onecli/onecli

What it solves

OneCLI provides a secure, team-oriented platform for deploying and managing AI agents. It addresses the difficulty of scaling autonomous agents from a single user to a company-wide deployment by centralizing secret management, permissioning, and identity provider integration, ensuring that agents operate within secure sandboxes without exposing sensitive credentials to the agent itself.

How it works

OneCLI creates a dedicated, sandboxed agent for every team member. These agents operate through a Rust-based gateway that intercepts outbound requests to inject credentials and enforce security policies. The architecture consists of a web dashboard for management, an API server as the control plane, a runner that manages agent sandboxes, and a channel adapter for Slack integration. Agents are equipped with their own filesystem, shell, and persistent memory, and can be run on private infrastructure via an outbound-only runner.

Who it’s for

Companies and teams that want to provide employees with personal AI agents capable of performing real work while maintaining strict security, governance, and human-in-the-loop approvals for critical actions.

Highlights

• Credential Injection: Secrets are injected by the gateway at request time, meaning agents never see the actual keys.
• Identity Integration: Provision agents automatically via company identity providers (IdP).
• Sandboxed Environments: Each agent has an isolated computer with its own filesystem and shell.
• Human-in-the-Loop: Supports deterministic approvals for high-risk actions like deleting data or sending emails.
• Multi-channel Access: Agents can be interacted with via a dashboard or dedicated Slack apps.
• Outbound-only Runner: Allows deployment on laptops or VPCs behind NAT without requiring inbound ports.