Now loading...
Starting today, Claude Managed Agents are now capable of running on a schedule, enabling secure access to command-line interface (CLI) tools and other authenticated services. These features have entered public beta on the Claude Platform.
The newly introduced scheduled execution allows agents to perform routine tasks automatically. With a scheduled deployment, agents are provided with a cron-like schedule, automatically launching a new session and completing designated tasks each time the schedule is triggered, without requiring users to create or manage their own scheduler.
This capability can be used for repetitive tasks such as nightly data synchronization, weekly compliance assessments, or daily summaries. Users have the flexibility to pause, resume, or archive these deployments as needed and can also initiate additional runs at any time.
Various teams are already using scheduled deployments to streamline their operations. For instance, Rakuten has implemented scheduled deployments to regularly analyze spreadsheet data and generate reports and presentations on a weekly or monthly basis. Additionally, they monitor production logs and metrics to provide product managers with insights into application performance without the need for a dedicated dashboard. Similarly, Actively AI uses Managed Agents to facilitate cross-account searches for sales teams, simplifying their infrastructure by automating answer updates. Moreover, Ando employs these agents to assist hiring and sales teams by monitoring channels for follow-up actions and sending meeting reminders.
In conjunction with these scheduling capabilities, agents have also been enhanced to securely store environment variables, facilitating authentication with external CLI tools. Now, teams can connect to various systems through direct API calls and other command-line interactions. This update allows for a streamlined integration process where an API key can be registered with an environment variable name and specified domains, enabling the agents’ CLIs to execute authenticated API requests.
Importantly, the API key remains concealed from the agent, as only a placeholder is stored in the sandbox environment. The actual key is accessed at the network boundary, and its usage is restricted to approved domains. When changes need to be made to a key, updating it in the vault will automatically refresh the value for ongoing sessions at their next request. Most CLIs that use HTTP requests for authentication, such as Browserbase, KERNEL, Notion, Ramp, and Sentry, operate within this framework. Additionally, Browserbase and KERNEL have introduced browser capabilities to Managed Agents for the first time, allowing them to navigate and interact with the web alongside other tools.
Organizations are taking advantage of environment variables stored in vaults to provide agents with secure access to essential tools. For example, Notion has integrated environment variables in vaults to expand its CLI functionalities, allowing file uploads by its agents without needing to expose API tokens. Meanwhile, Browserbase has developed a public catalog using the browse CLI, regularly updated through scheduled deployments to ensure accuracy. Additionally, KERNEL employs this method to connect its agents securely to databases that track usage and customer conversations, enabling prompt notifications of usage spikes. Lastly, Milana utilizes these environment variables to securely link its AI product engineer with customer codebases, allowing for faster automated bug fixes and large-scale data analysis.
