Self-hosted agent platform for auditable browser-driven internal automation and governance
OpenBot, from CopilotKit, is a self-hosted enterprise agent platform that turns AI assistants into sandboxed coworkers for internal automation and browser-driven workflows. It provides each agent with an isolated containerized browser, file workspace, and scheduled routines so agents can operate inside real back-office systems. The platform enforces governance via a CEL-based policy gateway and records actions in an audit trail, and it supports the AG-UI protocol for framework-agnostic agents. Engineering and security teams who need controlled, auditable agent automation benefit most.
What tasks can teams actually assign to it?
The app gives each agent an isolated containerized browser and file workspace, enabling tasks that need real-session interactions with internal web apps and back-office tools. Use cases include scheduled routines for recurring reports, bot-to-bot delegation to hand off specialized tasks, and temporary human takeover via "Take the Wheel" for manual intervention. These capabilities make it suitable for automation that requires persistent sessions rather than single API calls.
How does it guard sensitive actions and produce auditable outputs?
The tool enforces a policy gateway that evaluates agent intent, tool name, and metadata using Common Expression Language (CEL) rules before execution. By default the gateway is configured to refuse actions unless an administrator permits them, which implements a fail-closed security posture. Audit logging captures every permitted, refused, and failed action for compliance and debugging, producing a searchable trail administrators can inspect after the fact.
What does deployment and model integration require?
The tool is self-hosted and expects infrastructure management: local development uses Docker Compose while production deployments use Helm charts for Kubernetes. Administrators supply external models or local model providers because the platform is model-agnostic; it relies on MCP support to connect agents to tools and data sources. The project is in active development and has gathered community attention with over 3,500 GitHub stars, making it suitable for engineering teams comfortable operating containerized services.
A practical choice for technically capable teams that need on-premise agent control
The tool delivers a concrete pathway to embed agents into internal workflows while preserving control over credentials and activity. Expect operator effort to validate outputs and maintain governance during active development. The tool is a pragmatic option for engineering and security teams who need auditable, self-hosted agent automation, though its early-stage status suits organizations with deployment and operations expertise. Plan pilot projects before broad rollout.
