This repository publishes AgentBridge for Linux as a proper desktop application: the engine plus the Giraffe AI web client rendered in its own desktop window (WebKitGTK) with an application icon in the launcher — a normal Linux desktop app, not a console program.
Every AgentBridge GitHub release is picked up automatically (hourly check) and a new Flatpak is built and published here, so the engine stays current without any manual step.
Replace employees with easy, autonomous and effective Personal AI Assistant.
AgentBridge is a self-hosted server that runs its own AI agents behind two primary interfaces in a single process: a full-screen chat terminal (TUI) and an HTTP API that speaks the standard OpenAI protocol, so any OpenAI-compatible client drives the same agents.
Chat in the terminal (or in the desktop window, or with your own OpenAI client) while scripts, bots, and apps use the same agents on the same port: same process, same conversations, no bridges, no synchronization.
- You, in the desktop — the bundled Giraffe AI client opens in its own window, auto-connected to the local engine. Web apps run in "desktop mode": a web view with an application icon, not a browser tab.
- You, in the terminal — a modern full-screen chat UI (Terminal.Gui) with streaming
replies, a
/command palette, file attachments, voice dictation and in-process neural text-to-speech. - Any OpenAI-compatible client — SDKs, bots and scripts talk plain
POST /v1/chat/completionsto the same agents.
The agent core is our own engine (AIOrchestrator): an agent's abilities are compiled
.NET assemblies loaded as plugins from the Tools/ folder and driven in-process. The
agent drafts documents, works on spreadsheets and presentations, interacts with email and
Telegram, browses the web, produces podcasts and answers from your own knowledge base — a
plain folder on disk. Your files never leave your machine.
Everything runs on your own hardware, self-hosted. No cloud account, no uploads: connect the agent to your preferred AI — a local model (Ollama, ExLlamaV2, …) or any OpenAI-compatible provider.
Full documentation: graphenelab.it/AgentBridge · source: Graphene-Lab/AgentBridge · license: AGPL-3.0
Requirements: Flatpak with the Flathub remote (first install also downloads the GNOME runtime from Flathub).
# one time: add the Flathub remote (needed for the shared runtime)
flatpak remote-add --user --if-not-exists flathub https://dl.flathub.org/repo/flathub.flatpakrepo
# download + install the latest desktop bundle — x86_64 (Intel/AMD), or the
# aarch64 link below on ARM64 (Raspberry Pi 4/5, Apple-silicon Asahi, …)
flatpak install --user -y \
https://github.com/Graphene-Lab/AgentBridge-Linux/releases/latest/download/agentbridge-linux-x86_64.flatpak
# ARM64:
# flatpak install --user -y \
# https://github.com/Graphene-Lab/AgentBridge-Linux/releases/latest/download/agentbridge-linux-aarch64.flatpakLaunch it from your desktop environment (icon AgentBridge) or from the terminal:
flatpak run io.github.graphene_lab.agentbridgeUpdate — a new release appears here automatically at each AgentBridge release;
re-install with the same command (or --reinstall) to pick it up (use the -aarch64
asset URL on ARM64):
flatpak install --user -y --reinstall \
https://github.com/Graphene-Lab/AgentBridge-Linux/releases/latest/download/agentbridge-linux-x86_64.flatpak| Launcher entry | What it does |
|---|---|
| AgentBridge | Desktop app: starts the engine (headless) if needed and opens the Giraffe AI chat client in its own window, already connected to the local engine |
| AgentBridge (Terminal) | The classic full-screen terminal UI (TUI) + local server, in a terminal |
The bundled client is the latest Giraffe AI release at build time; the engine is the official self-contained AgentBridge linux-x64 / linux-arm64 release of the same build.
- Launch AgentBridge.
- The window asks you to configure an AI provider — point it at your local model
(
http://localhost:11434/for Ollama) or at any OpenAI-compatible API you use. (The AgentBridge engine itself is already pre-configured as a provider named "AgentBridge", selected by default.) - Start typing: the agent runs on your machine and reads/writes your own files.
The engine is healthy when it answers on http://localhost:5290/health (the standard
port, configurable in PersistentData/appsettings.json).
- Per-app data lives under
~/.var/app/io.github.graphene_lab.agentbridge/data/:agentbridge/engine/— the running engine copy (first launch materializes it from the read-only package;PersistentData/inside it holds all user-editable JSON config, exactly like a normal install).agentbridge/logs/— engine logs (desktop-headless.logfor the desktop mode).
- Version pinning: the packaged engine ships with updates disabled (
--no-update) — the new version arrives with the next Flatpak build (automatic). Power users can pass explicit engine arguments, e.g.flatpak run --command=agent-run io.github.graphene_lab.agentbridge --headless, and manage in-app auto-updates themselves. - Sandbox: this package keeps a deliberately broad sandbox (network, home folder,
audio, GPU) — it is the Flatpak equivalent of the classic
~/.agentbridgearchive install of a self-hosted server, not a restricted store app.
packaging/agentbridge.yml.tmpl flatpak manifest template (rendered per release)
packaging/branding/ launcher scripts, .desktop entries, metainfo, giraffe.svg
.github/workflows/linux-release.yml the automatic build + publication pipeline
Build and publication pipeline: see .github/workflows/linux-release.yml and AGENTS.md.
The desktop GUI is the Giraffe AI web client (Graphene-Lab/GiraffeAI); the giraffe logo asset is from that project.