OneCLI

OneCLI is an open-source agent platform built around a single security idea: give every employee an assistant that can act inside the company's tools, but never let that assistant hold a credential. Requests go through a gateway that injects scoped, short-lived credentials per call, so there is no long-lived secret sitting in an agent's environment to leak or be prompt-injected out.

How it works

Each agent is bound to one employee and inherits only the access that person already has. It works through multi-step jobs on its own and pauses for human approval when it reaches an action classified as sensitive.

Guardrails

• Scoped credentials per request rather than stored secrets.
• Lane enforcement so an agent cannot reach across into another department's tools.
• Approval gates on sensitive actions.
• Rate dampening — the platform slows an agent down automatically when it detects repetitive or unusually fast behaviour.

Integrations

The gateway fronts the usual operational surface: Postgres and other databases, Stripe, repository hosts, ticketing systems and internal services, with Slack used for agent notifications and hand-offs.

Compliance

OneCLI reports SOC 2 Type II in progress and GDPR compliance with a DPA available, and states that agents never store or access real passwords.

Pricing

A 7-day trial includes $5 of AI credits and needs no card (3 seats, 3 agents). Team is $149/month, or $99/month billed yearly, for 5 seats and 10 agents. Scale is $499/month, or $333/month yearly, for 10 seats and 20 agents, with extra seats at $49/user/month on bring-your-own-cloud keys or $199 with hosted models. Enterprise is custom. Team and above can bring their own Anthropic or OpenAI key.

Background

The project is Y Combinator-backed and open source, with the source at github.com/onecli/onecli and roughly 3,000 stars at launch. OneCLI lists Docker, MindsDB, Zoho, Coralogix and Kakao Entertainment among its users.