OpenBot

If you are looking for a pre-packaged SaaS agent platform, OpenBot is not that. Instead, OpenBot provides an open-source, self-hosted template designed to run AI coworkers inside isolated environments on your own infrastructure. Built to support AG-UI compatible agents, it gives each agent its own dedicated environment—including a browser, files, and specific tool grants—where every action is decided before execution and logged afterward. ### Architecture and Security Model OpenBot operates entirely within your infrastructure using Docker Compose, backed by PostgreSQL for data persistence. Its architecture decouples system orchestration from model provisioning: no AI models ship with the box. Administrators supply their own model credentials, which are encrypted at rest and never logged. A key structural aspect of OpenBot is that it is strictly a repository template rather than a published package or managed service. Every workspace in the repository is private. To deploy custom agent teams, you clone the repository, replace the example tenant package under `examples/` with your own channels, skills, and coworkers, and launch the stack. From an execution perspective, OpenBot accepts any AG-UI agent, whether built by hand or using an existing framework. Rather than returning standard markdown prose, agents respond in the UI using structured components. The interface also supports active monitoring: you can watch an agent work live on its screen, take manual control when it hits an action requiring human approval, and hand control back to the agent afterward. ### Environment Setup and Configuration Because OpenBot is engineered to run on a local machine, initial setup eliminates upfront authentication overhead for development. 1. Clone the repository and configure your environment files. 2. The provided `.env.example` includes the following flag: ```bash OPENBOT_SINGLE_USER=true ``` With single-user mode enabled, every incoming request is admitted as a single administrator. This allows you to test the platform locally without registering an OAuth client first. To expose the deployment to other users, single-user mode must be turned off and proper sign-in configured. 3. Launch the platform using Docker Compose. Defining agent workflows is handled via configuration rather than code. The example package includes three pre-configured coworkers defined in `agents.yaml`: * **General Assistant**: Configured for everyday tasks. * **Knowledge**: Configured to answer company-specific questions. * **Risk Analyst**: Configured for risk and compliance workflows. You can modify these coworkers directly inside `agents.yaml` or add new agents dynamically via the `/agents` route in the web UI. ### The Verdict OpenBot is currently in early alpha, meaning APIs and system structures are subject to change. However, its architectural commitment to local execution, self-hosted PostgreSQL data control, encrypted model credentials, and explicit human-in-the-loop overrides makes it a compelling starting point for developers who want full ownership over their agent infrastructure.