Local-first
Gabber is designed around models served by Ollama on your own machine. Your available Ollama models become the pool Gabber can work with.
What It Is
A roundtable for local AI.
Gabber is a Linux desktop application built around Ollama-managed language models. Rather than treating every LLM as an isolated chatbot, Gabber lets you choose multiple local models, arrange them in a deliberate sequence, and use that order to guide the conversation.
Multiple perspectives
Select several models for the same session so one model can introduce an idea, another can challenge it, and another can refine or extend it.
User-defined order
The order of the selected model cards matters. Drag them into the sequence you want Gabber to use during the conversation.
The Idea
Different models. One conversation.
Different LLMs often have different strengths, blind spots, writing styles, and reasoning habits. Gabber makes those differences useful by giving them a shared conversational workflow instead of forcing you to manually copy responses between separate terminals or chat windows.
The result is less like asking the same question three times and more like putting three different minds around the same digital coffee table.
Downloads
Choose your package.
Both packages are located in the root of this release page.
Linux x86_64 Binary
For users who want to run Gabber without building the application from source.
Download Binary Gabber_Beta_linux_x86_64.zipSource Package
For development, inspection, modification, rebuilding, or running Gabber directly from its Python environment.
Download Source Gabber_Beta_SRC.zipPrerequisites
Install Ollama and at least one LLM.
Gabber does not bundle language models. Ollama handles the model runtime and model library, while Gabber provides the multi-model conversation interface.
1. Install Ollama
On a typical Linux system, Ollama can be installed from its official install script:
curl -fsSL https://ollama.com/install.sh | sh
If Ollama is already installed and working, you can skip this step.
2. Start Ollama
Ollama normally runs as a local service. If it is not already running, start it manually:
ollama serve
Leave that service running while using Gabber if your system does not already manage it in the background.
3. Install a model
Pull one or more models into Ollama. For example:
ollama pull llama3.2 ollama pull qwen3
Model names and sizes vary. Choose models appropriate for your hardware.
4. Verify your models
Confirm Ollama can see the models you installed:
ollama list
Gabber uses the models managed by your Ollama installation, so a healthy Ollama setup is the important prerequisite.
Model Ordering
The order is part of the conversation.
Selected model cards can be dragged into the sequence you want. Gabber follows that selected-model order when coordinating model participation.
Example order
⋮⋮1Model AInitial analysis
⋮⋮2Model BCritique or alternative
⋮⋮3Model CSynthesis or refinement
Why change the order?
Model order can change the character of a session. A strong general model placed first can establish context. A coding model placed next can focus on implementation. A smaller or faster model placed last can summarize, critique, or tighten the result.
You can reorder the cards to experiment with different conversational roles without changing which models are selected.
Using Gabber
From launch to conversation.
Once Ollama is running and your models are installed, the normal workflow is straightforward.
- Launch Gabber. The application connects to the local Ollama environment and discovers the models Ollama manages.
- Select the models you want to use. You can build a session from one model or several, depending on the kind of conversation you want.
- Drag the selected model cards into order. The visible card order determines the sequence in which the selected models participate.
- Enter your prompt or topic. Give the models enough context to understand the task, question, problem, or discussion you want them to work through.
- Run the conversation. Gabber coordinates the selected models according to the arranged order so each can contribute to the evolving discussion.
- Experiment. Change the model mix or reorder the same models to compare how a different sequence changes the conversation.
Running the Binary
Unpack and launch.
Extract the binary release, enter the extracted directory, make the Gabber executable runnable if needed, and start it:
unzip Gabber_Beta_linux_x86_64.zip cd Gabber_Beta_linux_x86_64 chmod +x Gabber ./Gabber
The exact extracted folder name may vary with how the release archive is packaged. The important part is launching the included Gabber executable from the extracted release directory.
Build From Source
Use the included project scripts.
The source archive is intended for users who want to inspect, modify, run, or rebuild Gabber locally.
Extract the source
unzip Gabber_Beta_SRC.zip cd Gabber_Beta_SRC
From here, work inside the extracted Gabber project directory.
Prepare the environment
If the source package includes the project installer script, make it executable and run it:
chmod +x installer.sh ./installer.sh
This prepares the local Python environment and project dependencies used by Gabber.
Run from source
If the source package includes a run script:
chmod +x run.sh ./run.sh
You can also activate the project virtual environment and run the application entry point directly when developing.
Build the release
If the source package includes the Gabber build script:
chmod +x build.sh ./build.sh
The build output location is determined by the included build configuration. Review the build script before packaging or redistributing your own build.
Quick Answers
Gabber FAQ.
Does Gabber include its own LLMs?
No. Gabber uses language models managed by Ollama. Install the models you want with Ollama first.
Does model order actually matter?
Yes. The order of the selected model cards controls the sequence in which the participating models interact.
Can I use more than one model?
Yes. Multi-model participation is the central purpose of Gabber.
Can I change the model order?
Yes. Drag the selected model cards into the order you want before or while configuring your session.
What hardware do I need?
Gabber itself is comparatively lightweight. Your actual hardware requirements are mostly determined by the Ollama models you choose, their parameter count, quantization, context size, and whether inference runs on CPU, GPU, or both.