OneCLI is an open-source AI agent harness and execution gateway created by OneCLI (a Y Combinator-backed company) designed to safely deploy and orchestrate autonomous agents across engineering teams. The platform acts as a secure intermediary between AI models and underlying infrastructure, enabling agents to execute tasks while enforcing deterministic operational boundaries. In daily operations, team members interact with agents via chat interfaces such as Slack or through command interfaces. Agents operate within sealed sandboxes and perform automated tasks—such as diagnosing error spikes, opening pull requests, and querying databases—while all API requests and tool invocations are routed through the OneCLI gateway. The gateway dynamically injects scoped credentials per request, ensuring the LLM itself never has direct access to raw secrets or passwords. OneCLI differentiates itself from traditional CLI agents by enforcing security and governance outside the model's context window. Instead of relying on prompt instructions to prevent unintended behavior, OneCLI enforces hard rules at the proxy and network level, including rate limiting against runaway loops, explicit blocks on destructive commands, and mandatory human-in-the-loop approval prompts for sensitive operations. OneCLI is open-source software with source code available on GitHub. Hosted and team management services are provided via app.onecli.sh, offering a free tier that includes 500 calls per month and $5 in AI credits.
Tags: agent harness, ai agent, assistant, cli, OpenSource