Apps & Distribution
OpenAgent is not a single installer. It is a model-agnostic agent system with three independent apps that are released together but installed separately.
Most setups look like this:
- Install the Agent Server on the machine that should keep the agent running.
- Install the CLI Client or Desktop App on the machine where you want to control that agent.
- Point both clients at the same OpenAgent Gateway.
1. Agent Server
This is the actual OpenAgent runtime in openagent/.
- Runs the model provider, MCP tools, memory, channels, scheduler, and auto-update flow
- Installs as
openagent-framework - Lives on the machine where the agent should run continuously
Typical install:
bash
pip install openagent-framework[all]
openagent serve2. CLI Client
This is the terminal client for talking to a running OpenAgent server.
- Installs as
openagent-cli - Connects to any OpenAgent Gateway over WebSocket
- Useful for ops, remote access, and low-friction local usage
Typical install:
bash
pip install openagent-cli
openagent-cli connect localhost:8765 --token mysecret3. Desktop App
This is the Electron app for chat, memory browsing, graph exploration, MCP configuration, and settings.
- Distributed as platform-specific binaries on GitHub Releases
- Connects to a running Agent Server
- Does not replace the server install
Distribution Model
- Tagged GitHub releases are the shared download point for all three apps.
- The Agent Server and CLI Client are attached as Python package artifacts.
- The Desktop App is attached as macOS, Windows, and Linux installers.
- The Agent Server also supports install and auto-update flows after installation.
- The Downloads page resolves the newest available GitHub asset for each app separately, so one missing artifact family does not hide the others.
Model-Agnostic Behavior
OpenAgent keeps the same operating surface across providers:
- Same MCP tools
- Same memory behavior
- Same channels and scheduler
- Same clients
Changing the model should not change how the system is operated.