience in the services/local-llm directory.

On MacOS, make sure llama.cpp is installed and use llama.cpp scripts. As of making this, the brew package for llama.cpp didn't seem to be built with Metal support (GPU acceleration for MacOS), therefore it's recommended you build from source.

On Linux or WSL you can use the vllm scripts.

cd services/local-llm

LLAMA_SERVER_PATH=/path/to/llama.cpp/build/bin/llama-server ./qwen-omni-7b-q4km_llamacpp.sh

./qwen-omni-7b-awq_vllm.sh

Some nodes need to make use of secrets. For example, the OpenAICompatibleLLM node would need your LLM provider's API key to make requests. Place your secrets in a file called .secret. Once your secrets are in place, you will be able to see them in a a dropdown for any node that needs them.

Importantly, the actual secret values are not stored in the graph data. This ensures safe sharing of graphs with the community without risk of leaking your secrets.

An App is a graph consisting of nodes and their Pad connections. It's the highest level object in Gabber.

A Node is a functional building block within a Gabber app or subgraph. Each node performs a specific operation — like ingesting media, transcribing audio, analyzing emotions, calling an external API, or generating a response. Nodes can be composed into flows that define how your AI app behaves in real time.

Nodes have Inputs and Outputs called pads. Some pads have configurable properties that effect the behavior of the node.

A Pad is a connection point on a node that allows it to send or receive data.

There are two types of pads:

• Sink Pads: Receive data from upstream nodes
• Source Pads: Send data to downstream nodes

And two "modes":

• Property: Always has a value, either an initial value or the last value streamed
• Stateless: Streams values as they happen

Pads are typed, so only compatible typed pads can be linked. When a node emits output on a pad, any downstream nodes connected to that pad can process the result in real time.

A SubGraph is very similar to an App — it’s a collection of nodes and their pad connections. However, unlike an App, a SubGraph is designed to be embedded within other apps as a single node.

By using Proxy nodes, you can create entry and exit points from your subgraph that appear in your parent app. This makes it easy to pass data in and out of the SubGraph just like any other node.

A State Machine defines how your app moves between states using branching and conditional logic.

State Machines are made up of:

• Parameters — variables that the state machine listens to
• States — distinct phases in your application flow, starting from an initial state
• State Transitions — nodes that determine when to move from one state to another based on Parameters
• Transition Logic — transitions act as AND gates, and can be combined in parallel to create OR logic

Gabber consists of a frontend and three backend services: engine, editor, repository.

The frontend is a NextJS app and is the user interface for interacting with the backend services. The frontend can be accessed http://localhost:3000.

The editor is a backend service. It's the server for the frontend.

The engine is the service responsible for running apps.

The repository service is a very thin local http server reponsible for fetching and saving apps and subgraphs. All entities are stored in the .gabber directory. It runs on port 8001.

Gabber provides SDKs to help you integrate these services into your applications quickly.

Currently available SDKs include:

• JavaScript/TypeScript SDK — framework-agnostic client library for Node.js, browsers, Bun, and Deno. Ideal for backend services or non-React frontends.
• React SDK: prebuilt hooks, providers, and UI components for building Gabber-powered apps in React or React Native with minimal setup. Wraps the Typescript SDK.
• Python SDK: for backend integrations, prototyping, scripting, Rasberry Pi, etc. (Coming Soon)
• Unity SDK: for giving eyes, ears, and a voice to your Unity games & simulations. (Coming Soon)
• iOS SDK: (Coming Soon, let us know if this is a priority for you.)
• Android SDK: (Coming Soon, let us know if this is a priority for you.)
• React Native SDK: (Coming Soon, let us know if this is a priority for you.)
• Flutter SDK: (Coming Soon, let us know if this is a priority for you.)

Refer to the SDK documentation in the main Gabber repo for installation and usage details.

Help us develop our roadmap! Right now the Gabber foundation is more-or-less solidified and we will move towards creating more nodes and examples. Here is a non-ordered list of things we find interesting:

• Computer Use Nodes
• Unity SDK + Robotics Simulation
• SIP Nodes
• Discord, Zoom, Telegram, Google Meet nodes
• Faster Computer Vision Stuff (yolo, mediapipe, CLIP, etc)
• Multi-participant turn taking
• More sophisticated examples with our state-machine system to flex its capabilities

After starting Gabber using make all, you can navigate to the examples tab to play around with different pre-made apps. SDK minimal examples are located within the corresponding SDK directories when appropriate.

More elaborate examples (for example a Unity app) will be in their own dedicated repos.

Gabber is source-available and developer-first — we’d love for you to build with us.

• Questions or feedback? Open a GitHub Discussion or start a thread in Issues.
• See something missing? We welcome contributions — new nodes or bugfixes are all appreciated.
• Early access or enterprise? Reach out or file an issue with the label enterprise.
• Stay in the loop: Follow @gabberdev on Twitter/X or star the repo to get updates.
• Join the community Join the Discord to leave feedback, chat with the team & other builders, and stay up to date.

The Gabber engine and frontend code are fair-code distributed under the Sustainable Use License and Gabber Enterprise License.

This code follows the same license as n8n

• Source Available: Always visible source code
• Self-Hostable: Deploy anywhere
• Extensible: Add your own nodes and functionality

Code that isn't core to the Gabber engine and editor, such as examples and SDKs, are licensed as Apache 2.0 which is denoted by a LICENSE file in the corresponding directories.