OneCLI Offers Secure, Self-Hosted AI Agent Management for Teams

OneCLI is an open-source agent harness designed for teams, providing a secure, sandboxed environment for deploying and managing personal AI agents across an organization.

Developed by Jonathan and Guy, both veterans of the cybersecurity sector, OneCLI emerged from their work on an open-source database tool, ChartDB. They observed a significant vulnerability in autonomous agents like OpenClaw: their tendency to store sensitive credentials in memory or local files, making them susceptible to prompt injection attacks and unauthorized access.

To counter this, OneCLI was conceived as a 'vault for AI Agents,' initially built in Rust. Its core innovation lies in never allowing the agent itself to hold real secrets. Instead, credentials are injected at a secure gateway, per request, only after authorization, ensuring they never enter the agent's context, memory, or logs.

The platform enforces policies at the network layer, external to the agent and the underlying large language model (LLM). This architecture allows administrators to define rules for endpoint access, rate limiting, and required human approval, which the agent cannot bypass. Each agent operates within an isolated virtual machine, limiting the blast radius of any potential compromise.

OneCLI integrates with common business tools such as GitHub, Gmail, Notion, and Dropbox, enabling agents to perform real-world tasks. It also features deterministic human-in-the-loop approval for critical actions like sending emails or deleting tickets, ensuring oversight on sensitive operations.

For teams, OneCLI provides centralized policy management and global connections, such as shared LLM keys or service accounts. The entire platform is open-source under an Apache-2.0 license (with an enterprise exception), allowing companies to self-host and verify the code for security and control.

The system leverages 'jcode' for its agent engine, which the creators claim improves agent intelligence and speed. Use cases range from automating sales lifecycle management and CRM hygiene to orchestrating product development workflows, ensuring that every action is logged and bound to an employee identity for a full audit trail.

This approach provides a robust framework for enterprises to adopt AI agents without compromising on security or control, offering a clear path to scalable automation for sensitive business processes.